From jh at tutpro.com  Wed Mar  1 01:51:49 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Wed, 1 Mar 2006 08:51:49 +0200
Subject: [p2p-sip] FYI: I-D regarding decentralized identity
 infrastructure based on ENUM
In-Reply-To: <4404B6A7.3010706@cs.columbia.edu>
References: <44042657.8040009@enum.at>
	<4404B6A7.3010706@cs.columbia.edu>
Message-ID: <17413.17541.89771.641525@rautu.tutpro.com>

Kundan Singh writes:

 > This has a problem: phone numbers are too rigid and tied to a physical 
 > port/line. Users may get E.164 number but what about entities like 
 > virtual user identifiers or devices (e.g., sip:lamp at cs.columbia.edu 
 > belongs to an electric x10 controlled lamp in our lab).

i don't like phone numbers either.  they should go away as soon as
possible.  my requirement is that p2p architecture MUST allow use of
URI that equals user's mailto URI, but only differ in URI scheme.

-- juha

From philip_matthews at magma.ca  Wed Mar  1 09:44:37 2006
From: philip_matthews at magma.ca (Philip Matthews)
Date: Wed, 1 Mar 2006 09:44:37 -0500
Subject: [p2p-sip] New I-D: The effect of NATs on P2P SIP
	Overlay	Architecture
In-Reply-To: <4404C388.9090408@cs.columbia.edu>
References: <E3C3A9D5-2639-4F80-8922-F9DDBE866F02@magma.ca>
	<4404C388.9090408@cs.columbia.edu>
Message-ID: <12A2E483-9001-4E6F-B996-0646604ED7EA@magma.ca>

Hi Kundan:

Thanks for your comments!
A few replies inline.

- Philip

On 28-Feb-06, at 16:41 , Kundan Singh wrote:

>
>> The plain text version of the draft is available at
>> http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
>> and- overlays-00.txt
>
> I liked the analysis in the draft and agree with the conclusions.
Thanks!

>
> sec 4.4.3: I think Kademlia does symmetric that can fit the  
> description there.
Hmmm... I have the opposite impression, but I am not sure of my facts.
Certainly Kademlia has been very successful, so if it is symmetric,  
that would be a big plus.

>
> sec 4.4.2: I think the number 160 is misleading, since usually
> N << 2^160, and nodes will maintain only approx O(logN)  
> connections. For example, only 32 connections per peer for four  
> billion peers (2^32).
You are right. 160 is the theoretical maximum (assuming a 160-bit  
hash length), and the actual number is
about log N. This could be made clearer in the draft.

>
> sec 4.3: just to understand: is static connection == pointers in  
> DHT routing table? and dynamic == call setup? Otherwise, what does  
> static mean?

We see two types of connections.
The first type is used to maintain the overlay, do lookups, and for  
signaling. These follow the partial mesh
pattern described in the paper. We see these as static, meaning that  
they are long-lived (assuming
the peers at either end are long-lived) and do not change as calls  
are set up and torn down.
Since these are just a partial mesh, messages flowing over these may  
need a number of hops to reach
their destination.
The second type is used for the media in the call. These go from the  
calling peer to the called peer
and are dynamically set up when the call is established and torn down  
immediately afterwards.

It would be possible to have a third type. These would be set up when  
the called party receives the initial
Invite message and would be used for all the subsequent signaling  
messages. However, the cost to
set up a direct connection for the subsequent signaling messages  
seems to outweigh the benefits,
at least in my mind.

What section 4.3 discusses is the question of whether the first type  
of connections should be dynamic.
This is possible, but the overhead in setting up a connection seems  
to outweigh any advantages in
making the first type of connections dynamic.

Hope this makes things clear.

From marc1.bailly at francetelecom.com  Wed Mar  1 10:35:08 2006
From: marc1.bailly at francetelecom.com (BAILLY Marc RD-CORE-LAN)
Date: Wed, 1 Mar 2006 16:35:08 +0100
Subject: [p2p-sip] New I-D: The effect of NATs on P2P
	SIPOverlay	Architecture
Message-ID: <BAF83494CE653943A97B9F755016A066028EB3C7@ftrdmel1.rd.francetelecom.fr>


Hi Philip, 

Comments and questions inline.

> >
> > sec 4.3: just to understand: is static connection == pointers in  
> > DHT routing table? and dynamic == call setup? Otherwise, what does  
> > static mean?
> 
> We see two types of connections.
> The first type is used to maintain the overlay, do lookups, and for  
> signaling. These follow the partial mesh
> pattern described in the paper. We see these as static, meaning that  
> they are long-lived (assuming
> the peers at either end are long-lived) and do not change as calls  
> are set up and torn down.
> Since these are just a partial mesh, messages flowing over these may  
> need a number of hops to reach
> their destination.
> The second type is used for the media in the call. These go from the  
> calling peer to the called peer
> and are dynamically set up when the call is established and 
> torn down  
> immediately afterwards.
> 
> It would be possible to have a third type. These would be set 
> up when  
> the called party receives the initial
> Invite message and would be used for all the subsequent signaling  
> messages. However, the cost to
> set up a direct connection for the subsequent signaling messages  
> seems to outweigh the benefits,
> at least in my mind.
> 

If I understand well, you suppose that all the SIP signaling messages
should be routed via the overlay connection rather than using the
overlay as a location service and then establish a SIP connection?

In section 4.1 : Mesh -- Once connected into the peer network, nodes
exchange
      messages with selected other peers periodically to keep NAT
      pinholes open.  SIP messages are either sent directly to the
      destination peer, or are sent indirectly via intermediate peers.

I guess that the last sentence of this paragraph mention this two
architecture possibilities?
If it's right, maybe we could make it clearer.

> What section 4.3 discusses is the question of whether the first type  
> of connections should be dynamic.
> This is possible, but the overhead in setting up a connection seems  
> to outweigh any advantages in
> making the first type of connections dynamic.

I think dynamic connections might be used as an optimization (i.e. set
up additional dynamic connections with people you're often calling). Is
it your vision of dynamic connections? (I understand it as Kundan did).

Regards

Marc


From philip_matthews at magma.ca  Wed Mar  1 11:06:57 2006
From: philip_matthews at magma.ca (Philip Matthews)
Date: Wed, 1 Mar 2006 11:06:57 -0500
Subject: [p2p-sip] New I-D: The effect of NATs on P2P
	SIPOverlay	Architecture
In-Reply-To: <BAF83494CE653943A97B9F755016A066028EB3C7@ftrdmel1.rd.francetelecom.fr>
References: <BAF83494CE653943A97B9F755016A066028EB3C7@ftrdmel1.rd.francetelecom.fr>
Message-ID: <BC8E2510-505B-4A96-B428-7BC6CBF07B93@magma.ca>

Hi Marc:

See replies inline.

- Philip

On 1-Mar-06, at 10:35 , BAILLY Marc RD-CORE-LAN wrote:

>
> Hi Philip,
>
> Comments and questions inline.
>
>>>
>>> sec 4.3: just to understand: is static connection == pointers in
>>> DHT routing table? and dynamic == call setup? Otherwise, what does
>>> static mean?
>>
>> We see two types of connections.
>> The first type is used to maintain the overlay, do lookups, and for
>> signaling. These follow the partial mesh
>> pattern described in the paper. We see these as static, meaning that
>> they are long-lived (assuming
>> the peers at either end are long-lived) and do not change as calls
>> are set up and torn down.
>> Since these are just a partial mesh, messages flowing over these may
>> need a number of hops to reach
>> their destination.
>> The second type is used for the media in the call. These go from the
>> calling peer to the called peer
>> and are dynamically set up when the call is established and
>> torn down
>> immediately afterwards.
>>
>> It would be possible to have a third type. These would be set
>> up when
>> the called party receives the initial
>> Invite message and would be used for all the subsequent signaling
>> messages. However, the cost to
>> set up a direct connection for the subsequent signaling messages
>> seems to outweigh the benefits,
>> at least in my mind.
>>
>
> If I understand well, you suppose that all the SIP signaling messages
> should be routed via the overlay connection rather than using the
> overlay as a location service and then establish a SIP connection?

There is another option which we forgot to mention in the paper:
use the overlay as a location service, and when peer A wants to call  
peer B,
peer A sends a ConnectionRequest message (via the overlay) to peer B.
This tells peer B to start trying to punch a hole through the NAT  
back to
peer A. Once the direct connection between peer A and B is established,
it is used for all SIP signaling.

When considering all these options, the important point to note is  
that peer B
needs to know that peer A wants to establish a connection, so B can  
open a
pinhole in its local NAT. The message conveying this information to  
peer B
must go through the overlay, because that is the only way to reach  
peer B.
Whether this message is a ConnectionRequest (as described above) or is
a SIP Invite message is something we can decide.


>
> In section 4.1 : Mesh -- Once connected into the peer network, nodes
> exchange
>       messages with selected other peers periodically to keep NAT
>       pinholes open.  SIP messages are either sent directly to the
>       destination peer, or are sent indirectly via intermediate peers.
>
> I guess that the last sentence of this paragraph mention this two
> architecture possibilities?
> If it's right, maybe we could make it clearer.

You are correct.
We can try to make this clearer in the next revision.

>
>> What section 4.3 discusses is the question of whether the first type
>> of connections should be dynamic.
>> This is possible, but the overhead in setting up a connection seems
>> to outweigh any advantages in
>> making the first type of connections dynamic.
>
> I think dynamic connections might be used as an optimization (i.e. set
> up additional dynamic connections with people you're often  
> calling). Is
> it your vision of dynamic connections? (I understand it as Kundan  
> did).
>

If peer A is frequently calling peer B, then peer A could establish a  
direct connection
to peer B. Most obviously this would be used for SIP signaling, but  
it could
perhaps be used for lookups as well.

Off the top of my head, I would guess is that A would have to be VERY  
frequently calling B (e.g., once
every few seconds or so) to make this really worth while, but there  
may be applications
where this will happen and peer A can figure this out in advance.

Part of the decision on whether this is worth it or not hinges on how  
many total
connections we think that a single peer can support. This is a  
question that Eric
and I have been trying to look for data on. Some people have  
suggested that
a peer could not support more than about 10 connections, while other  
think the
number might be more like 100 or 200. I think this will be an  
important number
for the group to investigate and decide on.

If the number is only 10 or so, then setting up a dedicated signaling  
connection like
this would be much harder to justify. However, if the number is 200  
or so, then this is more
reasonable.


From quittek at netlab.nec.de  Thu Mar  2 04:59:47 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Thu, 02 Mar 2006 10:59:47 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <156450F9DBF9883343076F10@[192.168.1.128]>

Dear all,

We tried to analyze the NAT and firewall problem in order to identify
issues that would need consideration when standardizing P2P SIP.

Basically we identified four issues, three for SIP signaling and one for
middlebox detection.  Please find them at

<http://www.ietf.org/internet-drafts/draft-quittek-p2p-sip-middlebox-00.txt>

Here is the abstract:

   Middleboxes, particularly firewalls and network address translators,
   are essential components of today's Internet infrastructure.  They
   are designed to support client-server applications well, but very
   often they are obstacles for peer-to-peer communication.  This memo
   discusses middlebox traversal issues of SIP-based peer-to-peer
   communication.  Required communication steps are analyzed concerning
   their potential constraints caused by middleboxes.  The requirements
   derived from this analysis are compared with the capabilities of
   currently available middlebox traversal methods and SIP signaling
   standards.  The comparison identifies open issues that need to be
   considered when designing standards for a SIP-based peer-to-peer
   communication infrastructure.

The memo contains the essential facts, but we think it needs a more
detailed elaboration in the next version.

Thanks,

    Juergen
-- 
Juergen Quittek        quittek at netlab.nec.de       Tel: +49 6221 90511-15
NEC Europe Ltd.,       Network Laboratories        Fax: +49 6221 90511-55
Kurfuersten-Anlage 36, 69115 Heidelberg, Germany   http://www.netlab.nec.de

From jdrosen at cisco.com  Thu Mar  2 10:14:41 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Thu, 02 Mar 2006 10:14:41 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <156450F9DBF9883343076F10@[192.168.1.128]>
References: <156450F9DBF9883343076F10@[192.168.1.128]>
Message-ID: <44070BE1.3030205@cisco.com>

So, I read this with much interest.

I disagree with many of the assumptions and the analysis performed by 
this document. First and foremost, I think it is a dire, dire, dire 
mistake should any kind of nat traversal solution be proposed for p2p 
sip which is not directly interoperable with the "regular SIP" nat 
traversal strategies. This draft seems to hint that something totally 
new is needed, though I wasnt sure.

Secondly, the draft says its an analysis of existing nat traversal 
techniques for sip. However, it omits the main SIP nat traversal 
techinque in its entirety - ICE and sip-outbound. I dont know how an 
analysis of the applicability of existing nat traversal techniques can 
be considered complete when it has missed the main standards in this area.

Indeed, ICE is basically p2p NAT traversal, and so I believe ice fits 
into a p2p framework extremely well. The only trick you'll need to 
figure out is discovering turn and stun servers, since those presumably 
will be functions provided by other peers in the network (ICE allows for 
any means of discovery of relay services and also allows usage of 
non-turn ones too). If p2p sip endpoints do not use ICE for nat 
traversal, then we will not have interopreability for nat traversal 
between regular sip and p2p-sip, and that is not acceptable IMHO.

The draft presumes that any nat traversal solution requires a step where 
the presence of nat is detected, and typed (section 2.1.2). I strongly 
disagree with this. The ice solution does not attempt to discover or 
otherwise characterize nats at all. Indeed, one of our main experiences 
with rfc3489 was that the whole notion of nat discovery and type 
computation was flawed. That is why this functionality was removed from 
rfc3489bis almost two years ago.

Section 4.2.1 asserts that middlebox detection is needed for non-UDP 
flows. I half agree with this. I agree that we need to make tcp work, 
and support configurations where there is no intermediary, even if its 
another user in the p2p network. The draft asserts that there are no 
solutions for that, but its just not true. ICE-tcp 
(draft-ietf-mmusic-ice-tcp) will allow for creation of direct tcp 
connections through nat as long as the nats allow it; if they don't 
it'll end up falling back on a connection through a relay. Note that 
this relay can be dicsovered by any means, and thus can be another user 
on the p2p network.

Thanks,
Jonathan R.

Juergen Quittek wrote:

> Dear all,
> 
> We tried to analyze the NAT and firewall problem in order to identify
> issues that would need consideration when standardizing P2P SIP.
> 
> Basically we identified four issues, three for SIP signaling and one for
> middlebox detection.  Please find them at
> 
> <http://www.ietf.org/internet-drafts/draft-quittek-p2p-sip-middlebox-00.txt>
> 
> Here is the abstract:
> 
>    Middleboxes, particularly firewalls and network address translators,
>    are essential components of today's Internet infrastructure.  They
>    are designed to support client-server applications well, but very
>    often they are obstacles for peer-to-peer communication.  This memo
>    discusses middlebox traversal issues of SIP-based peer-to-peer
>    communication.  Required communication steps are analyzed concerning
>    their potential constraints caused by middleboxes.  The requirements
>    derived from this analysis are compared with the capabilities of
>    currently available middlebox traversal methods and SIP signaling
>    standards.  The comparison identifies open issues that need to be
>    considered when designing standards for a SIP-based peer-to-peer
>    communication infrastructure.
> 
> The memo contains the essential facts, but we think it needs a more
> detailed elaboration in the next version.
> 
> Thanks,
> 
>     Juergen

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From eunsoo at research.panasonic.com  Thu Mar  2 13:15:08 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Thu, 02 Mar 2006 13:15:08 -0500
Subject: [p2p-sip] I-D: An Architecture for P2P SIP
In-Reply-To: <4404BE75.9080908@research.panasonic.com>
References: <440389B2.5040801@research.panasonic.com> 
	<4404A1B4.10307@cs.columbia.edu>
	<4404BE75.9080908@research.panasonic.com>
Message-ID: <4407362C.6080808@research.panasonic.com>

Kundan, thanks for your comments.
Please see my comment about the record format inline.

Greg Daley wrote:

>Hi Kundan,
>
>I liked your tech report.
>
>Kundan Singh wrote:
>  
>
>>Some comments and feedback:
>>
>>- Seems like you propose to do P2P lookup first and then fall back to 
>>DNS (RFC3263). But since P2P lookup is slower, the reverse is probably 
>>more desirable, i.e., if DNS SRV/NAPTR exists, try that first. (or for 
>>INVITE do parallel forking to DNS and P2P destinations).
>>    
>>
>
>I take your point, although there may be other reasons to try
>P2P first rather than go for CS/SIP related DNS infrastructure
>(for example, cost, or more accurate local lookup information).
>
>
>  
>
>>- sec 8: regarding the format of records: It is probably better to reuse 
>>the existing SIP Contact header format. Since implementations will do 
>>SIP parsing anyway, and having multiple formats is just more work for 
>>translating from one to another. Thus instead of splitting IP, port, 
>>transport, etc., e.g., just put 
>><sip:user at 178.14.234.21:5060;transport=udp> if it supports only UDP, and 
>>ignore transport= param if it supports both. Multiple ports can be 
>>specified using multiple contacts. (appendix A of my tech report 
>>http://mice.cs.columbia.edu/getTechreport.php?techreportID=388 has 
>>another example of the data format for P2P-SIP).
>>    
>>
>
>Something very explicitly stated in the draft is the fact that we
>want to have a different (and possibly quite distinct) protocol layer
>undertaking the protocol operations for the P2P lookups.
>
>We may be able to make use of existing formats so long as this
>doesn't cause confusion about which components belong to which
>layer.
>  
>
Since the record will be interpreted by SIP layer, your suggestion to 
use the SIP Contact header format sounds reasonable also to me though I 
think the  simple format in the draft won't add much complexity.
We are quite open about the record formats. I just hope the formats are 
simple and flexible enough.

>  
>
>>Also a minor nit: calling it a 'resource record' may be confusing it 
>>with DNS RR.
>>    
>>
>
>Suggestions are welcome.
>
>  
>
>>- I am not quite clear about the purpose of 'login server'. Is this 
>>contacted every time the phone is started (like Skype's login server), 
>>in which case it is clearly not desirable unless it can be made p2p or 
>>distributed. Or is it like a certifying authority that is contacted only 
>>(once?) for issuing or renewing the certificate.
>>    
>>
>
>It may yet be too early to say, but may have some preference not
>to contact a server each time.  I'm actually not sure what
>other systems do right now with regard to login (few people are).
>
>Some people may think it is valuable to explicitly authorize
>participants to join the network.
>
>  
>
>>Minor corrections:
>>
>>- page 15: typo: ...peers may be able [to] use...
>>
>>- sec 5.3: I think we should add TLS (either separately or same line as 
>>TCP).
>>    
>>
>
>UDP and TCP can both use TLS/dTLS.  We describe transport security
>in the security considerations section.
>
>So we're not saying "in the clear" TCP.
>
>Once again, it may be too early to say.
>
>  
>
>>- 5.4: incomplete sentence 'After the association'
>>
>>- 5.4: '...credentials [MAY=>MUST] be needed...' I think. Similarly, 
>>'... the ON [will wish to=>MUST] authenticate the SN...
>>    
>>
>
>We need to go through these requirements individually.
>Within the authors team we had these discussions, and
>made the terms flexible for the initial description.
>
>  
>
>>- 7: typo: befind=>behind
>>
>>- 11.6: recorrds=>records
>>
>>- 12: as per my record 'Salman [Basset=>Baset]'
>>
>>    
>>
My apologies to Salman.

Thanks.

Eunsoo


From eunsoo at research.panasonic.com  Thu Mar  2 14:21:37 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Thu, 02 Mar 2006 14:21:37 -0500
Subject: [p2p-sip] I-D: An Architecture for P2P SIP
In-Reply-To: <4404A10D.1010709@Research.Panasonic.COM>
References: <BAF83494CE653943A97B9F755016A066028EAAF9@ftrdmel1.rd.francetele
	com.fr> <4404A10D.1010709@Research.Panasonic.COM>
Message-ID: <440745C1.3060201@research.panasonic.com>

|| That's why I'm wondering if it could be possible to use NAT traversal mechanisms in the P2P overlay in order to be able to || include NATed nodes in it (I know this is not a SIP problem but it has impact on the P2P SIP architecture).
|| Is it possible with a DHT? If possible, what are the impacts on the DHT efficiency?


I think it is possible to build a DHT including NATed nodes.
We did not consider it for simplicity.
Added complexity, in particular, for the DHT join process, seems to be a 
bigger concern than any decrease in lookup efficiency.
Philip asked about it in p2p-hackers mailing list and some people talked 
about on-going work.
Thanks.

Eunsoo


Sathya Narayanan wrote:

>>I have some questions about the supernode selection criteria. It is mentionned in your draft that "it is recommended that SNs should be on the public internet" in order to receive P2P messages.
>> 
>>
>>    
>>
>The text in the first half of this sentence provides the context to this
>recommendation. The complete line is,
>
>   When a P2P overlay network spans
>   over the global Internet and contains peers private IP addresses as
>   well as peers with public IP addresses, SNs need to perform STUN, and
>   possibly TURN server operations thus it is RECOMMENDED that SNs
>   should be on the public Internet, and publicly addressed [5][10].
>
>In this line, we are specifically talking about a P2P overlay spanning
>the global Internet. For use cases where the P2P overlay is within a
>private network, as pointed out my Phillip, this recommendation doesn't
>apply. The only mandatory requirement is,
>
>      1) MUST be able to receive messages on predetermined protocols
>      and ports from other SNs on the overlay network.
>
>If all the nodes are within a private network, they must be able to
>exchange messsages without any problem and most of the nodes can become
>a SN.
>
>We struggled with the wording for this requirement to make it applicable
>to all  use-cases. If somebody has a better suggestion, we will be happy
>to revise the text.
>
>- Sathya
>
>  
>
>>>-----Message d'origine-----
>>>De : p2p-sip-bounces at cs.columbia.edu 
>>>[mailto:p2p-sip-bounces at cs.columbia.edu] De la part de Eunsoo Shim
>>>Envoy? : mardi 28 f?vrier 2006 00:22
>>>? : P2P-SIP
>>>Objet : [p2p-sip] I-D: An Architecture for P2P SIP
>>>
>>>We submitted an I-D that describes an architecture for P2P SIP.
>>>You can get it from the link below until it appears in the IETF I-D 
>>>directory.
>>>We hope this draft can help the architecture and charter 
>>>discussion in BOF.
>>>Your comments will be appreciated.
>>>Thanks.
>>>
>>>Eunsoo
>>>
>>>http://www.comet.columbia.edu/~eunsoo/ietf/draft-shim-sipping-
>>>p2p-arch-00.txt
>>>
>>>Abstract
>>>
>>>  This document describes an architecture for peer-to-peer 
>>>SIP systems
>>>  in which a separate and independent peer-to-peer overlay layer
>>>  provides a distributed resource placement and search 
>>>service for SIP.
>>>  It also aims to help identify what should be specified for
>>>  interoperation.
>>>
>>>
>>>
>>>
>>>
>>>_______________________________________________
>>>p2p-sip mailing list
>>>p2p-sip at cs.columbia.edu
>>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>
>>>   
>>>
>>>      
>>>
>>_______________________________________________
>>p2p-sip mailing list
>>p2p-sip at cs.columbia.edu
>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>> 
>>
>>    
>>
>
>
>  
>



From bryan at ethernot.org  Fri Mar  3 16:19:44 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Fri, 3 Mar 2006 16:19:44 -0500
Subject: [p2p-sip] Updated BOF date and time
Message-ID: <8b2769930603031319o6e30cac7sefcccb4ea308fb13@mail.gmail.com>

All,

The actual BOF is now currently scheduled for Wed, March 22 from
9-11:30 am. There is still some chance this will change, since there
were a few scheduling conflicts that might require shuffling, and we
could be a casualty. Stay tuned.

At this point it does appears we will still have an ad-hoc meeting
slot on Friday, so if you are planning to come to IETF to discuss P2P
SIP, don't leave after the BOF. The exact time of the ad-hoc meeting
is still being nailed down. We'll let you know as soon we hear.

Draft agendas for both meetings are on p2psip.org/ietf65.php

David Bryan


From karstbj at yahoo.com  Sat Mar  4 02:30:42 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Fri, 3 Mar 2006 23:30:42 -0800 (PST)
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <44070BE1.3030205@cisco.com>
Message-ID: <20060304073042.93284.qmail@web53015.mail.yahoo.com>



--- Jonathan Rosenberg <jdrosen at cisco.com> wrote:

> Secondly, the draft says its an analysis of existing
> nat traversal 
> techniques for sip. However, it omits the main SIP
> nat traversal 
> techinque in its entirety - ICE and sip-outbound. I
> dont know how an 
> analysis of the applicability of existing nat
> traversal techniques can 
> be considered complete when it has missed the main
> standards in this area.

The draft also fails to mention IPv6. I believe this
is a bit short-sighted. Microsoft is making a big push
for IPv6 in Windows Vista. Vista will include an
implementation of Teredo (RFC 4380), which provides
some pretty interesting NAT traversal facilities.

-- Karst

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From dbarrett at quinthar.com  Sat Mar  4 03:12:44 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Sat, 4 Mar 2006 00:12:44 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <20060304073042.93284.qmail@web53015.mail.yahoo.com>
Message-ID: <200603040813.k248Cp3o012928@cs.columbia.edu>

Hi Karst, 

Do you know if there is any intent for Teredo to support symmetric NATs?
I'd suggest this be a base requirement for P2P-SIP.

-david

> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-
> bounces at cs.columbia.edu] On Behalf Of Karst Bjorgson
> Sent: Friday, March 03, 2006 11:31 PM
> To: Jonathan Rosenberg; Juergen Quittek
> Cc: P2P-SIP
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> 
> 
> --- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
> 
> > Secondly, the draft says its an analysis of existing
> > nat traversal
> > techniques for sip. However, it omits the main SIP
> > nat traversal
> > techinque in its entirety - ICE and sip-outbound. I
> > dont know how an
> > analysis of the applicability of existing nat
> > traversal techniques can
> > be considered complete when it has missed the main
> > standards in this area.
> 
> The draft also fails to mention IPv6. I believe this
> is a bit short-sighted. Microsoft is making a big push
> for IPv6 in Windows Vista. Vista will include an
> implementation of Teredo (RFC 4380), which provides
> some pretty interesting NAT traversal facilities.
> 
> -- Karst
> 
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From jdrosen at cisco.com  Sat Mar  4 10:20:56 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Sat, 04 Mar 2006 10:20:56 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <200603040813.k248Cp3o012928@cs.columbia.edu>
References: <200603040813.k248Cp3o012928@cs.columbia.edu>
Message-ID: <4409B058.8000501@cisco.com>

Its important to note that teredo fits fine into the ice framework. For 
a v4 host it allows it to get a v6 address that can be offered as a 
candidate.

-Jonathan R.

David Barrett wrote:

> Hi Karst, 
> 
> Do you know if there is any intent for Teredo to support symmetric NATs?
> I'd suggest this be a base requirement for P2P-SIP.
> 
> -david
> 
> 
>>-----Original Message-----
>>From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-
>>bounces at cs.columbia.edu] On Behalf Of Karst Bjorgson
>>Sent: Friday, March 03, 2006 11:31 PM
>>To: Jonathan Rosenberg; Juergen Quittek
>>Cc: P2P-SIP
>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>>
>>
>>--- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>>
>>
>>>Secondly, the draft says its an analysis of existing
>>>nat traversal
>>>techniques for sip. However, it omits the main SIP
>>>nat traversal
>>>techinque in its entirety - ICE and sip-outbound. I
>>>dont know how an
>>>analysis of the applicability of existing nat
>>>traversal techniques can
>>>be considered complete when it has missed the main
>>>standards in this area.
>>
>>The draft also fails to mention IPv6. I believe this
>>is a bit short-sighted. Microsoft is making a big push
>>for IPv6 in Windows Vista. Vista will include an
>>implementation of Teredo (RFC 4380), which provides
>>some pretty interesting NAT traversal facilities.
>>
>>-- Karst
>>
>>__________________________________________________
>>Do You Yahoo!?
>>Tired of spam?  Yahoo! Mail has the best spam protection around
>>http://mail.yahoo.com
>>_______________________________________________
>>p2p-sip mailing list
>>p2p-sip at cs.columbia.edu
>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From eunsoo at research.panasonic.com  Sat Mar  4 10:30:19 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Sat, 04 Mar 2006 10:30:19 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <200603040813.k248Cp3o012928@cs.columbia.edu>
References: <200603040813.k248Cp3o012928@cs.columbia.edu>
Message-ID: <4409B28B.2000004@research.panasonic.com>

I also think that support for symmetric NATs is a must.
The requirements draft can be revised to state this explicitly.
Thanks.

Eunsoo
 
David Barrett wrote:

>Hi Karst, 
>
>Do you know if there is any intent for Teredo to support symmetric NATs?
>I'd suggest this be a base requirement for P2P-SIP.
>
>-david
>
>  
>
>>-----Original Message-----
>>From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-
>>bounces at cs.columbia.edu] On Behalf Of Karst Bjorgson
>>Sent: Friday, March 03, 2006 11:31 PM
>>To: Jonathan Rosenberg; Juergen Quittek
>>Cc: P2P-SIP
>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>>
>>
>>--- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>>
>>    
>>
>>>Secondly, the draft says its an analysis of existing
>>>nat traversal
>>>techniques for sip. However, it omits the main SIP
>>>nat traversal
>>>techinque in its entirety - ICE and sip-outbound. I
>>>dont know how an
>>>analysis of the applicability of existing nat
>>>traversal techniques can
>>>be considered complete when it has missed the main
>>>standards in this area.
>>>      
>>>
>>The draft also fails to mention IPv6. I believe this
>>is a bit short-sighted. Microsoft is making a big push
>>for IPv6 in Windows Vista. Vista will include an
>>implementation of Teredo (RFC 4380), which provides
>>some pretty interesting NAT traversal facilities.
>>
>>-- Karst
>>
>>__________________________________________________
>>Do You Yahoo!?
>>Tired of spam?  Yahoo! Mail has the best spam protection around
>>http://mail.yahoo.com
>>_______________________________________________
>>p2p-sip mailing list
>>p2p-sip at cs.columbia.edu
>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>    
>>
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From edpimentl at gmail.com  Sat Mar  4 10:40:02 2006
From: edpimentl at gmail.com (Edpimentl)
Date: Sat, 04 Mar 2006 10:40:02 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4409B058.8000501@cisco.com>
References: <200603040813.k248Cp3o012928@cs.columbia.edu>
	<4409B058.8000501@cisco.com>
Message-ID: <4409B4D2.4090105@gmail.com>

Good comments, I like to point you to the Open-Source Miredo project
which is a Teredo Server
http://www.simphalempin.com/dev/miredo/
http://freshmeat.net/projects/miredo/?branch_id=51002&release_id=211721

We are planning to incorporate Miredo - Teredo Server with OpenSER.

Best.
E Pimentel
http://googaya.com

Jonathan Rosenberg wrote:

>Its important to note that teredo fits fine into the ice framework. For 
>a v4 host it allows it to get a v6 address that can be offered as a 
>candidate.
>
>-Jonathan R.
>
>David Barrett wrote:
>
>  
>
>>Hi Karst, 
>>
>>Do you know if there is any intent for Teredo to support symmetric NATs?
>>I'd suggest this be a base requirement for P2P-SIP.
>>
>>-david
>>
>>
>>    
>>
>>>-----Original Message-----
>>>From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-
>>>bounces at cs.columbia.edu] On Behalf Of Karst Bjorgson
>>>Sent: Friday, March 03, 2006 11:31 PM
>>>To: Jonathan Rosenberg; Juergen Quittek
>>>Cc: P2P-SIP
>>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>>
>>>
>>>
>>>--- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>>>
>>>
>>>      
>>>
>>>>Secondly, the draft says its an analysis of existing
>>>>nat traversal
>>>>techniques for sip. However, it omits the main SIP
>>>>nat traversal
>>>>techinque in its entirety - ICE and sip-outbound. I
>>>>dont know how an
>>>>analysis of the applicability of existing nat
>>>>traversal techniques can
>>>>be considered complete when it has missed the main
>>>>standards in this area.
>>>>        
>>>>
>>>The draft also fails to mention IPv6. I believe this
>>>is a bit short-sighted. Microsoft is making a big push
>>>for IPv6 in Windows Vista. Vista will include an
>>>implementation of Teredo (RFC 4380), which provides
>>>some pretty interesting NAT traversal facilities.
>>>
>>>-- Karst
>>>
>>>__________________________________________________
>>>Do You Yahoo!?
>>>Tired of spam?  Yahoo! Mail has the best spam protection around
>>>http://mail.yahoo.com
>>>_______________________________________________
>>>p2p-sip mailing list
>>>p2p-sip at cs.columbia.edu
>>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>      
>>>
>>_______________________________________________
>>p2p-sip mailing list
>>p2p-sip at cs.columbia.edu
>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
>>    
>>
>
>  
>


-- 

ENUM:   878107472467468
Mail:   edpimentl[at]gmail[dot]com
Web:    http://agileco.net
WebLog: http://googaya.com
IM:     [AOL | Jabber | Yahoo | MSN ] edpimentl
Voip:   [SKype | AgileSIP | GoogleTalk | FWD(14272) ] edpimentl



From dedinski at fmi.uni-passau.de  Thu Mar  2 05:25:56 2006
From: dedinski at fmi.uni-passau.de (Ivan Dedinski)
Date: Thu, 02 Mar 2006 11:25:56 +0100
Subject: [p2p-sip] Re.  NAT and Overlay Maintenance
Message-ID: <4406C834.8010008@fmi.uni-passau.de>

Hi,

I would like to draw your attention to a concept called AVP (Active 
Virtual Peer), some work on it was published in 2003:

H. De Meer, K. Tutschku, P. Tran-Gia
Dynamic Operation of Peer-to-Peer Overlay Networks
PIK (Praxis der Informationsverarbeitung und Kommunikation)
pages: 65-73, Volume: 26, No. 2, April-June 2003

The AVP concept could be eventually used to overcome the NAT problem by 
integrating some peers behind a NAT box together with globally reachable 
peers into virtual peers, which then do the overlay management. An 
advantage of this approach would be that the (few) globally reachable 
peers could gain support in terms of storage ressources and processing 
power from the (many) peers behind the NAT boxes. Also a VP instance 
behind a NAT box can easily serve the peers behind the same NAT box.

best regards

Ivan


From lowekamp at cs.wm.edu  Sun Mar  5 22:48:44 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Sun, 5 Mar 2006 22:48:44 -0500
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
Message-ID: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>

We have submitted an updated version of the draft "A P2P Approach to
SIP Registration and Resource Location", draft-bryan-sipping-p2p-02.
Until it is available in the archive, you can find it here:

http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html

or here

http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.txt

depending on the format you like.

David Bryan
Bruce Lowekamp
Cullen Jennings


From karstbj at yahoo.com  Mon Mar  6 01:36:51 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Sun, 5 Mar 2006 22:36:51 -0800 (PST)
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4409B28B.2000004@research.panasonic.com>
Message-ID: <20060306063651.52150.qmail@web53006.mail.yahoo.com>

> >Do you know if there is any intent for Teredo to
> support symmetric NATs?
> >I'd suggest this be a base requirement for P2P-SIP.

I am pretty sure that a Teredo implementation could be
tuned to work through a lot of symmetric NAT, or at
least in the symmetric to non-symmetric cases. But, as
Jonathan said, IPv6 should be supported as an option.
The combination of IPv4 and IPv6 will work more often
than IPv4 alone.

-- Karst

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From tutschku at informatik.uni-wuerzburg.de  Mon Mar  6 05:04:36 2006
From: tutschku at informatik.uni-wuerzburg.de (Kurt Tutschku)
Date: Mon, 6 Mar 2006 11:04:36 +0100
Subject: [p2p-sip] New paper about Skype
In-Reply-To: <BAF83494CE653943A97B9F755016A0660284D509@ftrdmel1.rd.francetelecom.fr>
Message-ID: <20060306101319.93A976F58E@nero.informatik.uni-wuerzburg.de>

Dear all,
 
perhaps a bit off topic. We have published a paper on the traffic
characteristic and voice quality of the Skype voice
service. You can access the paper at
 
http://www-info3.informatik.uni-wuerzburg.de/papers/conf_217.pdf
 
The login and the password is the same. It is:
 
ls3pubs
 
Best Regards
Kurt



  _____  

Von: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] Im Auftrag von BAILLY Marc
RD-CORE-LAN
Gesendet: Mittwoch, 22. Februar 2006 14:06
An: David A. Bryan
Cc: p2p-sip at cs.columbia.edu
Betreff: [p2p-sip] New paper about Skype




Hi, 

I just would like to mention a new paper about Skype analysis that could be
added to the "Research and Reading" section of p2psip.org

There are some interesting statistics about Skype p2p network. 
 <http://saikat.guha.cc/pub/iptps06-skype.pdf>
http://saikat.guha.cc/pub/iptps06-skype.pdf 

Marc 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060306/f7f53a23/attachment.html

From slavitch at objectworld.com  Mon Mar  6 12:48:27 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Mon, 6 Mar 2006 12:48:27 -0500
Subject: [p2p-sip] [Sip] Myth Busters: Episode 1,
	Is STUN request processing faster than a SIP PING
	requestprocessing? (P2PSIP possibilities)
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C598@photon.objectworld.com>

Hisham;

Love the code and the numbers. We'll try our luck here and compare. This
is the first time in a long time that our engineers and our testers were
equally happy (and not grumpy) after I forwarded them an email from an
IETF WG list.  

I hope that this is the start of a trend. It allows us to ask and answer
firm questions with real data in real implementations.

On another note, if 'it was not really part of SIP', but part of a 'P2P
Overlay mechanism' a lot of difficult things would begin to go away,
especially if part of the capability of P2PSIP offered a client adaptor
service for low-function SIP user agents or proxies. For that reason I'm
forwarding this email to the p2psip list, as people there will have an
interest in this discussion and especially the sample code and the test
results.

http://mice.cs.columbia.edu/getTechreport.php?techreportID=388  has an
interesting description of client adaptors in a peer to peer
environment, something that would offer great benefit if implemented in
P2PSIP.


Regards

Michael


Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: Bharrat, Shaun [mailto:SBharrat at sonusnet.com] 
Sent: March 6, 2006 10:35 AM
To: Hisham Khartabil; sip at ietf.org
Cc: Alan Duric; Dusan Stevanovic
Subject: RE: [Sip] Myth Busters: Episode 1,Is STUN request processing
faster than a SIP PING requestprocessing?

Hisham:

Glad to see actual code and test results. It is a _welcome_ sight to
have full information and numbers based on real code.

Now that I understand what is actually being done, I guess my question
is why you would have expected the numbers to be any different? Please
correct me if I am misunderstanding anything in what you have done. This
is not doing ANY sip processing - it just finds the first \r\n and then
copies everything beyond that (with a 200 + contact header prepended)
into the response. There is no VIA processing (guess we don't need to
add received= or rport?), no Content-Length processing (how does it work
for TCP?), etc.

If we are "allowed" to skip the SIP processing (i.e. it's not really a
SIP packet), then I agree that we can make the processing time of a PING
arbitrarily close to a STUN packet processing. However, if we are not, I
don't agree that this is a valid comparison of a SIP-based keepalive
versus a STUN based keepalive.

My 2c

Thanks and cheers,
Shaun

www.sonusnet.com

> -----Original Message-----
> From: Hisham Khartabil [mailto:hisham.khartabil at telio.no]
> Sent: Monday, March 06, 2006 10:02 AM
> To: sip at ietf.org List
> Cc: Alan Duric; Dusan Stevanovic
> Subject: [Sip] Myth Busters: Episode 1,Is STUN request processing 
> faster than a SIP PING requestprocessing?
> 
> 
> I promised to come back with hard real numbers on how long it will 
> take for a server to process and STUN request vs a SIP PING request
> 
> Myth: processing a STUN request is 10 to 100 times faster than a SIP 
> PING request
> 
> Assumptions:  - SIP PING request is processed early in the stack and 
> bypasses any unnecessary validations.
>                - SIP PING response carries in the contact header the 
> IP address and port as appears in the IP packet (source IP and port).
> 
> We implemented the PING request as described above in the reSIProcate 
> stack. Attached you will find the c++ file where the changes were 
> made.
> STUN was already implemented in the reSIProcate stack and used as is.
> 
> Below you will see the findings:
> 
> STUN request processing: 0.15 - 0.2 ms SIP PING request processing: 
> 0.1 - 0.2 ms SIP OPTIONS request: 7 ms SIP REGISTER request: 7 ms
> 
> The measurement was taken between the time a request appeared on 
> ethereal and its response appearing on ethereal. Average was taken 
> manually.
> 
> As you can see, SIP PING requests were sometimes faster to process 
> that STUN requests, but never slower. Actually the SIP PING requests 
> on average were faster to process.
> 
> Myth Busted!!!! :)
> 
> Regards,
> Hisham
> P.S. It took 15 mins to modify the sip stack to do the PING response.
> 
> 

_______________________________________________
Sip mailing list  https://www1.ietf.org/mailman/listinfo/sip
This list is for NEW development of the core SIP Protocol Use
sip-implementors at cs.columbia.edu for questions on current sip Use
sipping at ietf.org for new developments on the application of sip


--


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From fred at perspectiveip.ca  Mon Mar  6 19:33:53 2006
From: fred at perspectiveip.ca (Frederick Lefebvre)
Date: Mon, 6 Mar 2006 19:33:53 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <20060306063651.52150.qmail@web53006.mail.yahoo.com>
References: <4409B28B.2000004@research.panasonic.com>
	<20060306063651.52150.qmail@web53006.mail.yahoo.com>
Message-ID: <3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>

What about mandating the use IPv6 and leave the NAT traversal issue to
the IPv6 transition mechanism used?  Wouldn't it make the
implementation of p2p-sip easier if the protocol did not have to work
around NATs?
Of course, feel free to wonder if this is a realistic proposal or not.
 But with the upcoming Windows Vista, all of a sudden a fairly large
percentage of users will have V6 connectivity and it might not sound
so crazy anymore.

-- Fred

On 3/6/06, Karst Bjorgson <karstbj at yahoo.com> wrote:
> > >Do you know if there is any intent for Teredo to
> > support symmetric NATs?
> > >I'd suggest this be a base requirement for P2P-SIP.
>
> I am pretty sure that a Teredo implementation could be
> tuned to work through a lot of symmetric NAT, or at
> least in the symmetric to non-symmetric cases. But, as
> Jonathan said, IPv6 should be supported as an option.
> The combination of IPv4 and IPv6 will work more often
> than IPv4 alone.
>
> -- Karst
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From slavitch at gmail.com  Mon Mar  6 19:52:59 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 6 Mar 2006 19:52:59 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>
References: <4409B28B.2000004@research.panasonic.com>
	<20060306063651.52150.qmail@web53006.mail.yahoo.com>
	<3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>
Message-ID: <1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>

Are you serious? How many ISP's offer IPV6?  How many organizations outside
academe and a few large enterprises?
Do you realize how many people in the real world still use Windows 2000 and
Windows 98, including some very large organizations?
Given past history Vista will  run on the majority of Windows desktop
systems starting around  2010-2011, and that is optimistic.

In the meantime, Skype works now.

If P2PSIP is limited to IPV6 we might as well call it a day right now and
start licensing runtime modules from Skype.


On 3/6/06, Frederick Lefebvre <fred at perspectiveip.ca> wrote:
>
> What about mandating the use IPv6 and leave the NAT traversal issue to
> the IPv6 transition mechanism used?  Wouldn't it make the
> implementation of p2p-sip easier if the protocol did not have to work
> around NATs?
> Of course, feel free to wonder if this is a realistic proposal or not.
> But with the upcoming Windows Vista, all of a sudden a fairly large
> percentage of users will have V6 connectivity and it might not sound
> so crazy anymore.
>
> -- Fred
>
> On 3/6/06, Karst Bjorgson <karstbj at yahoo.com> wrote:
> > > >Do you know if there is any intent for Teredo to
> > > support symmetric NATs?
> > > >I'd suggest this be a base requirement for P2P-SIP.
> >
> > I am pretty sure that a Teredo implementation could be
> > tuned to work through a lot of symmetric NAT, or at
> > least in the symmetric to non-symmetric cases. But, as
> > Jonathan said, IPv6 should be supported as an option.
> > The combination of IPv4 and IPv6 will work more often
> > than IPv4 alone.
> >
> > -- Karst
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>



--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060306/8ed7df32/attachment.html

From eunsoo at research.panasonic.com  Mon Mar  6 21:33:22 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Mon, 06 Mar 2006 21:33:22 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
References: <4409B28B.2000004@research.panasonic.com> 
	<"20060306063651.52150.q mail"@web53006.mail.yahoo.com> 
	<1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
Message-ID: <440CF0F2.8060702@research.panasonic.com>

I strongly believe support for IPv4 is a must.
Thanks.

Eunsoo

Michael Slavitch wrote:

> Are you serious? How many ISP's offer IPV6?  How many organizations
> outside academe and a few large enterprises?
> Do you realize how many people in the real world still use Windows
> 2000 and Windows 98, including some very large organizations?
> Given past history Vista will  run on the majority of Windows desktop
> systems starting around  2010-2011, and that is optimistic.
>
> In the meantime, Skype works now.
>
> If P2PSIP is limited to IPV6 we might as well call it a day right now
> and start licensing runtime modules from Skype.
>
>
> On 3/6/06, * Frederick Lefebvre* <fred at perspectiveip.ca
> <mailto:fred at perspectiveip.ca>> wrote:
>
>     What about mandating the use IPv6 and leave the NAT traversal issue to
>     the IPv6 transition mechanism used?  Wouldn't it make the
>     implementation of p2p-sip easier if the protocol did not have to work
>     around NATs?
>     Of course, feel free to wonder if this is a realistic proposal or not.
>     But with the upcoming Windows Vista, all of a sudden a fairly large
>     percentage of users will have V6 connectivity and it might not sound
>     so crazy anymore.
>
>     -- Fred
>
>     On 3/6/06, Karst Bjorgson <karstbj at yahoo.com
>     <mailto:karstbj at yahoo.com>> wrote:
>     > > >Do you know if there is any intent for Teredo to
>     > > support symmetric NATs?
>     > > >I'd suggest this be a base requirement for P2P-SIP.
>     >
>     > I am pretty sure that a Teredo implementation could be
>     > tuned to work through a lot of symmetric NAT, or at
>     > least in the symmetric to non-symmetric cases. But, as
>     > Jonathan said, IPv6 should be supported as an option.
>     > The combination of IPv4 and IPv6 will work more often
>     > than IPv4 alone.
>     >
>     > -- Karst
>     >
>     > __________________________________________________
>     > Do You Yahoo!?
>     > Tired of spam?  Yahoo! Mail has the best spam protection around
>     > http://mail.yahoo.com
>     > _______________________________________________
>     > p2p-sip mailing list
>     > p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
>     > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>     <https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip>
>     >
>
>     _______________________________________________
>     p2p-sip mailing list
>     p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
>     https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>
>
>
> -- 
> Michael Slavitch
> Ottawa, Ontario Canada
>
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From RBrown at talkswitch.com  Mon Mar  6 21:39:57 2006
From: RBrown at talkswitch.com (Rob Brown)
Date: Mon, 6 Mar 2006 21:39:57 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <FA4B9088F1833A4FAAAA5BBB3A20A43B071040@omega.ctpdomain.ctp>

Agreed.  IPv4 must be supported.

Regards,
Robert Brown
TalkSwitch



> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu 
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Eunsoo Shim
> Sent: March 6, 2006 9:33 PM
> To: Michael Slavitch
> Cc: P2P-SIP
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> I strongly believe support for IPv4 is a must.
> Thanks.
> 
> Eunsoo
> 
> Michael Slavitch wrote:
> 
> > Are you serious? How many ISP's offer IPV6?  How many organizations 
> > outside academe and a few large enterprises?
> > Do you realize how many people in the real world still use Windows 
> > 2000 and Windows 98, including some very large organizations?
> > Given past history Vista will  run on the majority of 
> Windows desktop 
> > systems starting around  2010-2011, and that is optimistic.
> >
> > In the meantime, Skype works now.
> >
> > If P2PSIP is limited to IPV6 we might as well call it a day 
> right now 
> > and start licensing runtime modules from Skype.
> >
> >
> > On 3/6/06, * Frederick Lefebvre* <fred at perspectiveip.ca 
> > <mailto:fred at perspectiveip.ca>> wrote:
> >
> >     What about mandating the use IPv6 and leave the NAT 
> traversal issue to
> >     the IPv6 transition mechanism used?  Wouldn't it make the
> >     implementation of p2p-sip easier if the protocol did 
> not have to work
> >     around NATs?
> >     Of course, feel free to wonder if this is a realistic 
> proposal or not.
> >     But with the upcoming Windows Vista, all of a sudden a 
> fairly large
> >     percentage of users will have V6 connectivity and it 
> might not sound
> >     so crazy anymore.
> >
> >     -- Fred
> >
> >     On 3/6/06, Karst Bjorgson <karstbj at yahoo.com
> >     <mailto:karstbj at yahoo.com>> wrote:
> >     > > >Do you know if there is any intent for Teredo to
> >     > > support symmetric NATs?
> >     > > >I'd suggest this be a base requirement for P2P-SIP.
> >     >
> >     > I am pretty sure that a Teredo implementation could be
> >     > tuned to work through a lot of symmetric NAT, or at
> >     > least in the symmetric to non-symmetric cases. But, as
> >     > Jonathan said, IPv6 should be supported as an option.
> >     > The combination of IPv4 and IPv6 will work more often
> >     > than IPv4 alone.
> >     >
> >     > -- Karst
> >     >
> >     > __________________________________________________
> >     > Do You Yahoo!?
> >     > Tired of spam?  Yahoo! Mail has the best spam 
> protection around
> >     > http://mail.yahoo.com
> >     > _______________________________________________
> >     > p2p-sip mailing list
> >     > p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
> >     > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >     <https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip>
> >     >
> >
> >     _______________________________________________
> >     p2p-sip mailing list
> >     p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
> >     https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
> >
> >
> > --
> > Michael Slavitch
> > Ottawa, Ontario Canada
> >
> >-------------------------------------------------------------
> ----------
> >-
> >
> >_______________________________________________
> >p2p-sip mailing list
> >p2p-sip at cs.columbia.edu
> >https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >  
> >
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From tme at multicasttech.com  Mon Mar  6 21:49:01 2006
From: tme at multicasttech.com (Marshall Eubanks)
Date: Mon, 6 Mar 2006 21:49:01 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <FA4B9088F1833A4FAAAA5BBB3A20A43B071040@omega.ctpdomain.ctp>
References: <FA4B9088F1833A4FAAAA5BBB3A20A43B071040@omega.ctpdomain.ctp>
Message-ID: <3815E04A-24EC-4BA4-B41E-0DEEFF1A01B4@multicasttech.com>

I agree too. Anyone who does not should audit the current discussion  
on NANOG.

Regards
Marshall Eubanks

On Mar 6, 2006, at 9:39 PM, Rob Brown wrote:

> Agreed.  IPv4 must be supported.
>
> Regards,
> Robert Brown
> TalkSwitch
>
>
>
>> -----Original Message-----
>> From: p2p-sip-bounces at cs.columbia.edu
>> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Eunsoo Shim
>> Sent: March 6, 2006 9:33 PM
>> To: Michael Slavitch
>> Cc: P2P-SIP
>> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>> I strongly believe support for IPv4 is a must.
>> Thanks.
>>
>> Eunsoo
>>
>> Michael Slavitch wrote:
>>
>>> Are you serious? How many ISP's offer IPV6?  How many organizations
>>> outside academe and a few large enterprises?
>>> Do you realize how many people in the real world still use Windows
>>> 2000 and Windows 98, including some very large organizations?
>>> Given past history Vista will  run on the majority of
>> Windows desktop
>>> systems starting around  2010-2011, and that is optimistic.
>>>
>>> In the meantime, Skype works now.
>>>
>>> If P2PSIP is limited to IPV6 we might as well call it a day
>> right now
>>> and start licensing runtime modules from Skype.
>>>
>>>
>>> On 3/6/06, * Frederick Lefebvre* <fred at perspectiveip.ca
>>> <mailto:fred at perspectiveip.ca>> wrote:
>>>
>>>     What about mandating the use IPv6 and leave the NAT
>> traversal issue to
>>>     the IPv6 transition mechanism used?  Wouldn't it make the
>>>     implementation of p2p-sip easier if the protocol did
>> not have to work
>>>     around NATs?
>>>     Of course, feel free to wonder if this is a realistic
>> proposal or not.
>>>     But with the upcoming Windows Vista, all of a sudden a
>> fairly large
>>>     percentage of users will have V6 connectivity and it
>> might not sound
>>>     so crazy anymore.
>>>
>>>     -- Fred
>>>
>>>     On 3/6/06, Karst Bjorgson <karstbj at yahoo.com
>>>     <mailto:karstbj at yahoo.com>> wrote:
>>>>>> Do you know if there is any intent for Teredo to
>>>>> support symmetric NATs?
>>>>>> I'd suggest this be a base requirement for P2P-SIP.
>>>>
>>>> I am pretty sure that a Teredo implementation could be
>>>> tuned to work through a lot of symmetric NAT, or at
>>>> least in the symmetric to non-symmetric cases. But, as
>>>> Jonathan said, IPv6 should be supported as an option.
>>>> The combination of IPv4 and IPv6 will work more often
>>>> than IPv4 alone.
>>>>
>>>> -- Karst
>>>>
>>>> __________________________________________________
>>>> Do You Yahoo!?
>>>> Tired of spam?  Yahoo! Mail has the best spam
>> protection around
>>>> http://mail.yahoo.com
>>>> _______________________________________________
>>>> p2p-sip mailing list
>>>> p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
>>>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>     <https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip>
>>>>
>>>
>>>     _______________________________________________
>>>     p2p-sip mailing list
>>>     p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
>>>     https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>
>>>
>>>
>>>
>>> --
>>> Michael Slavitch
>>> Ottawa, Ontario Canada
>>>
>>> -------------------------------------------------------------
>> ----------
>>> -
>>>
>>> _______________________________________________
>>> p2p-sip mailing list
>>> p2p-sip at cs.columbia.edu
>>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>
>>>
>>
>> _______________________________________________
>> p2p-sip mailing list
>> p2p-sip at cs.columbia.edu
>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From henry at pulver.com  Mon Mar  6 22:04:55 2006
From: henry at pulver.com (Henry Sinnreich)
Date: Mon, 6 Mar 2006 21:04:55 -0600
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <440CF0F2.8060702@research.panasonic.com>
Message-ID: <200603070308.k27354Bp026005@cs.columbia.edu>

>> Are you serious? How many ISP's offer IPV6?

I agree with Michael. Groupthink and wishful thinking can be serious
impediments for P2P SIP. Skype was original.

 Thanks, Henry

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Eunsoo Shim
Sent: Monday, March 06, 2006 8:33 PM
To: Michael Slavitch
Cc: P2P-SIP
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal

I strongly believe support for IPv4 is a must.
Thanks.

Eunsoo

Michael Slavitch wrote:

> Are you serious? How many ISP's offer IPV6?  How many organizations
> outside academe and a few large enterprises?
> Do you realize how many people in the real world still use Windows
> 2000 and Windows 98, including some very large organizations?
> Given past history Vista will  run on the majority of Windows desktop
> systems starting around  2010-2011, and that is optimistic.
>
> In the meantime, Skype works now.
>
> If P2PSIP is limited to IPV6 we might as well call it a day right now
> and start licensing runtime modules from Skype.
>
>
> On 3/6/06, * Frederick Lefebvre* <fred at perspectiveip.ca
> <mailto:fred at perspectiveip.ca>> wrote:
>
>     What about mandating the use IPv6 and leave the NAT traversal issue to
>     the IPv6 transition mechanism used?  Wouldn't it make the
>     implementation of p2p-sip easier if the protocol did not have to work
>     around NATs?
>     Of course, feel free to wonder if this is a realistic proposal or not.
>     But with the upcoming Windows Vista, all of a sudden a fairly large
>     percentage of users will have V6 connectivity and it might not sound
>     so crazy anymore.
>
>     -- Fred
>
>     On 3/6/06, Karst Bjorgson <karstbj at yahoo.com
>     <mailto:karstbj at yahoo.com>> wrote:
>     > > >Do you know if there is any intent for Teredo to
>     > > support symmetric NATs?
>     > > >I'd suggest this be a base requirement for P2P-SIP.
>     >
>     > I am pretty sure that a Teredo implementation could be
>     > tuned to work through a lot of symmetric NAT, or at
>     > least in the symmetric to non-symmetric cases. But, as
>     > Jonathan said, IPv6 should be supported as an option.
>     > The combination of IPv4 and IPv6 will work more often
>     > than IPv4 alone.
>     >
>     > -- Karst
>     >
>     > __________________________________________________
>     > Do You Yahoo!?
>     > Tired of spam?  Yahoo! Mail has the best spam protection around
>     > http://mail.yahoo.com
>     > _______________________________________________
>     > p2p-sip mailing list
>     > p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
>     > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>     <https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip>
>     >
>
>     _______________________________________________
>     p2p-sip mailing list
>     p2p-sip at cs.columbia.edu <mailto:p2p-sip at cs.columbia.edu>
>     https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>
>
>
> -- 
> Michael Slavitch
> Ottawa, Ontario Canada
>
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip




From enrico.marocco at telecomitalia.it  Tue Mar  7 04:25:17 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Tue, 07 Mar 2006 10:25:17 +0100
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
In-Reply-To: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
References: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
Message-ID: <440D517D.90204@telecomitalia.it>

I'm happy to see that the work is going on! Anyways, after reading the
new version of the draft, I have some doubts:

+ conflicts on Node-ID: as noticed in "Open Issues", conflicts on node
   ids are likely to happen.  Unfortunatly, they can be due not only to
   natted addresses; indeed nothing prevents a well balanced hash
   function to return the same value for two totally different
   "hostports".  Just off the top of my head, I would generate node ids
   as hash("sip:<addr>:<port>;idtag=<rand>") and would define/pick a
   response code for conflict reporting.  In this way, a node would be
   able to generate unique identifiers;

+ redundancy: if I understood well, the proposed mechanism is quite
   different from the chord's one.  Rather than storing copies of each
   record in responsible node's successors, it stores replica in random
   positions.  Thus, the number of messages required for "put" (always)
   and "get" (upon failure) operations doubles or trebles depending on
   the redundancy level.  I think that the (weak) security benefits that
   this solution provides, could also be achieved with a significant
   reduction in the number of messages if the registering node directly
   asks responsible's successors to store replica.

How do you think to deal with them?

Enrico

Bruce Lowekamp wrote:
> We have submitted an updated version of the draft "A P2P Approach to
> SIP Registration and Resource Location", draft-bryan-sipping-p2p-02.
> Until it is available in the archive, you can find it here:
> 
> http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html
> 
> or here
> 
> http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.txt
> 
> depending on the format you like.
> 
> David Bryan
> Bruce Lowekamp
> Cullen Jennings
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip

From lowekamp at cs.wm.edu  Tue Mar  7 09:05:48 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 7 Mar 2006 09:05:48 -0500
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
In-Reply-To: <440D517D.90204@telecomitalia.it>
References: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
	<440D517D.90204@telecomitalia.it>
Message-ID: <e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>

Enrico,

Thanks for your comments.  Both of these points raise the important
issue of how to balance a desire for efficiency with securing the DHT
from attacks.   We have tried to push the protocol toward something
that we think can run on open, public networks that may be subject to
attacks.  Of course, it's still not completely secure, but I think
it's making progress.  Also, it may be possible that some people will
want to set up a DHT that is more efficient rather than secure from
attackers (and, if using HMAC NodeIDs, may simply wish to rely on the
shared secret for security).   In that case, other algorithms might be
more appropriate.

Specific comments inline:

On 3/7/06, Enrico Marocco <enrico.marocco at telecomitalia.it> wrote:
> I'm happy to see that the work is going on! Anyways, after reading the
> new version of the draft, I have some doubts:
>
> + conflicts on Node-ID: as noticed in "Open Issues", conflicts on node
>    ids are likely to happen.  Unfortunatly, they can be due not only to
>    natted addresses; indeed nothing prevents a well balanced hash
>    function to return the same value for two totally different
>    "hostports".  Just off the top of my head, I would generate node ids
>    as hash("sip:<addr>:<port>;idtag=<rand>") and would define/pick a
>    response code for conflict reporting.  In this way, a node would be
>    able to generate unique identifiers;

The problem with allowing a node to generate a unique identifier is
that it opens the DHT up to various attacks, such as the Sybill
attack.  Essentially, if you can generate your own NodeID, you can
insert yourself anywhere (and in as many places) in the ring as you
would like.  Even worse, you can let the ring operate perfectly
normally, but carefully insert yourself to be the primary location for
the ID of a particular user who you wish to attack.  I actually am
leaning toward the need to remove the port number from the hashed
portion altogether, as it gives a naive attacker tens of thousands of
possible IDs to use.  Obviously we would have to deal with collisions
on IDs there.

>
> + redundancy: if I understood well, the proposed mechanism is quite
>    different from the chord's one.  Rather than storing copies of each
>    record in responsible node's successors, it stores replica in random
>    positions.  Thus, the number of messages required for "put" (always)
>    and "get" (upon failure) operations doubles or trebles depending on
>    the redundancy level.  I think that the (weak) security benefits that
>    this solution provides, could also be achieved with a significant
>    reduction in the number of messages if the registering node directly
>    asks responsible's successors to store replica.

This is the most obvious tradeoff of performance vs security.  Again,
consider an attempt to attack the ID of a particular user by acquiring
a node with a NodeID that makes you responsible for storing that
user's registration.  Now you receive a query and respond to it with
your corrupt pointer (assuming a reasonable security system,
presumably you can only attack with DOS rather than impersonation). 
If the successors are storing the replicas, then you can also prevent
the querying node from checking a replica by using false entries in
your finger table.  However, if the replicas are stored at other
locations around the ring, you don't have nearly as much power to
influence those entries or affect how the querying node will locate
them.

You make an excellent point that this scheme has more overhead than
simply using successors.  Our goal is that if we make it difficult
(but not impossible) to choose a NodeID for an attack, and make it
difficult (but not impossible) to misdirect a querying node looking
for a replica, then the combined effect will make it essentially
impossible to use certain kinds of attacks against the network.

Bruce


>
> How do you think to deal with them?
>
> Enrico
>
> Bruce Lowekamp wrote:
> > We have submitted an updated version of the draft "A P2P Approach to
> > SIP Registration and Resource Location", draft-bryan-sipping-p2p-02.
> > Until it is available in the archive, you can find it here:
> >
> > http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html
> >
> > or here
> >
> > http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.txt
> >
> > depending on the format you like.
> >
> > David Bryan
> > Bruce Lowekamp
> > Cullen Jennings
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>


From bryan at ethernot.org  Tue Mar  7 10:16:02 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Tue, 7 Mar 2006 10:16:02 -0500
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
In-Reply-To: <e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>
References: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
	<440D517D.90204@telecomitalia.it>
	<e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>
Message-ID: <8b2769930603070716r395490f0gdc0b8c059cc8840d@mail.gmail.com>

One other option that has been discussed is to use as Node ID the
result of a hash with some number of digits for the port, un-hashed
appended to the has result. This has the result of placing multiple
nodes behind a NAT close to each other in the overlay, which obviously
has implications, but allows users to have multiple nodes behind a
NAT, while preventing the Sybil attack Bruce mentioned. The positive
is that the nodes that are likely to be most affected by attacks are
yoour neighbors, which in this case are (likely) to be your own node
if you are creating multiple identities, but the down side is that you
can thus create long sequences of successors, which also may be a
problem (depends on how your DHT is architected)

I don't think I have the original Sybil paper on p2psip.org, but I'll
put it up in the next week or so for those that haven't read it:

www.cs.rice.edu/Conferences/IPTPS02/101.pdf

--David

On 3/7/06, Bruce Lowekamp <lowekamp at cs.wm.edu> wrote:
> Enrico,
>
> Thanks for your comments.  Both of these points raise the important
> issue of how to balance a desire for efficiency with securing the DHT
> from attacks.   We have tried to push the protocol toward something
> that we think can run on open, public networks that may be subject to
> attacks.  Of course, it's still not completely secure, but I think
> it's making progress.  Also, it may be possible that some people will
> want to set up a DHT that is more efficient rather than secure from
> attackers (and, if using HMAC NodeIDs, may simply wish to rely on the
> shared secret for security).   In that case, other algorithms might be
> more appropriate.
>
> Specific comments inline:
>
> On 3/7/06, Enrico Marocco <enrico.marocco at telecomitalia.it> wrote:
> > I'm happy to see that the work is going on! Anyways, after reading the
> > new version of the draft, I have some doubts:
> >
> > + conflicts on Node-ID: as noticed in "Open Issues", conflicts on node
> >    ids are likely to happen.  Unfortunatly, they can be due not only to
> >    natted addresses; indeed nothing prevents a well balanced hash
> >    function to return the same value for two totally different
> >    "hostports".  Just off the top of my head, I would generate node ids
> >    as hash("sip:<addr>:<port>;idtag=<rand>") and would define/pick a
> >    response code for conflict reporting.  In this way, a node would be
> >    able to generate unique identifiers;
>
> The problem with allowing a node to generate a unique identifier is
> that it opens the DHT up to various attacks, such as the Sybill
> attack.  Essentially, if you can generate your own NodeID, you can
> insert yourself anywhere (and in as many places) in the ring as you
> would like.  Even worse, you can let the ring operate perfectly
> normally, but carefully insert yourself to be the primary location for
> the ID of a particular user who you wish to attack.  I actually am
> leaning toward the need to remove the port number from the hashed
> portion altogether, as it gives a naive attacker tens of thousands of
> possible IDs to use.  Obviously we would have to deal with collisions
> on IDs there.
>
> >
> > + redundancy: if I understood well, the proposed mechanism is quite
> >    different from the chord's one.  Rather than storing copies of each
> >    record in responsible node's successors, it stores replica in random
> >    positions.  Thus, the number of messages required for "put" (always)
> >    and "get" (upon failure) operations doubles or trebles depending on
> >    the redundancy level.  I think that the (weak) security benefits that
> >    this solution provides, could also be achieved with a significant
> >    reduction in the number of messages if the registering node directly
> >    asks responsible's successors to store replica.
>
> This is the most obvious tradeoff of performance vs security.  Again,
> consider an attempt to attack the ID of a particular user by acquiring
> a node with a NodeID that makes you responsible for storing that
> user's registration.  Now you receive a query and respond to it with
> your corrupt pointer (assuming a reasonable security system,
> presumably you can only attack with DOS rather than impersonation).
> If the successors are storing the replicas, then you can also prevent
> the querying node from checking a replica by using false entries in
> your finger table.  However, if the replicas are stored at other
> locations around the ring, you don't have nearly as much power to
> influence those entries or affect how the querying node will locate
> them.
>
> You make an excellent point that this scheme has more overhead than
> simply using successors.  Our goal is that if we make it difficult
> (but not impossible) to choose a NodeID for an attack, and make it
> difficult (but not impossible) to misdirect a querying node looking
> for a replica, then the combined effect will make it essentially
> impossible to use certain kinds of attacks against the network.
>
> Bruce
>
>
> >
> > How do you think to deal with them?
> >
> > Enrico
> >
> > Bruce Lowekamp wrote:
> > > We have submitted an updated version of the draft "A P2P Approach to
> > > SIP Registration and Resource Location", draft-bryan-sipping-p2p-02.
> > > Until it is available in the archive, you can find it here:
> > >
> > > http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html
> > >
> > > or here
> > >
> > > http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.txt
> > >
> > > depending on the format you like.
> > >
> > > David Bryan
> > > Bruce Lowekamp
> > > Cullen Jennings
> > >
> > > _______________________________________________
> > > p2p-sip mailing list
> > > p2p-sip at cs.columbia.edu
> > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From enrico.marocco at telecomitalia.it  Tue Mar  7 11:37:38 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Tue, 07 Mar 2006 17:37:38 +0100
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
In-Reply-To: <e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>
References: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
	<440D517D.90204@telecomitalia.it>
	<e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>
Message-ID: <440DB6D2.7020801@telecomitalia.it>

> The problem with allowing a node to generate a unique identifier is
> that it opens the DHT up to various attacks, such as the Sybill
> attack.

Partially agreed.  This way you are allowing the user to register an
indefinite number of nodes from the same machine; the draft limits that
number to 65534.  Probably, for addressing security threats like the
Sybill attack, other measures are needed.

>          Essentially, if you can generate your own NodeID, you can
> insert yourself anywhere (and in as many places) in the ring as you
> would like.  Even worse, you can let the ring operate perfectly
> normally, but carefully insert yourself to be the primary location for
> the ID of a particular user who you wish to attack.

This is not trivial: it implies breaking of sha-1 (which is known to
require ~2**69 operations). If we think that sha-1 is not secure for our
purpose, the right thing would be switching to sha-2.

>                                                                Again,
> consider an attempt to attack the ID of a particular user by acquiring
> a node with a NodeID that makes you responsible for storing that
> user's registration.  Now you receive a query and respond to it with
> your corrupt pointer (assuming a reasonable security system,
> presumably you can only attack with DOS rather than impersonation). 
> If the successors are storing the replicas, then you can also prevent
> the querying node from checking a replica by using false entries in
> your finger table.  However, if the replicas are stored at other
> locations around the ring, you don't have nearly as much power to
> influence those entries or affect how the querying node will locate
> them.

If you think a node is corrupted (maliciously or not) you can reach both
its predecessor and its successors starting from an arbitrary point of
the network without asking it. Moreover, browsing the finger table and
the successor list provided by the node queried just before the bad one,
you can reach them with very few messages.

The most dangerous case would be when malicious nodes could occupy
contiguous positions; luckily, this could be avoided using a strong hash
function.

Enrico


Bruce Lowekamp wrote:
> Enrico,
> 
> Thanks for your comments.  Both of these points raise the important
> issue of how to balance a desire for efficiency with securing the DHT
> from attacks.   We have tried to push the protocol toward something
> that we think can run on open, public networks that may be subject to
> attacks.  Of course, it's still not completely secure, but I think
> it's making progress.  Also, it may be possible that some people will
> want to set up a DHT that is more efficient rather than secure from
> attackers (and, if using HMAC NodeIDs, may simply wish to rely on the
> shared secret for security).   In that case, other algorithms might be
> more appropriate.
> 
> Specific comments inline:
> 
> On 3/7/06, Enrico Marocco <enrico.marocco at telecomitalia.it> wrote:
> 
>>I'm happy to see that the work is going on! Anyways, after reading the
>>new version of the draft, I have some doubts:
>>
>>+ conflicts on Node-ID: as noticed in "Open Issues", conflicts on node
>>   ids are likely to happen.  Unfortunatly, they can be due not only to
>>   natted addresses; indeed nothing prevents a well balanced hash
>>   function to return the same value for two totally different
>>   "hostports".  Just off the top of my head, I would generate node ids
>>   as hash("sip:<addr>:<port>;idtag=<rand>") and would define/pick a
>>   response code for conflict reporting.  In this way, a node would be
>>   able to generate unique identifiers;
> 
> 
> The problem with allowing a node to generate a unique identifier is
> that it opens the DHT up to various attacks, such as the Sybill
> attack.  Essentially, if you can generate your own NodeID, you can
> insert yourself anywhere (and in as many places) in the ring as you
> would like.  Even worse, you can let the ring operate perfectly
> normally, but carefully insert yourself to be the primary location for
> the ID of a particular user who you wish to attack.  I actually am
> leaning toward the need to remove the port number from the hashed
> portion altogether, as it gives a naive attacker tens of thousands of
> possible IDs to use.  Obviously we would have to deal with collisions
> on IDs there.
> 
> 
>>+ redundancy: if I understood well, the proposed mechanism is quite
>>   different from the chord's one.  Rather than storing copies of each
>>   record in responsible node's successors, it stores replica in random
>>   positions.  Thus, the number of messages required for "put" (always)
>>   and "get" (upon failure) operations doubles or trebles depending on
>>   the redundancy level.  I think that the (weak) security benefits that
>>   this solution provides, could also be achieved with a significant
>>   reduction in the number of messages if the registering node directly
>>   asks responsible's successors to store replica.
> 
> 
> This is the most obvious tradeoff of performance vs security.  Again,
> consider an attempt to attack the ID of a particular user by acquiring
> a node with a NodeID that makes you responsible for storing that
> user's registration.  Now you receive a query and respond to it with
> your corrupt pointer (assuming a reasonable security system,
> presumably you can only attack with DOS rather than impersonation). 
> If the successors are storing the replicas, then you can also prevent
> the querying node from checking a replica by using false entries in
> your finger table.  However, if the replicas are stored at other
> locations around the ring, you don't have nearly as much power to
> influence those entries or affect how the querying node will locate
> them.
> 
> You make an excellent point that this scheme has more overhead than
> simply using successors.  Our goal is that if we make it difficult
> (but not impossible) to choose a NodeID for an attack, and make it
> difficult (but not impossible) to misdirect a querying node looking
> for a replica, then the combined effect will make it essentially
> impossible to use certain kinds of attacks against the network.
> 
> Bruce
> 
> 
> 
>>How do you think to deal with them?
>>
>>Enrico
>>
>>Bruce Lowekamp wrote:
>>
>>>We have submitted an updated version of the draft "A P2P Approach to
>>>SIP Registration and Resource Location", draft-bryan-sipping-p2p-02.
>>>Until it is available in the archive, you can find it here:
>>>
>>>http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html
>>>
>>>or here
>>>
>>>http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.txt
>>>
>>>depending on the format you like.
>>>
>>>David Bryan
>>>Bruce Lowekamp
>>>Cullen Jennings
>>>
>>>_______________________________________________
>>>p2p-sip mailing list
>>>p2p-sip at cs.columbia.edu
>>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
>>

From enrico.marocco at telecomitalia.it  Tue Mar  7 11:54:16 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Tue, 07 Mar 2006 17:54:16 +0100
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
In-Reply-To: <8b2769930603070716r395490f0gdc0b8c059cc8840d@mail.gmail.com>
References: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
	<440D517D.90204@telecomitalia.it>
	<e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>
	<8b2769930603070716r395490f0gdc0b8c059cc8840d@mail.gmail.com>
Message-ID: <440DBAB8.7080800@telecomitalia.it>

David A. Bryan wrote:
> One other option that has been discussed is to use as Node ID the
> result of a hash with some number of digits for the port, un-hashed
> appended to the has result. This has the result of placing multiple
> nodes behind a NAT close to each other in the overlay, which obviously
> has implications, but allows users to have multiple nodes behind a
> NAT, while preventing the Sybil attack Bruce mentioned. The positive
> is that the nodes that are likely to be most affected by attacks are
> yoour neighbors, which in this case are (likely) to be your own node
> if you are creating multiple identities, but the down side is that you
> can thus create long sequences of successors, which also may be a
> problem (depends on how your DHT is architected)

This doesn't prevents from conflicts on node ids.

Just off the top of my head, if the problem is avoiding too much uses of 
the same address, why not just storing the information in the dht 
itself? Up on a node join, the accepting node could insert/update a pair 
like <"ip address", counter> and could allow the registration only if 
counter is below a given threshold.

Enrico

From dean.willis at softarmor.com  Tue Mar  7 12:42:14 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 7 Mar 2006 11:42:14 -0600
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
References: <4409B28B.2000004@research.panasonic.com>
	<20060306063651.52150.qmail@web53006.mail.yahoo.com>
	<3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>
	<1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
Message-ID: <2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>


On Mar 6, 2006, at 6:52 PM, Michael Slavitch wrote:

> Are you serious? How many ISP's offer IPV6?  How many organizations  
> outside academe and a few large enterprises?
> Do you realize how many people in the real world still use Windows  
> 2000 and Windows 98, including some very large organizations?
> Given past history Vista will  run on the majority of Windows  
> desktop systems starting around  2010-2011, and that is optimistic.
>
> In the meantime, Skype works now.
>
> If P2PSIP is limited to IPV6 we might as well call it a day right  
> now and start licensing runtime modules from Skype.
>
>
> On 3/6/06, Frederick Lefebvre <fred at perspectiveip.ca> wrote: What  
> about mandating the use IPv6 and leave the NAT traversal issue to
> the IPv6 transition mechanism used?  Wouldn't it make the
> implementation of p2p-sip easier if the protocol did not have to work
> around NATs?
> Of course, feel free to wonder if this is a realistic proposal or not.
> But with the upcoming Windows Vista, all of a sudden a fairly large
> percentage of users will have V6 connectivity and it might not sound
> so crazy anymore.



I believe the suggestion was that we not develop a P2P-SIP specific  
mechanism for NAT traversal on IPV6 networks, but leave that to the  
IPV6 transition mechanism.

This is far different from suggesting that we not support IPV4.

--
Dean

>

From slavitch at gmail.com  Tue Mar  7 12:49:46 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Tue, 7 Mar 2006 12:49:46 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>
References: <4409B28B.2000004@research.panasonic.com>
	<20060306063651.52150.qmail@web53006.mail.yahoo.com>
	<3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>
	<1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
	<2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>
Message-ID: <1fc32c8f0603070949x1edf2a4ocafa0c7bfbf85fc9@mail.gmail.com>

"What about mandating the use (of) IPv6" and "Wouldn't it make the
implementation of p2p-sip easier if the protocol did not have to work around
NATs?"  sound very clear to me.


On 3/7/06, Dean Willis <dean.willis at softarmor.com> wrote:
>
>
> On Mar 6, 2006, at 6:52 PM, Michael Slavitch wrote:
>
> > Are you serious? How many ISP's offer IPV6?  How many organizations
> > outside academe and a few large enterprises?
> > Do you realize how many people in the real world still use Windows
> > 2000 and Windows 98, including some very large organizations?
> > Given past history Vista will  run on the majority of Windows
> > desktop systems starting around  2010-2011, and that is optimistic.
> >
> > In the meantime, Skype works now.
> >
> > If P2PSIP is limited to IPV6 we might as well call it a day right
> > now and start licensing runtime modules from Skype.
> >
> >
> > On 3/6/06, Frederick Lefebvre <fred at perspectiveip.ca> wrote: What
> > about mandating the use IPv6 and leave the NAT traversal issue to
> > the IPv6 transition mechanism used?  Wouldn't it make the
> > implementation of p2p-sip easier if the protocol did not have to work
> > around NATs?
> > Of course, feel free to wonder if this is a realistic proposal or not.
> > But with the upcoming Windows Vista, all of a sudden a fairly large
> > percentage of users will have V6 connectivity and it might not sound
> > so crazy anymore.
>
>
>
> I believe the suggestion was that we not develop a P2P-SIP specific
> mechanism for NAT traversal on IPV6 networks, but leave that to the
> IPV6 transition mechanism.
>
> This is far different from suggesting that we not support IPV4.
>
> --
> Dean
>
> >
>



--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060307/b1a23de2/attachment.html

From marc1.bailly at francetelecom.com  Tue Mar  7 13:02:26 2006
From: marc1.bailly at francetelecom.com (BAILLY Marc RD-CORE-LAN)
Date: Tue, 7 Mar 2006 19:02:26 +0100
Subject: [p2p-sip] I-D: An Architecture for P2P SIP
Message-ID: <BAF83494CE653943A97B9F755016A06602998D3A@ftrdmel1.rd.francetelecom.fr>


Hi Eunsoo,

In section 8, regarding the storage of ressource records for SIP, you include a timestamp and a lifetime. I agree that these information are needed. 
It is not mentioned in the draft on what is based this timestamp. If this timestamp is based on the node clock, it requires that this clock is correct : if possible shifts in nodes clocks are not taken into account, I think we will have bad behaviors in the management of records in DHT.
Maybe it could be interesting to mention this point in your draft.

Marc

> -----Message d'origine-----
> De : p2p-sip-bounces at cs.columbia.edu 
> [mailto:p2p-sip-bounces at cs.columbia.edu] De la part de Eunsoo Shim
> Envoy? : mardi 28 f?vrier 2006 00:22
> ? : P2P-SIP
> Objet : [p2p-sip] I-D: An Architecture for P2P SIP
> 
> We submitted an I-D that describes an architecture for P2P SIP.
> You can get it from the link below until it appears in the IETF I-D 
> directory.
> We hope this draft can help the architecture and charter 
> discussion in BOF.
> Your comments will be appreciated.
> Thanks.
> 
> Eunsoo
> 
> http://www.comet.columbia.edu/~eunsoo/ietf/draft-shim-sipping-
> p2p-arch-00.txt
> 
> Abstract
> 
>    This document describes an architecture for peer-to-peer 
> SIP systems
>    in which a separate and independent peer-to-peer overlay layer
>    provides a distributed resource placement and search 
> service for SIP.
>    It also aims to help identify what should be specified for
>    interoperation.
> 
> 
> 
> 
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From eunsoo at research.panasonic.com  Tue Mar  7 14:41:15 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Tue, 07 Mar 2006 14:41:15 -0500
Subject: [p2p-sip] I-D: An Architecture for P2P SIP
In-Reply-To: <BAF83494CE653943A97B9F755016A06602998D3A@ftrdmel1.rd.francetel
	ecom.fr>
References: <BAF83494CE653943A97B9F755016A06602998D3A@ftrdmel1.rd.francetele
	com.fr>
Message-ID: <440DE1DB.7020207@research.panasonic.com>

I think the draft does not provide enough details about the use of the 
user location record.
Actually it was not out intention to propose protocol details by this 
draft. We intended to propose a possible architecture, describe how the 
system will work with the architecture, and show what specifications we 
need. Once the group has a consensus on those, detailed specifications 
can be worked out more easily.

You pointed out one of many that were not elaborated in enough details 
in the draft.

We don't assume clock synchronization among peers.
The timestamp is just for identifying which record is more fresh (or 
up-to-date) when multiple records, conflicting each other, are discovered.
It can be considered as a version number. So the timestampe could be the 
system time of the peer that generated the record or a number 
incremented at every update of the record.

Thanks.

Eunsoo


BAILLY Marc RD-CORE-LAN wrote:

>Hi Eunsoo,
>
>In section 8, regarding the storage of ressource records for SIP, you include a timestamp and a lifetime. I agree that these information are needed. 
>It is not mentioned in the draft on what is based this timestamp. If this timestamp is based on the node clock, it requires that this clock is correct : if possible shifts in nodes clocks are not taken into account, I think we will have bad behaviors in the management of records in DHT.
>Maybe it could be interesting to mention this point in your draft.
>
>Marc
>
>  
>
>>-----Message d'origine-----
>>De : p2p-sip-bounces at cs.columbia.edu 
>>[mailto:p2p-sip-bounces at cs.columbia.edu] De la part de Eunsoo Shim
>>Envoy? : mardi 28 f?vrier 2006 00:22
>>? : P2P-SIP
>>Objet : [p2p-sip] I-D: An Architecture for P2P SIP
>>
>>We submitted an I-D that describes an architecture for P2P SIP.
>>You can get it from the link below until it appears in the IETF I-D 
>>directory.
>>We hope this draft can help the architecture and charter 
>>discussion in BOF.
>>Your comments will be appreciated.
>>Thanks.
>>
>>Eunsoo
>>
>>http://www.comet.columbia.edu/~eunsoo/ietf/draft-shim-sipping-
>>p2p-arch-00.txt
>>
>>Abstract
>>
>>   This document describes an architecture for peer-to-peer 
>>SIP systems
>>   in which a separate and independent peer-to-peer overlay layer
>>   provides a distributed resource placement and search 
>>service for SIP.
>>   It also aims to help identify what should be specified for
>>   interoperation.
>>
>>
>>
>>
>>
>>_______________________________________________
>>p2p-sip mailing list
>>p2p-sip at cs.columbia.edu
>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
>>    
>>


From csp at csperkins.org  Tue Mar  7 15:00:42 2006
From: csp at csperkins.org (Colin Perkins)
Date: Tue, 7 Mar 2006 20:00:42 +0000
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>
References: <4409B28B.2000004@research.panasonic.com>
	<20060306063651.52150.qmail@web53006.mail.yahoo.com>
	<3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>
	<1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
	<2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>
Message-ID: <6D84DFFE-12B3-40C5-B1F3-2C6BB3A32792@csperkins.org>

On 7 Mar 2006, at 17:42, Dean Willis wrote:
> I believe the suggestion was that we not develop a P2P-SIP specific
> mechanism for NAT traversal on IPV6 networks, but leave that to the
> IPV6 transition mechanism.
>
> This is far different from suggesting that we not support IPV4.

Indeed -- there are interesting approaches that build a peer-to-peer  
IP overlay network, perhaps IPv6-based, as the NAT traversal  
mechanism, and then run normal applications on that overlay network.

Colin

From salman at cs.columbia.edu  Tue Mar  7 18:25:33 2006
From: salman at cs.columbia.edu (Salman Abdul Baset)
Date: Tue, 7 Mar 2006 18:25:33 -0500 (EST)
Subject: [p2p-sip] 50% of broadband US/UK MSN users behind NAT
Message-ID: <Pine.GSO.4.58.0603071823480.27087@flame.cs.columbia.edu>

http://research.microsoft.com/~pablo/papers/CCR.pdf (CCR Jan 2006)
(pointed in a post on p2p-hackers list)

-salman

From lowekamp at cs.wm.edu  Wed Mar  8 02:00:03 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Wed, 8 Mar 2006 02:00:03 -0500
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
In-Reply-To: <440DB6D2.7020801@telecomitalia.it>
References: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
	<440D517D.90204@telecomitalia.it>
	<e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>
	<440DB6D2.7020801@telecomitalia.it>
Message-ID: <e9132f820603072300u2ab67453qb5f585413eda3b8d@mail.gmail.com>

On 3/7/06, Enrico Marocco <enrico.marocco at telecomitalia.it> wrote:
> >          Essentially, if you can generate your own NodeID, you can
> > insert yourself anywhere (and in as many places) in the ring as you
> > would like.  Even worse, you can let the ring operate perfectly
> > normally, but carefully insert yourself to be the primary location for
> > the ID of a particular user who you wish to attack.
>
> This is not trivial: it implies breaking of sha-1 (which is known to
> require ~2**69 operations). If we think that sha-1 is not secure for our
> purpose, the right thing would be switching to sha-2.

Well, no, you don't need to break sha-1.  All you need to do is
calculate an address that hashes to an ID in between the ID of the
user you wish to DOS and the ID of the current node.  Assuming uniform
distribution of hashes (on average, true with sha1), your chances of
being able to do this scale with the number of nodes in the network. 
For small networks (a few hundred), you can trivially insert yourself
into almost any point in the network by using different port numbers. 
Obviously as the network grows, this type of attack (against a
predetermined user) becomes harder.

>
> >                                                                Again,
> > consider an attempt to attack the ID of a particular user by acquiring
> > a node with a NodeID that makes you responsible for storing that
> > user's registration.  Now you receive a query and respond to it with
> > your corrupt pointer (assuming a reasonable security system,
> > presumably you can only attack with DOS rather than impersonation).
> > If the successors are storing the replicas, then you can also prevent
> > the querying node from checking a replica by using false entries in
> > your finger table.  However, if the replicas are stored at other
> > locations around the ring, you don't have nearly as much power to
> > influence those entries or affect how the querying node will locate
> > them.
>
> If you think a node is corrupted (maliciously or not) you can reach both
> its predecessor and its successors starting from an arbitrary point of
> the network without asking it. Moreover, browsing the finger table and
> the successor list provided by the node queried just before the bad one,
> you can reach them with very few messages.

True, but is this solution more efficient than simply storing a
replica at a random location?  At a random location, you still benefit
from log(n) search, whereas if you start searching linearly.  I'm
really not sure of the answer here.

Also, how do you know which node is bad?  An advantage of the random
replica is that you are unlikely to need to worry about correlation
between your two queries.  If you use successive replicas, you now
need to worry about which node is corrupt, which makes the problem
much harder.


> The most dangerous case would be when malicious nodes could occupy
> contiguous positions; luckily, this could be avoided using a strong hash
> function.

As above, this depends on the number of nodes in the network (if your
port number is included in the hash).

Bruce


From enrico.marocco at telecomitalia.it  Wed Mar  8 04:11:56 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Wed, 08 Mar 2006 10:11:56 +0100
Subject: [p2p-sip] update to draft-bryan-sipping-p2p
In-Reply-To: <e9132f820603072300u2ab67453qb5f585413eda3b8d@mail.gmail.com>
References: <e9132f820603051948m2677fc1br1d8fc9b9a9038da8@mail.gmail.com>
	<440D517D.90204@telecomitalia.it>
	<e9132f820603070605k19fb43b8n24b5cb670c1af179@mail.gmail.com>
	<440DB6D2.7020801@telecomitalia.it>
	<e9132f820603072300u2ab67453qb5f585413eda3b8d@mail.gmail.com>
Message-ID: <440E9FDC.4010500@telecomitalia.it>

Bruce Lowekamp wrote:
> On 3/7/06, Enrico Marocco <enrico.marocco at telecomitalia.it> wrote:
>>>         Essentially, if you can generate your own NodeID, you can
>>>insert yourself anywhere (and in as many places) in the ring as you
>>>would like.  Even worse, you can let the ring operate perfectly
>>>normally, but carefully insert yourself to be the primary location for
>>>the ID of a particular user who you wish to attack.
>>
>>This is not trivial: it implies breaking of sha-1 (which is known to
>>require ~2**69 operations). If we think that sha-1 is not secure for our
>>purpose, the right thing would be switching to sha-2.
> 
> Well, no, you don't need to break sha-1.  All you need to do is
> calculate an address that hashes to an ID in between the ID of the
> user you wish to DOS and the ID of the current node.  Assuming uniform
> distribution of hashes (on average, true with sha1), your chances of
> being able to do this scale with the number of nodes in the network. 
> For small networks (a few hundred), you can trivially insert yourself
> into almost any point in the network by using different port numbers. 
> Obviously as the network grows, this type of attack (against a
> predetermined user) becomes harder.

Right. But a mechanism for aviding id spoofing (or whatever we call it)
could still be easily deployed:
+ define the node URI as: "sip:<h>@<addr>[:<port>];idtag=<cnt>", where:
   - h = hash("sip:<addr>;idtag=<cnt>") is the hash value;
+ when a new register is received for a node URL with idtag > 1, perform
   a lookup of the same URL but with a decremented idtag. If it fails,
   kick out the joining node.

>>>                                                               Again,
>>>consider an attempt to attack the ID of a particular user by acquiring
>>>a node with a NodeID that makes you responsible for storing that
>>>user's registration.  Now you receive a query and respond to it with
>>>your corrupt pointer (assuming a reasonable security system,
>>>presumably you can only attack with DOS rather than impersonation).
>>>If the successors are storing the replicas, then you can also prevent
>>>the querying node from checking a replica by using false entries in
>>>your finger table.  However, if the replicas are stored at other
>>>locations around the ring, you don't have nearly as much power to
>>>influence those entries or affect how the querying node will locate
>>>them.
>>
>>If you think a node is corrupted (maliciously or not) you can reach both
>>its predecessor and its successors starting from an arbitrary point of
>>the network without asking it. Moreover, browsing the finger table and
>>the successor list provided by the node queried just before the bad one,
>>you can reach them with very few messages.
> 
> 
> True, but is this solution more efficient than simply storing a
> replica at a random location?  At a random location, you still benefit
> from log(n) search, whereas if you start searching linearly.  I'm
> really not sure of the answer here.

But this solution introduces overhead only when a inconsistent state is
detected. Obviously, as you noticed, it relyes on the difficulty of
spoofing ids and, in particular, on the difficulty of placing chains of
malicious nodes.

> Also, how do you know which node is bad?  An advantage of the random
> replica is that you are unlikely to need to worry about correlation
> between your two queries.  If you use successive replicas, you now
> need to worry about which node is corrupt, which makes the problem
> much harder.

In each scenarios, when you are looking for successive replicas, you are 
supposing the first query was performed on a bad node; I don't think 
anyone wants to seek replicas whenever a call cannot be established.

>>The most dangerous case would be when malicious nodes could occupy
>>contiguous positions; luckily, this could be avoided using a strong hash
>>function.
> 
> As above, this depends on the number of nodes in the network (if your
> port number is included in the hash).

You're right.
Again, I think that we can find a good and still simple solution for 
resolving conflicts, avoiding spoofing and avoiding multiple identities 
relying on the dht itself.

Enrico

From hu.xinyu at huawei.com  Wed Mar  8 04:47:46 2006
From: hu.xinyu at huawei.com (Seaward Hu)
Date: Wed, 08 Mar 2006 17:47:46 +0800
Subject: [p2p-sip] I-D: An Architecture for P2P SIP
In-Reply-To: <440389B2.5040801@research.panasonic.com>
Message-ID: <000001c64295$5dd442f0$3837a40a@china.huawei.com>

Hello Mr.Shim,
        Can the DHT support content search? I mean the key word which
putted by ON may be not as accurate as those stored in the DHT network.

 
 
 
 
 
 
Sincerely yours,
Seaward

 
************************************************************************
***************
This e-mail and its attachments contain confidential information from
HUAWEI, which is intended only for the person or entity whose address is
listed above. Any use of the information contained herein in any way
(including, but not limited to, total or partial disclosure,
reproduction, or dissemination) by persons other than the intended
recipient(s) is prohibited. If you receive this e-mail in error, please
notify the sender by phone or email immediately and delete it!
************************************************************************
*****************


> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu 
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Eunsoo Shim
> Sent: Tuesday, February 28, 2006 7:22 AM
> To: P2P-SIP
> Subject: [p2p-sip] I-D: An Architecture for P2P SIP
> 
> 
> We submitted an I-D that describes an architecture for P2P SIP.
> You can get it from the link below until it appears in the IETF I-D 
> directory.
> We hope this draft can help the architecture and charter 
> discussion in BOF.
> Your comments will be appreciated.
> Thanks.
> 
> Eunsoo
> 
http://www.comet.columbia.edu/~eunsoo/ietf/draft-shim-sipping-p2p-arch-0
0.txt

Abstract

   This document describes an architecture for peer-to-peer SIP systems
   in which a separate and independent peer-to-peer overlay layer
   provides a distributed resource placement and search service for SIP.
   It also aims to help identify what should be specified for
   interoperation.





_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip



From stiemerling at netlab.nec.de  Wed Mar  8 05:46:46 2006
From: stiemerling at netlab.nec.de (Martin Stiemerling)
Date: Wed, 8 Mar 2006 11:46:46 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <44070BE1.3030205@cisco.com>
References: <156450F9DBF9883343076F10@[192.168.1.128]>
	<44070BE1.3030205@cisco.com>
Message-ID: <6E8ADCC4-1FE5-47AA-90CD-F1592D777F0D@netlab.nec.de>

Hi Jonathan, all,


Am 02.03.2006 um 16:14 schrieb Jonathan Rosenberg:

> So, I read this with much interest.
>
> I disagree with many of the assumptions and the analysis performed by
> this document. First and foremost, I think it is a dire, dire, dire
> mistake should any kind of nat traversal solution be proposed for p2p
> sip which is not directly interoperable with the "regular SIP" nat
> traversal strategies. This draft seems to hint that something totally
> new is needed, though I wasnt sure.

Nope, the draft is NOT suggesting to use something totally new. Indeed
it tries to identify the possible ways to go in p2p sip. And definitely,
all techniques of NAT traversal mentioned in the document should
work just fine with the "regular SIP". There is no intention to open
the box for yet another NAT traversal technique.

>
> Secondly, the draft says its an analysis of existing nat traversal
> techniques for sip. However, it omits the main SIP nat traversal
> techinque in its entirety - ICE and sip-outbound. I dont know how an
> analysis of the applicability of existing nat traversal techniques can
> be considered complete when it has missed the main standards in  
> this area.

Granted. We have missed ICE, even though it is well-known. This is not
because of ignorance but due to a very fast approaching -00 deadline.
As said above, the document's intention is not to open the box for a
new protocols, but to explorer the existing NAT traversal space.

>
> Indeed, ICE is basically p2p NAT traversal, and so I believe ice fits
> into a p2p framework extremely well. The only trick you'll need to
> figure out is discovering turn and stun servers, since those  
> presumably
> will be functions provided by other peers in the network (ICE  
> allows for
> any means of discovery of relay services and also allows usage of

I share this view.

> non-turn ones too). If p2p sip endpoints do not use ICE for nat
> traversal, then we will not have interopreability for nat traversal
> between regular sip and p2p-sip, and that is not acceptable IMHO.
>
> The draft presumes that any nat traversal solution requires a step  
> where
> the presence of nat is detected, and typed (section 2.1.2). I strongly
> disagree with this. The ice solution does not attempt to discover or
> otherwise characterize nats at all. Indeed, one of our main  
> experiences
> with rfc3489 was that the whole notion of nat discovery and type
> computation was flawed. That is why this functionality was removed  
> from
> rfc3489bis almost two years ago.

Sure, it is quite difficult to characterize NATs in general, but  
nodes should
be free to run their NAT classification procedure. Where  
classification does
not refer to "this is kind of NAT type X" but more "does it work what  
I want
to achieve".

>
> Section 4.2.1 asserts that middlebox detection is needed for non-UDP
> flows. I half agree with this. I agree that we need to make tcp work,
> and support configurations where there is no intermediary, even if its
> another user in the p2p network. The draft asserts that there are no
> solutions for that, but its just not true. ICE-tcp
> (draft-ietf-mmusic-ice-tcp) will allow for creation of direct tcp

Same here as with the ICE above.

Thanks,

   Martin

>
> connections through nat as long as the nats allow it; if they don't
> it'll end up falling back on a connection through a relay. Note that
> this relay can be dicsovered by any means, and thus can be another  
> user
> on the p2p network.
>
> Thanks,
> Jonathan R.
>
> Juergen Quittek wrote:
>
>> Dear all,
>>
>> We tried to analyze the NAT and firewall problem in order to identify
>> issues that would need consideration when standardizing P2P SIP.
>>
>> Basically we identified four issues, three for SIP signaling and  
>> one for
>> middlebox detection.  Please find them at
>>
>> <http://www.ietf.org/internet-drafts/draft-quittek-p2p-sip- 
>> middlebox-00.txt>
>>
>> Here is the abstract:
>>
>>    Middleboxes, particularly firewalls and network address  
>> translators,
>>    are essential components of today's Internet infrastructure.  They
>>    are designed to support client-server applications well, but very
>>    often they are obstacles for peer-to-peer communication.  This  
>> memo
>>    discusses middlebox traversal issues of SIP-based peer-to-peer
>>    communication.  Required communication steps are analyzed  
>> concerning
>>    their potential constraints caused by middleboxes.  The  
>> requirements
>>    derived from this analysis are compared with the capabilities of
>>    currently available middlebox traversal methods and SIP signaling
>>    standards.  The comparison identifies open issues that need to be
>>    considered when designing standards for a SIP-based peer-to-peer
>>    communication infrastructure.
>>
>> The memo contains the essential facts, but we think it needs a more
>> detailed elaboration in the next version.
>>
>> Thanks,
>>
>>     Juergen
>
> -- 
> Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
> Cisco Fellow                                   Parsippany, NJ  
> 07054-2711
> Cisco Systems
> jdrosen at cisco.com                              FAX:   (973) 952-5050
> http://www.jdrosen.net                         PHONE: (973) 952-5000
> http://www.cisco.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From stiemerling at netlab.nec.de  Wed Mar  8 05:51:33 2006
From: stiemerling at netlab.nec.de (Martin Stiemerling)
Date: Wed, 8 Mar 2006 11:51:33 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <20060304073042.93284.qmail@web53015.mail.yahoo.com>
References: <20060304073042.93284.qmail@web53015.mail.yahoo.com>
Message-ID: <CE77544B-6904-4414-A2BB-C510C59A7480@netlab.nec.de>


Am 04.03.2006 um 08:30 schrieb Karst Bjorgson:

>
>
> --- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>
>> Secondly, the draft says its an analysis of existing
>> nat traversal
>> techniques for sip. However, it omits the main SIP
>> nat traversal
>> techinque in its entirety - ICE and sip-outbound. I
>> dont know how an
>> analysis of the applicability of existing nat
>> traversal techniques can
>> be considered complete when it has missed the main
>> standards in this area.
>
> The draft also fails to mention IPv6. I believe this

Well, is there a NAT for IPv6? Or do you refer to NAT
traversal of IPv6 itself?

If it is NAT traversal of IPv6 itself, your SIP application would
probably not see much of it. In this case the operating system
would get you an IPv6 address, which is basically located at
the Teredo server.

   Martin

> is a bit short-sighted. Microsoft is making a big push
> for IPv6 in Windows Vista. Vista will include an
> implementation of Teredo (RFC 4380), which provides
> some pretty interesting NAT traversal facilities.
>
> -- Karst
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From quittek at netlab.nec.de  Wed Mar  8 06:27:25 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Wed, 08 Mar 2006 12:27:25 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <6E8ADCC4-1FE5-47AA-90CD-F1592D777F0D@netlab.nec.de>
References: <156450F9DBF9883343076F10@[192.168.1.128]>	<44070BE1.3030205@cisco.com>
	<6E8ADCC4-1FE5-47AA-90CD-F1592D777F0D@netlab.nec.de>
Message-ID: <FE2F347265F6D37957DF496F@[10.1.1.171]>

Jonathan and Martin,

--On 3/8/06 11:46 AM +0100 Martin Stiemerling wrote:

> Hi Jonathan, all,
>
>
> Am 02.03.2006 um 16:14 schrieb Jonathan Rosenberg:
>
>> So, I read this with much interest.
>>
>> I disagree with many of the assumptions and the analysis performed by
>> this document. First and foremost, I think it is a dire, dire, dire
>> mistake should any kind of nat traversal solution be proposed for p2p
>> sip which is not directly interoperable with the "regular SIP" nat
>> traversal strategies. This draft seems to hint that something totally
>> new is needed, though I wasnt sure.
>
> Nope, the draft is NOT suggesting to use something totally new. Indeed
> it tries to identify the possible ways to go in p2p sip. And definitely,
> all techniques of NAT traversal mentioned in the document should
> work just fine with the "regular SIP". There is no intention to open
> the box for yet another NAT traversal technique.

The point of confusion might be that the draft tries to decompose functions
that are nicely integrated into existing NAT traversal techniques.
These functions are listed separately, but there is not intention
to propose that the selected NAT traversal methods for P2P SIP should
be individual solutions for each function. The drafts says (probably
not strong enough) that this is not required (and probably not
recommendable).

>> Secondly, the draft says its an analysis of existing nat traversal
>> techniques for sip. However, it omits the main SIP nat traversal
>> techinque in its entirety - ICE and sip-outbound. I dont know how an
>> analysis of the applicability of existing nat traversal techniques can
>> be considered complete when it has missed the main standards in
>> this area.
>
> Granted. We have missed ICE, even though it is well-known. This is not
> because of ignorance but due to a very fast approaching -00 deadline.
> As said above, the document's intention is not to open the box for a
> new protocols, but to explorer the existing NAT traversal space.

Yes, the lack of ICE coverage is really a flaw.  But it should be no
problem to fix it for the next version.  Also, I do not see a big impact
on the identified issues that we should take care of when developing
P2P SIP.

>> Indeed, ICE is basically p2p NAT traversal, and so I believe ice fits
>> into a p2p framework extremely well. The only trick you'll need to
>> figure out is discovering turn and stun servers, since those
>> presumably
>> will be functions provided by other peers in the network (ICE
>> allows for
>> any means of discovery of relay services and also allows usage of
>
> I share this view.
>
>> non-turn ones too). If p2p sip endpoints do not use ICE for nat
>> traversal, then we will not have interopreability for nat traversal
>> between regular sip and p2p-sip, and that is not acceptable IMHO.
>>
>> The draft presumes that any nat traversal solution requires a step
>> where
>> the presence of nat is detected, and typed (section 2.1.2). I strongly
>> disagree with this. The ice solution does not attempt to discover or
>> otherwise characterize nats at all. Indeed, one of our main
>> experiences
>> with rfc3489 was that the whole notion of nat discovery and type
>> computation was flawed. That is why this functionality was removed
>> from
>> rfc3489bis almost two years ago.
>
> Sure, it is quite difficult to characterize NATs in general, but
> nodes should
> be free to run their NAT classification procedure. Where
> classification does
> not refer to "this is kind of NAT type X" but more "does it work what
> I want
> to achieve".

Yes, the draft should state this more clearly.
Since we do not have a common NAT taxonomy, it would be very difficult
to clearly identify a NAT type.  But it will be comparably easy to identify
which kind of communication is possible and how it can be achieved.

>> Section 4.2.1 asserts that middlebox detection is needed for non-UDP
>> flows. I half agree with this. I agree that we need to make tcp work,
>> and support configurations where there is no intermediary, even if its
>> another user in the p2p network. The draft asserts that there are no
>> solutions for that, but its just not true. ICE-tcp
>> (draft-ietf-mmusic-ice-tcp) will allow for creation of direct tcp

Please note that draft-ietf-mmusic-ice-tcp-00 has the same publication
date as the P2P SIP NAT traversal draft. So, it is no wonder that ICE TCP
is not yet included.

The draft does not state or assert that there is no solution for TCP
NAT traversal.  But it states that there are no known tools for
performing middlebox detection for protocols other than UDP. And
after reading draft-ietf-mmusic-ice-tcp-00 I still think this is correct.

Thanks,

    Juergen

> Same here as with the ICE above.
>
> Thanks,
>
>    Martin
>
>>
>> connections through nat as long as the nats allow it; if they don't
>> it'll end up falling back on a connection through a relay. Note that
>> this relay can be dicsovered by any means, and thus can be another
>> user
>> on the p2p network.
>>
>> Thanks,
>> Jonathan R.
>>
>> Juergen Quittek wrote:
>>
>>> Dear all,
>>>
>>> We tried to analyze the NAT and firewall problem in order to identify
>>> issues that would need consideration when standardizing P2P SIP.
>>>
>>> Basically we identified four issues, three for SIP signaling and
>>> one for
>>> middlebox detection.  Please find them at
>>>
>>> <http://www.ietf.org/internet-drafts/draft-quittek-p2p-sip-
>>> middlebox-00.txt>
>>>
>>> Here is the abstract:
>>>
>>>    Middleboxes, particularly firewalls and network address
>>> translators,
>>>    are essential components of today's Internet infrastructure.  They
>>>    are designed to support client-server applications well, but very
>>>    often they are obstacles for peer-to-peer communication.  This
>>> memo
>>>    discusses middlebox traversal issues of SIP-based peer-to-peer
>>>    communication.  Required communication steps are analyzed
>>> concerning
>>>    their potential constraints caused by middleboxes.  The
>>> requirements
>>>    derived from this analysis are compared with the capabilities of
>>>    currently available middlebox traversal methods and SIP signaling
>>>    standards.  The comparison identifies open issues that need to be
>>>    considered when designing standards for a SIP-based peer-to-peer
>>>    communication infrastructure.
>>>
>>> The memo contains the essential facts, but we think it needs a more
>>> detailed elaboration in the next version.
>>>
>>> Thanks,
>>>
>>>     Juergen
>>
>> --
>> Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
>> Cisco Fellow                                   Parsippany, NJ
>> 07054-2711
>> Cisco Systems
>> jdrosen at cisco.com                              FAX:   (973) 952-5050
>> http://www.jdrosen.net                         PHONE: (973) 952-5000
>> http://www.cisco.com
>> _______________________________________________
>> p2p-sip mailing list
>> p2p-sip at cs.columbia.edu
>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip



From quittek at netlab.nec.de  Wed Mar  8 06:32:59 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Wed, 08 Mar 2006 12:32:59 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <20060304073042.93284.qmail@web53015.mail.yahoo.com>
References: <20060304073042.93284.qmail@web53015.mail.yahoo.com>
Message-ID: <0CBEFB65075A3619961CE12B@[10.1.1.171]>

Karst,

The draft does not make any specific assumption about IPv4 or IPv6.
Particularly, it does not make the (potentially wrong) assumption
that if the entire Internet had moved to IPv6, there would be no
NATs anymore.

Thanks,

    Juergen

--On 3/4/06 8:30 AM +0100 Karst Bjorgson wrote:

>
>
>
> --- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>
>> Secondly, the draft says its an analysis of existing
>> nat traversal
>> techniques for sip. However, it omits the main SIP
>> nat traversal
>> techinque in its entirety - ICE and sip-outbound. I
>> dont know how an
>> analysis of the applicability of existing nat
>> traversal techniques can
>> be considered complete when it has missed the main
>> standards in this area.
>
> The draft also fails to mention IPv6. I believe this
> is a bit short-sighted. Microsoft is making a big push
> for IPv6 in Windows Vista. Vista will include an
> implementation of Teredo (RFC 4380), which provides
> some pretty interesting NAT traversal facilities.
>
> -- Karst
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip



From quittek at netlab.nec.de  Wed Mar  8 06:34:42 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Wed, 08 Mar 2006 12:34:42 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>
References: <4409B28B.2000004@research.panasonic.com>	<20060306063651.52150.qmail@web53006.mail.yahoo.com>	<3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>	<1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
	<2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>
Message-ID: <B2E4533F10A0C5BB0D809B73@[10.1.1.171]>

--On 3/7/06 11:42 AM -0600 Dean Willis wrote:
>
> On Mar 6, 2006, at 6:52 PM, Michael Slavitch wrote:
>
>> Are you serious? How many ISP's offer IPV6?  How many organizations
>> outside academe and a few large enterprises?
>> Do you realize how many people in the real world still use Windows
>> 2000 and Windows 98, including some very large organizations?
>> Given past history Vista will  run on the majority of Windows
>> desktop systems starting around  2010-2011, and that is optimistic.
>>
>> In the meantime, Skype works now.
>>
>> If P2PSIP is limited to IPV6 we might as well call it a day right
>> now and start licensing runtime modules from Skype.
>>
>>
>> On 3/6/06, Frederick Lefebvre <fred at perspectiveip.ca> wrote: What
>> about mandating the use IPv6 and leave the NAT traversal issue to
>> the IPv6 transition mechanism used?  Wouldn't it make the
>> implementation of p2p-sip easier if the protocol did not have to work
>> around NATs?
>> Of course, feel free to wonder if this is a realistic proposal or not.
>> But with the upcoming Windows Vista, all of a sudden a fairly large
>> percentage of users will have V6 connectivity and it might not sound
>> so crazy anymore.
>
>
>
> I believe the suggestion was that we not develop a P2P-SIP specific
> mechanism for NAT traversal on IPV6 networks, but leave that to the
> IPV6 transition mechanism.
>
> This is far different from suggesting that we not support IPV4.

Whatever we develop, it should fully cover both, IPv4 and IPv6.

Thanks,

    Juergen


From Saverio.Niccolini at netlab.nec.de  Wed Mar  8 09:11:36 2006
From: Saverio.Niccolini at netlab.nec.de (Saverio Niccolini)
Date: Wed, 8 Mar 2006 15:11:36 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <6D28EBC684A4D94096217AD2FE40087324AA2A@venus.office>

Hi all,

> Karst,
> 
> The draft does not make any specific assumption about IPv4 or IPv6.
> Particularly, it does not make the (potentially wrong) assumption
> that if the entire Internet had moved to IPv6, there would be no
> NATs anymore.

and moreover if all the workd has IPv6 without NAT and Firewalls, then
there
is no problem for P2P-SIP. You just get your IPv6 address and use this.
Thus no reason to include it is our draft since we are writing about
"Problem Statement for SIP-signalled Peer-to-Peer Communication across
Middleboxes".

Saverio

> Thanks,
> 
>     Juergen
> 
> --On 3/4/06 8:30 AM +0100 Karst Bjorgson wrote:
> 
> >
> >
> >
> > --- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
> >
> >> Secondly, the draft says its an analysis of existing
> >> nat traversal
> >> techniques for sip. However, it omits the main SIP
> >> nat traversal
> >> techinque in its entirety - ICE and sip-outbound. I
> >> dont know how an
> >> analysis of the applicability of existing nat
> >> traversal techniques can
> >> be considered complete when it has missed the main
> >> standards in this area.
> >
> > The draft also fails to mention IPv6. I believe this
> > is a bit short-sighted. Microsoft is making a big push
> > for IPv6 in Windows Vista. Vista will include an
> > implementation of Teredo (RFC 4380), which provides
> > some pretty interesting NAT traversal facilities.
> >
> > -- Karst
> >
> > __________________________________________________
> > Do You Yahoo!?
> > Tired of spam?  Yahoo! Mail has the best spam protection around
> > http://mail.yahoo.com
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From slavitch at gmail.com  Wed Mar  8 09:27:08 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Wed, 8 Mar 2006 09:27:08 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <B2E4533F10A0C5BB0D809B73@10.1.1.171>
References: <4409B28B.2000004@research.panasonic.com>
	<20060306063651.52150.qmail@web53006.mail.yahoo.com>
	<3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>
	<1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>
	<2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>
	<B2E4533F10A0C5BB0D809B73@10.1.1.171>
Message-ID: <1fc32c8f0603080627g20b4ae5h2519297ca3ccbe76@mail.gmail.com>

A participating P2P node MUST NOT neet to support IPV6. IPV6 may exist in
the mesh, but it must not be the defining characteristics of the mesh
itself, otherwise we are limiting ourselves once again to orthodoxy.  There
are several well-known techniques already discussed in this list that makes
the underlying addressing protocols abstract entitites.  They work both in
theory and in practise, are well-founded, and should be considered. They are
not open research topics, they are new elaborations of well-understood
techniques.  The only matter left is determining if the underlying
abstraction identifiers are robust enough and can map into SIP without any
changes or harm.

P2P overlays can traverse any protocol that has the following
characteristics:

1. A known consistant datalink layer.
2. A known consistant socket mechanism.
3. An addressing scheme that can be overloaded by some well-published or
well-known identifier.

That's it.

There are mathematical proofs for this.  This is not guesswork.

The protocol underneath need not be IPV6. It can be IPV4, IPX,  or even
DECNET.  With a little work and crossed fingers you could support X.25 and
HDLC.

I have personal experience in this matter as a I developed similar overlay
systems for peer to peer communications nearly a decade ago, systems that
completely abstracted the underlying addressing schemes with unique
identifiers in order to transparently traverse not only NATs but protocol
gateways and without changes to the network layer or existing protocol
addressing schemes.  It not only worked for IPV4 and IPV6 but worked for
IPX, DECnet,  and other proprietary protocols, and required minor changes
only to top-line application and database schemas, not the underlying
systems themselves.
In that industry the procurement cycles take upwards of a decade and the
equipment is used for a very long time in very distant places, so upgrades
are not possible.  If an obscure programmer could do this a decade ago there
is no excuse for the IETF to say it is not possible nor a good idea now.

Michael

On 3/8/06, Juergen Quittek <quittek at netlab.nec.de> wrote:
>
>
> Whatever we develop, it should fully cover both, IPv4 and IPv6.
>
> Thanks,
>
>     Juergen
>
>


--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060308/8bb62201/attachment.html

From henry at pulver.com  Wed Mar  8 09:55:00 2006
From: henry at pulver.com (Henry Sinnreich)
Date: Wed, 8 Mar 2006 08:55:00 -0600
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603080627g20b4ae5h2519297ca3ccbe76@mail.gmail.com>
Message-ID: <200603081455.k28Et7Cx008329@cs.columbia.edu>

Michael,

I wish you could submit an I-D to the hopefully soon to be formed P2P SIP WG
discussing these issues.

Henry
  

  _____  

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Michael Slavitch
Sent: Wednesday, March 08, 2006 8:27 AM
To: Juergen Quittek; P2P-SIP
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal

 

A participating P2P node MUST NOT neet to support IPV6. IPV6 may exist in
the mesh, but it must not be the defining characteristics of the mesh
itself, otherwise we are limiting ourselves once again to orthodoxy.  There
are several well-known techniques already discussed in this list that makes
the underlying addressing protocols abstract entitites.  They work both in
theory and in practise, are well-founded, and should be considered. They are
not open research topics, they are new elaborations of well-understood
techniques.  The only matter left is determining if the underlying
abstraction identifiers are robust enough and can map into SIP without any
changes or harm.  

P2P overlays can traverse any protocol that has the following
characteristics:

1. A known consistant datalink layer.
2. A known consistant socket mechanism.
3. An addressing scheme that can be overloaded by some well-published or
well-known identifier. 

That's it.   

There are mathematical proofs for this.  This is not guesswork.

The protocol underneath need not be IPV6. It can be IPV4, IPX,  or even
DECNET.  With a little work and crossed fingers you could support X.25 and
HDLC.

I have personal experience in this matter as a I developed similar overlay
systems for peer to peer communications nearly a decade ago, systems that
completely abstracted the underlying addressing schemes with unique
identifiers in order to transparently traverse not only NATs but protocol
gateways and without changes to the network layer or existing protocol
addressing schemes.  It not only worked for IPV4 and IPV6 but worked for
IPX, DECnet,  and other proprietary protocols, and required minor changes
only to top-line application and database schemas, not the underlying
systems themselves.  
In that industry the procurement cycles take upwards of a decade and the
equipment is used for a very long time in very distant places, so upgrades
are not possible.  If an obscure programmer could do this a decade ago there
is no excuse for the IETF to say it is not possible nor a good idea now. 

Michael

On 3/8/06, Juergen Quittek < <mailto:quittek at netlab.nec.de>
quittek at netlab.nec.de> wrote:


Whatever we develop, it should fully cover both, IPv4 and IPv6.

Thanks,

    Juergen




-- 
Michael Slavitch
Ottawa, Ontario Canada 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060308/7625522d/attachment.html

From gregd at research.panasonic.com  Wed Mar  8 10:04:04 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Wed, 08 Mar 2006 10:04:04 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <FE2F347265F6D37957DF496F@[10.1.1.171]>
References: <156450F9DBF9883343076F10@[192.168.1.128]> 
	<6E8ADCC4-1FE5-47AA-90CD-F1592D777F0D@netlab.nec.de> 
	<FE2F347265F6D37957DF496F@[10.1.1.171]>
Message-ID: <440EF264.5030603@research.panasonic.com>

Hi Juergen.

Juergen Quittek wrote:
[cut]
>>> Section 4.2.1 asserts that middlebox detection is needed for non-UDP
>>> flows. I half agree with this. I agree that we need to make tcp work,
>>> and support configurations where there is no intermediary, even if its
>>> another user in the p2p network. The draft asserts that there are no
>>> solutions for that, but its just not true. ICE-tcp
>>> (draft-ietf-mmusic-ice-tcp) will allow for creation of direct tcp
> 
> Please note that draft-ietf-mmusic-ice-tcp-00 has the same publication
> date as the P2P SIP NAT traversal draft. So, it is no wonder that ICE TCP
> is not yet included.
> 
> The draft does not state or assert that there is no solution for TCP
> NAT traversal.  But it states that there are no known tools for
> performing middlebox detection for protocols other than UDP. And
> after reading draft-ietf-mmusic-ice-tcp-00 I still think this is correct.

This is the WG document.  There were prior individual submissions
which could be referred to.

Let's be very careful not to re-invent things in our rush to
define requirements for P2P-SIP.

I know this is not the intention, but if we fail to regard the
body of work available, we will make too much work for ourselves.

Greg

From slavitch at objectworld.com  Wed Mar  8 10:00:44 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Wed, 8 Mar 2006 10:00:44 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C62D@photon.objectworld.com>

Will do Henry.  
 
M
 
 
Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1 613 599 9698 225       Ottawa, Ontario, Canada
 

________________________________

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Henry Sinnreich
Sent: March 8, 2006 9:55 AM
To: 'Michael Slavitch'; 'Juergen Quittek'; 'P2P-SIP'
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal



Michael,

I wish you could submit an I-D to the hopefully soon to be formed P2P
SIP WG discussing these issues.

Henry
  

________________________________

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Michael Slavitch
Sent: Wednesday, March 08, 2006 8:27 AM
To: Juergen Quittek; P2P-SIP
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal

 

A participating P2P node MUST NOT neet to support IPV6. IPV6 may exist
in the mesh, but it must not be the defining characteristics of the mesh
itself, otherwise we are limiting ourselves once again to orthodoxy.
There are several well-known techniques already discussed in this list
that makes the underlying addressing protocols abstract entitites.  They
work both in theory and in practise, are well-founded, and should be
considered. They are not open research topics, they are new elaborations
of well-understood techniques.  The only matter left is determining if
the underlying abstraction identifiers are robust enough and can map
into SIP without any changes or harm.  

P2P overlays can traverse any protocol that has the following
characteristics:

1. A known consistant datalink layer.
2. A known consistant socket mechanism.
3. An addressing scheme that can be overloaded by some well-published or
well-known identifier. 

That's it.   

There are mathematical proofs for this.  This is not guesswork.

The protocol underneath need not be IPV6. It can be IPV4, IPX,  or even
DECNET.  With a little work and crossed fingers you could support X.25
and HDLC.

I have personal experience in this matter as a I developed similar
overlay systems for peer to peer communications nearly a decade ago,
systems that completely abstracted the underlying addressing schemes
with unique identifiers in order to transparently traverse not only NATs
but protocol gateways and without changes to the network layer or
existing protocol addressing schemes.  It not only worked for IPV4 and
IPV6 but worked for IPX, DECnet,  and other proprietary protocols, and
required minor changes only to top-line application and database
schemas, not the underlying systems themselves.  
In that industry the procurement cycles take upwards of a decade and the
equipment is used for a very long time in very distant places, so
upgrades are not possible.  If an obscure programmer could do this a
decade ago there is no excuse for the IETF to say it is not possible nor
a good idea now. 

Michael

On 3/8/06, Juergen Quittek < quittek at netlab.nec.de
<mailto:quittek at netlab.nec.de> > wrote:


Whatever we develop, it should fully cover both, IPv4 and IPv6.

Thanks,

    Juergen




-- 
Michael Slavitch
Ottawa, Ontario Canada 

________________________________


Spam
> 
Not spam
> 
Forget previous vote
> 


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060308/9e332f05/attachment-0001.html

From quittek at netlab.nec.de  Wed Mar  8 10:15:10 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Wed, 08 Mar 2006 16:15:10 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603080627g20b4ae5h2519297ca3ccbe76@mail.gmail.com>
References: <4409B28B.2000004@research.panasonic.com>	
	<20060306063651.52150.qmail@web53006.mail.yahoo.com>	
	<3f96a9130603061633m2c45627al8c41e8c5fbe47dba@mail.gmail.com>	
	<1fc32c8f0603061652h582d667et90877e48e7c1535a@mail.gmail.com>	
	<2B5DDC25-DE17-4F7C-8E81-7DAD57CFB113@softarmor.com>	
	<B2E4533F10A0C5BB0D809B73@10.1.1.171>
	<1fc32c8f0603080627g20b4ae5h2519297ca3ccbe76@mail.gmail.com>
Message-ID: <02EC53B2449729F5C513BA45@[10.1.1.171]>

Hi Michael,

--On 3/8/06 9:27 AM -0500 Michael Slavitch wrote:

> A participating P2P node MUST NOT neet to support IPV6. IPV6 may

Probably I should have phrased more carefully.  My apologies not
not being clear enough.

I just wanted to state that whatever we develop, it should be
possible to implement it for an IPv4 environment as well as for
an IPv6 environment (probably in different ways).  This does not
imply that every implementation must fully support both (which
would not make too much sense).

Still, we can think about whether or not, an IPv6 client needs
to be become aware that its public IP address is an IPv4 address
(or vice versa). The current SIP/SDP specification would require
this.

> exist in the mesh, but it must not be the defining characteristics
> of the mesh itself, otherwise we are limiting ourselves once again
> to orthodoxy.  There are several well-known
> techniques already discussed in this list that makes the underlying
> addressing protocols abstract entitites.  They work both in theory

This is a nice way of dealing with the IPv4/IPv6 problem that
certainly should be considered as design choice for our P2P SIP
solution.

> and in practise, are well-founded, and should be considered. They
> are not open research topics, they are new
> elaborations of well-understood techniques.  The only matter left
> is determining if the underlying abstraction identifiers are robust
> enough and can map into SIP without any changes or harm.
>
> P2P overlays can traverse any protocol that has the following characteristics:
>
> 1. A known consistant datalink layer.
> 2. A known consistant socket mechanism.
> 3. An addressing scheme that can be overloaded by some well-published
> or well-known identifier.
>
> That's it.

Right ... except for setting it up. when you bootstrap the overlay
or connect a new node, you might have to care about the ugly NATted
world underneath.

> There are mathematical proofs for this.  This is not guesswork.
>
> The protocol underneath need not be IPV6. It can be IPV4, IPX,  or
> even DECNET.  With a little work and crossed fingers you could
> support X.25 and HDLC.
>
> I have personal experience in this matter as a I developed similar
> overlay systems for peer to peer communications nearly a decade ago,
> systems that completely abstracted the underlying addressing schemes
> with unique identifiers in order to
> transparently traverse not only NATs but protocol gateways and without
> changes to the network layer or existing protocol addressing schemes.
> It not only worked for IPV4 and IPV6 but worked for IPX, DECnet,  and
> other proprietary protocols, and
> required minor changes only to top-line application and database
> schemas, not the underlying systems themselves.
> In that industry the procurement cycles take upwards of a decade and
> the equipment is used for a very long time in very distant places, so
> upgrades are not possible.  If an obscure programmer could do this a
> decade ago there is no excuse for the IETF
> to say it is not possible nor a good idea now.

Did anyone say so?

> Michael

I do not feel to be in any disagreement with you.
Please consider that we are in an early phase right now
and many design decisions have not been made yet.

If you refer to the draft on problem statements for NAT traversal:
It is not intended to limit the solution space for P2P SIP in any way.
I just lists potential issues with SIP and middlebox detection when
designing the P2P SIP solution.  We will make sure that we state this
very clearly in the next version.

Thanks,

    Juergen

> On 3/8/06, Juergen Quittek < quittek at netlab.nec.de> wrote:
>
>
> Whatever we develop, it should fully cover both, IPv4 and IPv6.
>
> Thanks,
>
>     Juergen
>
>
>
>
>
> --
> Michael Slavitch
> Ottawa, Ontario Canada



From slavitch at objectworld.com  Wed Mar  8 10:38:20 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Wed, 8 Mar 2006 10:38:20 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C631@photon.objectworld.com>

Juergen, good feedback. Thank you.

Comments inline. Apologies in advance for any misquotes.

Juergen:  

>I just wanted to state that whatever we develop, it should be possible
to implement it for an IPv4 environment as well as >for an IPv6
environment (probably in different ways).  This does not imply that
every implementation must fully support both (which would not make too
much sense).

Me:

I agree. But we have the power to make supporting 'any' underlying
transport transparent.

>Still, we can think about whether or not, an IPv6 client needs to be
become aware that its public IP address is an IPv4 
>address (or vice versa). The current SIP/SDP specification would
require this.

Me:  

If you force a UA to use numeric addressing that is true. IF a UA can
use name-based addressing it is different.
This may require subtle changes to the SIP/SDP specifications. If that
name can be abstracted the world opens up.
Note that the name could be a numeric address, a DNS address, or some
kind of overloaded address. Once the addressing is abstract so is the
underlying transport. This is the essence of a possible solution. 

(my original email):
>> exist in the mesh, but it must not be the defining characteristics of

>> the mesh itself, otherwise we are limiting ourselves once again to 
>> orthodoxy.  There are several well-known techniques already discussed

>> in this list that makes the underlying addressing protocols abstract 
>> entitites.  They work both in theory

Juergen:

>This is a nice way of dealing with the IPv4/IPv6 problem that certainly
should be considered as design choice for our P2P 
>SIP solution.

I agree. I'm also very glad that other people have already done most of
the work. :%)

> and in practise, are well-founded, and should be considered. They are 
> not open research topics, they are new elaborations of well-understood

> techniques.  The only matter left is determining if the underlying 
> abstraction identifiers are robust enough and can map into SIP without

> any changes or harm.
>
> P2P overlays can traverse any protocol that has the following
characteristics:
>
> 1. A known consistant datalink layer.
> 2. A known consistant socket mechanism.
> 3. An addressing scheme that can be overloaded by some well-published 
> or well-known identifier.
>
> That's it.

>>Right ... except for setting it up. when you bootstrap the overlay or
connect a new node, you might have to care about >>the ugly NATted world
underneath.

Michael (my key comment):  

Not if the bootstrapping is done out of band. Each device we use can
fetch things out of band with different protocols.
OOB can mean anything, from a slip of paper with a code in it to an
http:// url, to an SMS message, or anything.

Done well bootstrapping can be done by carrier pigeon. It doesn't have
to be fancy or elegant. Elegance may be a bad idea.
It should however be simple and flexible.

For example, a whole system can be simultaneously bootstrapped by an
email sent to a list. It's ugly, but it is good. 
In existing rugged implementations multiple modes of bootstrapping are a
requirement in case the standard method is quite literally blown up out
of existence. In that case responsibility must be delegated to secondary
methods, no matter how inelegant or distasteful.

M


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From quittek at netlab.nec.de  Wed Mar  8 10:40:48 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Wed, 08 Mar 2006 16:40:48 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <440EF264.5030603@research.panasonic.com>
References: <156450F9DBF9883343076F10@[192.168.1.128]>
	<6E8ADCC4-1FE5-47AA-90CD-F1592D777F0D@netlab.nec.de>
	<FE2F347265F6D37957DF496F@[10.1.1.171]>
	<440EF264.5030603@research.panasonic.com>
Message-ID: <D5FD05E03071BD384E0818A4@[10.1.1.171]>

Greg,

--On 3/8/06 10:04 AM -0500 Greg Daley wrote:

> Hi Juergen.
>
> Juergen Quittek wrote:
> [cut]
>>>> Section 4.2.1 asserts that middlebox detection is needed for non-UDP
>>>> flows. I half agree with this. I agree that we need to make tcp work,
>>>> and support configurations where there is no intermediary, even if its
>>>> another user in the p2p network. The draft asserts that there are no
>>>> solutions for that, but its just not true. ICE-tcp
>>>> (draft-ietf-mmusic-ice-tcp) will allow for creation of direct tcp
>>
>> Please note that draft-ietf-mmusic-ice-tcp-00 has the same publication
>> date as the P2P SIP NAT traversal draft. So, it is no wonder that ICE TCP
>> is not yet included.
>>
>> The draft does not state or assert that there is no solution for TCP
>> NAT traversal.  But it states that there are no known tools for
>> performing middlebox detection for protocols other than UDP. And
>> after reading draft-ietf-mmusic-ice-tcp-00 I still think this is correct.
>
> This is the WG document.  There were prior individual submissions
> which could be referred to.

Yes, there was Jonathan's draft out since October.
Also we are missing the entire ICE story.
Still, in the draft we talked about 'existing tools'.
I hope that ICE TCP will soon be something I can use as tool.

> Let's be very careful not to re-invent things in our rush to
> define requirements for P2P-SIP.

That's the point of misunderstanding.  We tried in the draft
not to limit the solution space in any way. Intentionally, we
avoided talking about 'requirements'. We just listed 'issues
to be considered' developing P2P SIP.
I'm sorry for communicating this badly in the draft.

> I know this is not the intention, but if we fail to regard the
> body of work available, we will make too much work for ourselves.

Completely agreed.  The draft charter intentionally says "Select
firewall/NAT traversal technique(s)" and not "Develop ...".

> Greg

Thanks,

    Juergen

> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip



From quittek at netlab.nec.de  Wed Mar  8 10:59:19 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Wed, 08 Mar 2006 16:59:19 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4E2DC1ACE541B94A946B8D8657E372C528C631@photon.objectworld.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C631@photon.objectworld.com>
Message-ID: <73FC95839281061783492703@[10.1.1.171]>

Michael,

I like your view.  We have implemented this concept already
and are promoting it.  But currently, we limit promotion to
research projects.  I am not sure, this technology is already
mature enough for standardization at the IETF.  Let's see.

And let's see what the (future) working group thinks about it.

Thanks,

    Juergen


--On 3/8/06 10:38 AM -0500 Michael Slavitch wrote:

> Juergen, good feedback. Thank you.
>
> Comments inline. Apologies in advance for any misquotes.
>
> Juergen:
>
>> I just wanted to state that whatever we develop, it should be possible
> to implement it for an IPv4 environment as well as >for an IPv6
> environment (probably in different ways).  This does not imply that
> every implementation must fully support both (which would not make too
> much sense).
>
> Me:
>
> I agree. But we have the power to make supporting 'any' underlying
> transport transparent.
>
>> Still, we can think about whether or not, an IPv6 client needs to be
> become aware that its public IP address is an IPv4
>> address (or vice versa). The current SIP/SDP specification would
> require this.
>
> Me:
>
> If you force a UA to use numeric addressing that is true. IF a UA can
> use name-based addressing it is different.
> This may require subtle changes to the SIP/SDP specifications. If that
> name can be abstracted the world opens up.
> Note that the name could be a numeric address, a DNS address, or some
> kind of overloaded address. Once the addressing is abstract so is the
> underlying transport. This is the essence of a possible solution.
>
> (my original email):
>>> exist in the mesh, but it must not be the defining characteristics of
>
>>> the mesh itself, otherwise we are limiting ourselves once again to
>>> orthodoxy.  There are several well-known techniques already discussed
>
>>> in this list that makes the underlying addressing protocols abstract
>>> entitites.  They work both in theory
>
> Juergen:
>
>> This is a nice way of dealing with the IPv4/IPv6 problem that certainly
> should be considered as design choice for our P2P
>> SIP solution.
>
> I agree. I'm also very glad that other people have already done most of
> the work. :%)
>
>> and in practise, are well-founded, and should be considered. They are
>> not open research topics, they are new elaborations of well-understood
>
>> techniques.  The only matter left is determining if the underlying
>> abstraction identifiers are robust enough and can map into SIP without
>
>> any changes or harm.
>>
>> P2P overlays can traverse any protocol that has the following
> characteristics:
>>
>> 1. A known consistant datalink layer.
>> 2. A known consistant socket mechanism.
>> 3. An addressing scheme that can be overloaded by some well-published
>> or well-known identifier.
>>
>> That's it.
>
>>> Right ... except for setting it up. when you bootstrap the overlay or
> connect a new node, you might have to care about >>the ugly NATted world
> underneath.
>
> Michael (my key comment):
>
> Not if the bootstrapping is done out of band. Each device we use can
> fetch things out of band with different protocols.
> OOB can mean anything, from a slip of paper with a code in it to an
> http:// url, to an SMS message, or anything.
>
> Done well bootstrapping can be done by carrier pigeon. It doesn't have
> to be fancy or elegant. Elegance may be a bad idea.
> It should however be simple and flexible.
>
> For example, a whole system can be simultaneously bootstrapped by an
> email sent to a list. It's ugly, but it is good.
> In existing rugged implementations multiple modes of bootstrapping are a
> requirement in case the standard method is quite literally blown up out
> of existence. In that case responsibility must be delegated to secondary
> methods, no matter how inelegant or distasteful.
>
> M
>
>
> -----------------------------------------------
>
> This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this
> information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or
> reproduction of this transmission by unintended recipients is not authorized and may be unlawful.
>
> Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp.
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip



From lowekamp at cs.wm.edu  Wed Mar  8 11:01:02 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Wed, 8 Mar 2006 11:01:02 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <FE2F347265F6D37957DF496F@10.1.1.171>
References: <156450F9DBF9883343076F10@192.168.1.128>
	<44070BE1.3030205@cisco.com>
	<6E8ADCC4-1FE5-47AA-90CD-F1592D777F0D@netlab.nec.de>
	<FE2F347265F6D37957DF496F@10.1.1.171>
Message-ID: <e9132f820603080801q586408e0w64162b0d8bc984bb@mail.gmail.com>

On 3/8/06, Juergen Quittek <quittek at netlab.nec.de> wrote:
> >> The draft presumes that any nat traversal solution requires a step
> >> where
> >> the presence of nat is detected, and typed (section 2.1.2). I strongly
> >> disagree with this. The ice solution does not attempt to discover or
> >> otherwise characterize nats at all. Indeed, one of our main
> >> experiences
> >> with rfc3489 was that the whole notion of nat discovery and type
> >> computation was flawed. That is why this functionality was removed
> >> from
> >> rfc3489bis almost two years ago.
> >
> > Sure, it is quite difficult to characterize NATs in general, but
> > nodes should
> > be free to run their NAT classification procedure. Where
> > classification does
> > not refer to "this is kind of NAT type X" but more "does it work what
> > I want
> > to achieve".
>
> Yes, the draft should state this more clearly.
> Since we do not have a common NAT taxonomy, it would be very difficult
> to clearly identify a NAT type.  But it will be comparably easy to identify
> which kind of communication is possible and how it can be achieved.
>

I've found draft-ietf-behave-nat-udp-04 to be very useful in terms of
describing the behavior of NATs.  As others have pointed out,
ultimately the only test is whether the NAT does what you need it to 
(and I have a couple on my desk that characterize the right way but
fail applications in bizarre ways), but this draft (which has been
submitted for publication) is useful for the common terminology it
provides.

Ultimately, I too hope that whatever P2P SIP protocol we come up with
will simply say "use ICE" (or something similar), and we can all go
out and have a beer when it comes time to discuss NATs.  Sadly, I
suspect getting to that point will take a lot of work on requirements.

Bruce


From eunsoo at research.panasonic.com  Wed Mar  8 11:51:44 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Wed, 08 Mar 2006 11:51:44 -0500
Subject: [p2p-sip] 50% of broadband US/UK MSN users behind NAT
In-Reply-To: <Pine.GSO.4.58.0603071823480.27087@flame.cs.columbia.edu>
References: <Pine.GSO.4.58.0603071823480.27087@flame.cs.columbia.edu>
Message-ID: <440F0BA0.20707@research.panasonic.com>

Thanks for the interesting information.
I thought actually more (broadband) users were behind NAT.
Still a significant portion of broadband users have public IP addresses. 
It's good.
Thanks.

Eunsoo

Salman Abdul Baset wrote:

>http://research.microsoft.com/~pablo/papers/CCR.pdf (CCR Jan 2006)
>(pointed in a post on p2p-hackers list)
>
>-salman
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From eunsoo at research.panasonic.com  Wed Mar  8 11:55:52 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Wed, 08 Mar 2006 11:55:52 -0500
Subject: [p2p-sip] I-D: An Architecture for P2P SIP
In-Reply-To: <000001c64295$5dd442f0$3837a40a@china.huawei.com>
References: <000001c64295$5dd442f0$3837a40a@china.huawei.com>
Message-ID: <440F0C98.3050602@research.panasonic.com>

Hi,

Content search is not a focus of P2P SIP and currently out of scope by 
the proposed charter.
So this list is not a good place to discuss about content search.
Having said that, please let me say a few words about DHT and content 
search.
The basic DHT is not good when the resource name is not known precisely.
But if right keywords can be easily guessed for resources, DHT can be 
used for keyword-based (content) search.
There are a number of literature about how to design content search over 
DHT.
eDonkey's eMule client is using Kademlia, a DHT.
Thanks.

Eunsoo

Seaward Hu wrote:

>Hello Mr.Shim,
>        Can the DHT support content search? I mean the key word which
>putted by ON may be not as accurate as those stored in the DHT network.
>
> 
> 
> 
> 
> 
> 
>Sincerely yours,
>Seaward
>
> 
>************************************************************************
>***************
>This e-mail and its attachments contain confidential information from
>HUAWEI, which is intended only for the person or entity whose address is
>listed above. Any use of the information contained herein in any way
>(including, but not limited to, total or partial disclosure,
>reproduction, or dissemination) by persons other than the intended
>recipient(s) is prohibited. If you receive this e-mail in error, please
>notify the sender by phone or email immediately and delete it!
>************************************************************************
>*****************
>
>
>  
>
>>-----Original Message-----
>>From: p2p-sip-bounces at cs.columbia.edu 
>>[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Eunsoo Shim
>>Sent: Tuesday, February 28, 2006 7:22 AM
>>To: P2P-SIP
>>Subject: [p2p-sip] I-D: An Architecture for P2P SIP
>>
>>
>>We submitted an I-D that describes an architecture for P2P SIP.
>>You can get it from the link below until it appears in the IETF I-D 
>>directory.
>>We hope this draft can help the architecture and charter 
>>discussion in BOF.
>>Your comments will be appreciated.
>>Thanks.
>>
>>Eunsoo
>>
>>    
>>
>http://www.comet.columbia.edu/~eunsoo/ietf/draft-shim-sipping-p2p-arch-0
>0.txt
>
>Abstract
>
>   This document describes an architecture for peer-to-peer SIP systems
>   in which a separate and independent peer-to-peer overlay layer
>   provides a distributed resource placement and search service for SIP.
>   It also aims to help identify what should be specified for
>   interoperation.
>
>
>
>
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From henry at pulver.com  Wed Mar  8 12:04:28 2006
From: henry at pulver.com (Henry Sinnreich)
Date: Wed, 8 Mar 2006 11:04:28 -0600
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <73FC95839281061783492703@[10.1.1.171]>
Message-ID: <200603081704.k28H4Z7n014326@cs.columbia.edu>

> I am not sure, this technology is already mature enough for 
> standardization at the IETF.

About 80% of the Internet traffic is P2P and it does NAT traversal. 
This should be good enough.

Thanks, Henry


 
-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Juergen Quittek
Sent: Wednesday, March 08, 2006 9:59 AM
To: Michael Slavitch; Michael Slavitch; P2P-SIP
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal

Michael,

I like your view.  We have implemented this concept already
and are promoting it.  But currently, we limit promotion to
research projects.  I am not sure, this technology is already
mature enough for standardization at the IETF.  Let's see.

And let's see what the (future) working group thinks about it.

Thanks,

    Juergen




From quittek at netlab.nec.de  Wed Mar  8 12:37:29 2006
From: quittek at netlab.nec.de (Juergen Quittek)
Date: Wed, 08 Mar 2006 18:37:29 +0100
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <200603081704.k28H4Z7n014326@cs.columbia.edu>
References: <200603081704.k28H4Z7n014326@cs.columbia.edu>
Message-ID: <61AE1B22A8E1F10D55F34E9D@[10.1.1.171]>

Henry,

--On 3/8/06 11:04 AM -0600 Henry Sinnreich wrote:

>> I am not sure, this technology is already mature enough for
>> standardization at the IETF.
>
> About 80% of the Internet traffic is P2P and it does NAT traversal.
> This should be good enough.

But 95% or more of this P2P traffic deals with concrete IP(v4)
addresses and is a little bit more down to earth compared to
Michael's view when it solves the NAT traversal problem.
This could be good enough.

Thanks,

    Juergen

> Thanks, Henry
>
>
>
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Juergen Quittek
> Sent: Wednesday, March 08, 2006 9:59 AM
> To: Michael Slavitch; Michael Slavitch; P2P-SIP
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>
> Michael,
>
> I like your view.  We have implemented this concept already
> and are promoting it.  But currently, we limit promotion to
> research projects.  I am not sure, this technology is already
> mature enough for standardization at the IETF.  Let's see.
>
> And let's see what the (future) working group thinks about it.
>
> Thanks,
>
>     Juergen
>
>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip



From slavitch at objectworld.com  Wed Mar  8 12:55:41 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Wed, 8 Mar 2006 12:55:41 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C644@photon.objectworld.com>

All the better! This means is that there are no excuses! 


Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: Juergen Quittek [mailto:quittek at netlab.nec.de] 
Sent: March 8, 2006 12:37 PM
To: henry at pulver.com; Michael Slavitch; 'Michael Slavitch'; 'P2P-SIP'
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal

Henry,

--On 3/8/06 11:04 AM -0600 Henry Sinnreich wrote:

>> I am not sure, this technology is already mature enough for 
>> standardization at the IETF.
>
> About 80% of the Internet traffic is P2P and it does NAT traversal.
> This should be good enough.

But 95% or more of this P2P traffic deals with concrete IP(v4) addresses
and is a little bit more down to earth compared to Michael's view when
it solves the NAT traversal problem.
This could be good enough.

Thanks,

    Juergen

> Thanks, Henry
>
>
>
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Juergen Quittek
> Sent: Wednesday, March 08, 2006 9:59 AM
> To: Michael Slavitch; Michael Slavitch; P2P-SIP
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>
> Michael,
>
> I like your view.  We have implemented this concept already and are 
> promoting it.  But currently, we limit promotion to research projects.

> I am not sure, this technology is already mature enough for 
> standardization at the IETF.  Let's see.
>
> And let's see what the (future) working group thinks about it.
>
> Thanks,
>
>     Juergen
>
>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip



-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From slavitch at gmail.com  Wed Mar  8 13:25:50 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Wed, 8 Mar 2006 13:25:50 -0500
Subject: [p2p-sip] 50% of broadband US/UK MSN users behind NAT
In-Reply-To: <440F0BA0.20707@research.panasonic.com>
References: <Pine.GSO.4.58.0603071823480.27087@flame.cs.columbia.edu>
	<440F0BA0.20707@research.panasonic.com>
Message-ID: <1fc32c8f0603081025g69052096ja6ca4a676eb11012@mail.gmail.com>

A very helpful paper.  Some food for thought:

1. Many broadband users connect directly from their personal computers,
either direct or by PPPOE, which would explain the lower numbers.

2. Many enthusiasts open up ports to one DMZ host endpoint as a courtesy to
others.  For serious MMRPG gamers this is a standard practice.
    It helps their teams win.   Keep an eye on the MMRPG efforts as they are
the ones truly pushing the envelope.

3. The other nodes in such networks still must follow NAT principles.



On 3/8/06, Eunsoo Shim <eunsoo at research.panasonic.com> wrote:
>
> Thanks for the interesting information.
> I thought actually more (broadband) users were behind NAT.
> Still a significant portion of broadband users have public IP addresses.
> It's good.
> Thanks.
>
> Eunsoo
>
> Salman Abdul Baset wrote:
>
> >http://research.microsoft.com/~pablo/papers/CCR.pdf<http://research.microsoft.com/%7Epablo/papers/CCR.pdf>(CCR Jan 2006)
> >(pointed in a post on p2p-hackers list)
> >
> >-salman
> >_______________________________________________
> >p2p-sip mailing list
> >p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>



--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060308/2347d2ee/attachment.html

From slavitch at gmail.com  Wed Mar  8 13:30:20 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Wed, 8 Mar 2006 13:30:20 -0500
Subject: [p2p-sip] 50% of broadband US/UK MSN users behind NAT
In-Reply-To: <1fc32c8f0603081025g69052096ja6ca4a676eb11012@mail.gmail.com>
References: <Pine.GSO.4.58.0603071823480.27087@flame.cs.columbia.edu>
	<440F0BA0.20707@research.panasonic.com>
	<1fc32c8f0603081025g69052096ja6ca4a676eb11012@mail.gmail.com>
Message-ID: <1fc32c8f0603081030r2dbe1531m308b02c43f8898f1@mail.gmail.com>

One example: http://solipsis.netofpeers.net/wiki2/index.php/Main_Page

On 3/8/06, Michael Slavitch <slavitch at gmail.com> wrote:
>
> A very helpful paper.  Some food for thought:
>
> 1. Many broadband users connect directly from their personal computers,
> either direct or by PPPOE, which would explain the lower numbers.
>
> 2. Many enthusiasts open up ports to one DMZ host endpoint as a courtesy
> to others.  For serious MMRPG gamers this is a standard practice.
>     It helps their teams win.   Keep an eye on the MMRPG efforts as they
> are the ones truly pushing the envelope.
>
> 3. The other nodes in such networks still must follow NAT principles.
>
>
>
>
> On 3/8/06, Eunsoo Shim < eunsoo at research.panasonic.com> wrote:
> >
> > Thanks for the interesting information.
> > I thought actually more (broadband) users were behind NAT.
> > Still a significant portion of broadband users have public IP addresses.
> > It's good.
> > Thanks.
> >
> > Eunsoo
> >
> > Salman Abdul Baset wrote:
> >
> > >http://research.microsoft.com/~pablo/papers/CCR.pdf
> > <http://research.microsoft.com/%7Epablo/papers/CCR.pdf> (CCR Jan 2006)
> > >(pointed in a post on p2p-hackers list)
> > >
> > >-salman
> > >_______________________________________________
> > >p2p-sip mailing list
> > >p2p-sip at cs.columbia.edu
> > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > >
> > >
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
>
>
>
> --
> Michael Slavitch
> Ottawa, Ontario Canada
>



--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060308/7eae4aae/attachment.html

From huang-ming.pan at comcast.net  Thu Mar  9 19:49:32 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Thu, 9 Mar 2006 16:49:32 -0800
Subject: [p2p-sip] p2p-sip Digest, Vol 8, Issue 14
References: <mailman.38.1141923624.23074.p2p-sip@cs.columbia.edu>
Message-ID: <001901c643dc$7f8beae0$6400a8c0@comcast.net>

A p2p overlay weaver, with emulator and visualizer. FYI.
http://overlayweaver.sourceforge.net/doc/overview/

peter
----- Original Message -----
From: <p2p-sip-request at cs.columbia.edu>
To: <p2p-sip at cs.columbia.edu>
Sent: Thursday, March 09, 2006 9:00 AM
Subject: p2p-sip Digest, Vol 8, Issue 14


> Send p2p-sip mailing list submissions to
> p2p-sip at cs.columbia.edu
>
> To subscribe or unsubscribe via the World Wide Web, visit
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> or, via email, send a message with subject or body 'help' to
> p2p-sip-request at cs.columbia.edu
>
> You can reach the person managing the list at
> p2p-sip-owner at cs.columbia.edu
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of p2p-sip digest..."
>


----------------------------------------------------------------------------
----


> Today's Topics:
>
>    1. Re: Problem statement for P2P SIP NAT traversal (Henry Sinnreich)
>    2. Re: Problem statement for P2P SIP NAT traversal (Juergen Quittek)
>    3. Re: Problem statement for P2P SIP NAT traversal (Michael Slavitch)
>    4. Re: 50% of broadband US/UK MSN users behind NAT (Michael Slavitch)
>    5. Re: 50% of broadband US/UK MSN users behind NAT (Michael Slavitch)
>


----------------------------------------------------------------------------
----


> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>



From ekr at networkresonance.com  Sun Mar 12 21:35:38 2006
From: ekr at networkresonance.com (Eric Rescorla)
Date: Sun, 12 Mar 2006 18:35:38 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
Message-ID: <20060313024122.82E97222418@laser.networkresonance.com>

OVERVIEW
This draft describes a protocol and architecture for doing
"P2P SIP". What this means in practice is that registration
information is stored in a Distributed Hash Table, thus
in principle removing the need for centralized servers.


GENERAL COMMENTS

Use of SIP for DHT Functions
----------------------------
The protocol used in this draft uses SIP both for the usual
signalling functions and for DHT maintenance. This strikes
me as a poor design decision.

1. It means that while you will be able to use generic 
   DHT algorithms you won't be able to use a generic
   DHT service. There's nothing particularly special
   that this application requires from a DHT, so it seems
   like a classic architectural mistake to tie yourself
   to something specific.

2. It obscures what should be a clean architectural separation
   between lookup services (what the DHT is being used for)
   and signalling services (what SIP is used for now). This
   made some sense in the previous draft which actually 
   punned on the SIP messages (using INVITE to find the
   UA for the desired contact) but now that you're using
   REGISTER in order to explicitly separate these functions,
   there's really no value added here.

3. It's really confusing.



Security
--------
There are a number of known security issues with DHTs. In
the specific design described here:

1. There is no authentication for names. The authors write:

   "The lack of central authority to resolve name disputes in the
   namespace means that at some level this problem is unsolved.  The
   approach has tended to be to allow everyone to call themselves Wally
   then let the certificates sort them out.  Users with names that are
   often "stolen" by others will learn that theirs is a poor choice of
   name because it is too valuable, and they will select a less valuable
   name.  Equilibrium will prevail, or chaos."

I don't really understand how this works. If the attacker can 
insert itself at a single replica, it can create a systematic
inconsistency about the appropriate holder of a name. This
makes it very easy to mount an attack on a specific target. 
This draft seems to assume that it's specific names that are
valuable, but in many cases it's not the name itself but rather
the identity of the name holder that's valuable, so changing
your name won't help, at last not for long.

2. You need some mechanism for resolving which replica is 
correct. Otherwise, any inconsistency leads to what...?

3. The protection against Sybil attacks depends entirely on the
attacker not having access to a large block of routable
addresses. This is questionable at best with Pv4 and silly with
v6. Designing a protocol which is only secure on v4 seems unwise.
The discussion of this in 10.11 is clearly inadequate and I
don't really see how it can work unless you assume specific
address boundaries. 

It's not clear to me what constraints lead to this design. 
The use cases documents talk a lot about not necessarily
having access to a global nameservice, but the kind of
non-real-time access required for secure naming (e.g.,
to get DNSSEC "certificates") is very different from requiring
a full-time service. Having any kind of distributed naming
service obviates many of the security problems listed above.


These aren't security problems that can be solved by gluing on 
some crypto at the end of the day. Security for DHTs is a
major unsolved research problem, so this is rather worrisome.


DETAILED COMMENTS
S 7.2.1:

   ipport         =  IPV4address ":" port

And v6?


S 10.4:

   A P2P environment can do little to protect against an individual node
   compromising the registrations it is responsible for.  Accordingly, a
   UA cannot trust a response from a single node, whether it indicates a
   successful search or an error.  In the absence of other methods of
   verifying the response (such as having a certificate of the user
   being searched for and a signed registration that can be verified
   with the certificate) a UA should search for the primary registration
   and at least one replica.  Because the locations the replicas are
   stored are unrelated to the location of the primary registration, a
   single attacker is unlikely to be able to compromise both entries.
   As the overlay gains more nodes and more replicas are searched for,
   the odds of a compromise are reduced.

It's not clear to me why you think your replica model has superior
security properties to the "next N successors" model that chord uses.
In both cases, the attacker needs to compromise (or compromise
routing to) N nodes. And since the preimages of the next N node IDs
are randomly distributed in IP space, I don't see an advantage
here. Please explain.

S 10.7:

   computer.  Cao et al, [9] describe an approach that separates the
   storage of resource location information from the actual storage of
   the offline research.

Is "research" really the word you want here?











   











From dbarrett at quinthar.com  Sun Mar 12 23:24:51 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Sun, 12 Mar 2006 20:24:51 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <20060313024122.82E97222418@laser.networkresonance.com>
Message-ID: <200603130426.k2D4QGOZ012536@cs.columbia.edu>

> -----Original Message-----
> From: Eric Rescorla
> Sent: Sunday, March 12, 2006 6:36 PM
> To: p2p-sip at cs.columbia.edu
> Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
> 
> 1. It means that while you will be able to use generic
>    DHT algorithms you won't be able to use a generic
>    DHT service. There's nothing particularly special
>    that this application requires from a DHT, so it seems
>    like a classic architectural mistake to tie yourself
>    to something specific.

It's my impression we're attempting to create P2P SIP, not a general DHT
service (though such a thing would be cool).  If we succeed in the former
while failing in the latter, that's still success in my eyes.


> 2. It obscures what should be a clean architectural separation
>    between lookup services (what the DHT is being used for)
>    and signalling services (what SIP is used for now). This
>    made some sense in the previous draft which actually
>    punned on the SIP messages (using INVITE to find the
>    UA for the desired contact) but now that you're using
>    REGISTER in order to explicitly separate these functions,
>    there's really no value added here.

While I can understand this perspective, I disagree.  I think a key failing
of SIP today is that it has far, far too many "clean architectural
separations" -- so much so that it's impossible to do anything without
reading hundreds of pages of highly redundant documentation.

I rather like the idea of just extending the base SIP protocol to include
the DHT, rather than rolling an entirely new protocol (*cough* ICE *cough*)
from scratch, using its own parsers, own state machines, own eccentricities,
etc.


> These aren't security problems that can be solved by gluing on
> some crypto at the end of the day. Security for DHTs is a
> major unsolved research problem, so this is rather worrisome.

Given that a secure DHT is an unsolved problem, and given the IETF only
deals with solved problems, should we:

1) Standardize the best practices today, despite their insecurity.

	-- OR --

2) Wait until a secure DHT has been invented and reconvene.

I vote for #1.  I certainly don't vote for the unstated #3: "Simultaneously
invent and standardize a new variant of DHT that is secure (for some
definition of 'secure')."

-david


From dbarrett at quinthar.com  Sun Mar 12 23:24:51 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Sun, 12 Mar 2006 20:24:51 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <FE2F347265F6D37957DF496F@[10.1.1.171]>
Message-ID: <200603130426.k2D4Q9OZ012512@cs.columbia.edu>

> -----Original Message-----
> From: Juergen Quittek
> Sent: Wednesday, March 08, 2006 3:27 AM
> To: Martin Stiemerling; Jonathan Rosenberg
> Cc: P2P-SIP
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> >> Secondly, the draft says its an analysis of existing nat traversal
> >> techniques for sip. However, it omits the main SIP nat traversal
> >> techinque in its entirety - ICE and sip-outbound. I dont know how an
> >> analysis of the applicability of existing nat traversal techniques can
> >> be considered complete when it has missed the main standards in
> >> this area.
> >
> > Granted. We have missed ICE, even though it is well-known. This is not
> > because of ignorance but due to a very fast approaching -00 deadline.
> > As said above, the document's intention is not to open the box for a
> > new protocols, but to explorer the existing NAT traversal space.
> 
> Yes, the lack of ICE coverage is really a flaw.  But it should be no
> problem to fix it for the next version.  Also, I do not see a big impact
> on the identified issues that we should take care of when developing
> P2P SIP.

My impression was that we would define what we need from NAT traversal, and
then evaluate all the options (ICE included).  Furthermore, once our
requirements are specified, I thought we were going to remain option to new
NAT traversal options, and not presuppose ICE as the final choice (otherwise
we should just pick it now).

For example, would we be open to the following chain of events?

1) We publish what P2P-SIP requires in the way of NAT traversal
2) Somebody invents a protocol that meets these requirements
3) We adopt this new protocol, despite it not being ICE

If we're not, then let's just say "P2P-SIP will use ICE" and be done with
it.  Granted, I believe that'd be a mistake, but I respect that reasonable
minds can disagree on this point.

-david



From dbarrett at quinthar.com  Sun Mar 12 23:26:07 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Sun, 12 Mar 2006 20:26:07 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <FE2F347265F6D37957DF496F@[10.1.1.171]>
Message-ID: <200603130426.k2D4QGOZ012537@cs.columbia.edu>

> -----Original Message-----
> From: Juergen Quittek
> Sent: Wednesday, March 08, 2006 3:27 AM
> To: Martin Stiemerling; Jonathan Rosenberg
> Cc: P2P-SIP
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> >> Secondly, the draft says its an analysis of existing nat traversal
> >> techniques for sip. However, it omits the main SIP nat traversal
> >> techinque in its entirety - ICE and sip-outbound. I dont know how an
> >> analysis of the applicability of existing nat traversal techniques can
> >> be considered complete when it has missed the main standards in
> >> this area.
> >
> > Granted. We have missed ICE, even though it is well-known. This is not
> > because of ignorance but due to a very fast approaching -00 deadline.
> > As said above, the document's intention is not to open the box for a
> > new protocols, but to explorer the existing NAT traversal space.
> 
> Yes, the lack of ICE coverage is really a flaw.  But it should be no
> problem to fix it for the next version.  Also, I do not see a big impact
> on the identified issues that we should take care of when developing
> P2P SIP.

My impression was that we would define what we need from NAT traversal, and
then evaluate all the options (ICE included).  Furthermore, once our
requirements are specified, I thought we were going to remain option to new
NAT traversal options, and not presuppose ICE as the final choice (otherwise
we should just pick it now).

For example, would we be open to the following chain of events?

1) We publish what P2P-SIP requires in the way of NAT traversal
2) Somebody invents a protocol that meets these requirements
3) We adopt this new protocol, despite it not being ICE

If we're not, then let's just say "P2P-SIP will use ICE" and be done with
it.  Granted, I believe that'd be a mistake, but I respect that reasonable
minds can disagree on this point.

-david



From bryan at ethernot.org  Sun Mar 12 23:54:04 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Sun, 12 Mar 2006 23:54:04 -0500
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <20060313024122.82E97222418@laser.networkresonance.com>
References: <20060313024122.82E97222418@laser.networkresonance.com>
Message-ID: <8b2769930603122054x6f6b5b8ej85d67452f18300f9@mail.gmail.com>

I'll comment a bit about the SIP for DHTs question, REGISTER vs INVITE
and the replica key vs. successor choice. See inline...

On 3/12/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> OVERVIEW
> This draft describes a protocol and architecture for doing
> "P2P SIP". What this means in practice is that registration
> information is stored in a Distributed Hash Table, thus
> in principle removing the need for centralized servers.
>
>
> GENERAL COMMENTS
>
> Use of SIP for DHT Functions
> ----------------------------
> The protocol used in this draft uses SIP both for the usual
> signalling functions and for DHT maintenance. This strikes
> me as a poor design decision.
>
> 1. It means that while you will be able to use generic
>    DHT algorithms you won't be able to use a generic
>    DHT service. There's nothing particularly special
>    that this application requires from a DHT, so it seems
>    like a classic architectural mistake to tie yourself
>    to something specific.
>
> 2. It obscures what should be a clean architectural separation
>    between lookup services (what the DHT is being used for)
>    and signalling services (what SIP is used for now). This
>    made some sense in the previous draft which actually
>    punned on the SIP messages (using INVITE to find the
>    UA for the desired contact) but now that you're using
>    REGISTER in order to explicitly separate these functions,
>    there's really no value added here.

There are sort of 2 poins here. Using SIP for the DHT, and using
REGISTER to locate resources independently of INVITE.

So for point 1, there simply isn't, to my knowledge, a defined IETF
protocol for a DHT or to encode DHT messages. There is no specified
DHT protocol, so we either have to 1) use SIP or 2) invent a new
protocol or encoding. Now of course we can use all kinds of things --
HTTP to MSRP -- to push the info around, but why not SIP? The devices
I have in mind for this already speak SIP.

Now if there is a standard protocol for DHT messages (not a library
like OpenDHT, but a true protocol) I can definitely get on board with
that and think we could (and should) use that. As far as I know, there
isn't one approved by the IETF, and it seemed a task quite a bit
beyond my expertise (and the scope of a draft intended for SIPPING) to
try to suggest a generic DHT. So in summary, yep -- I agree if we had
a really good DHT protocol defined this should definitely use it, but
I don't know if one and I'm not sure designing it here is the right
approach. Any suggestions on this one?

In light of above perhaps the point would be moot if we had a DHT
protocol, but assuming for a moment that we don't have a generic DHT
protocol and we use SIP, and I'll talk about point 2.  Why the change
to REGISTER for search and not the original INVITE? We made that
change for two reasons. It actually seems more in line with the SIP
semantics. REGISTER is used in RFC 3261 to store and query
registrations, which is really what we are doing here, so we went to
using REGISTER for it. It also makes implementations much easier than
our previous INVITE doing DHT thing, since an unmodified UA on top of
a P2P lookup functions normally, and similarly, the only "odd"
messages -- that is, messages about DHTs -- floating around are
REGISTERS -- something a UA that is not P2P is likely to reject if it
receives, so interaction with non-P2P endpoints seems cleaner.

>
> 3. It's really confusing.
>

I suppose I can't really argue that ;) Suggestions to make it cleaner?
I sort of hoped that breaking it so things were done with register --
essentially only adding and querying registrations -- it would be
easier to understand.

>
>
> Security
> --------
> There are a number of known security issues with DHTs. In
> the specific design described here:
>
> 1. There is no authentication for names. The authors write:
>
>    "The lack of central authority to resolve name disputes in the
>    namespace means that at some level this problem is unsolved.  The
>    approach has tended to be to allow everyone to call themselves Wally
>    then let the certificates sort them out.  Users with names that are
>    often "stolen" by others will learn that theirs is a poor choice of
>    name because it is too valuable, and they will select a less valuable
>    name.  Equilibrium will prevail, or chaos."
>
> I don't really understand how this works. If the attacker can
> insert itself at a single replica, it can create a systematic
> inconsistency about the appropriate holder of a name. This
> makes it very easy to mount an attack on a specific target.
> This draft seems to assume that it's specific names that are
> valuable, but in many cases it's not the name itself but rather
> the identity of the name holder that's valuable, so changing
> your name won't help, at last not for long.
>
> 2. You need some mechanism for resolving which replica is
> correct. Otherwise, any inconsistency leads to what...?
>
> 3. The protection against Sybil attacks depends entirely on the
> attacker not having access to a large block of routable
> addresses. This is questionable at best with Pv4 and silly with
> v6. Designing a protocol which is only secure on v4 seems unwise.
> The discussion of this in 10.11 is clearly inadequate and I
> don't really see how it can work unless you assume specific
> address boundaries.
>
> It's not clear to me what constraints lead to this design.
> The use cases documents talk a lot about not necessarily
> having access to a global nameservice, but the kind of
> non-real-time access required for secure naming (e.g.,
> to get DNSSEC "certificates") is very different from requiring
> a full-time service. Having any kind of distributed naming
> service obviates many of the security problems listed above.
>
>
> These aren't security problems that can be solved by gluing on
> some crypto at the end of the day. Security for DHTs is a
> major unsolved research problem, so this is rather worrisome.
>
>
> DETAILED COMMENTS
> S 7.2.1:
>
>    ipport         =  IPV4address ":" port
>
> And v6?
>
>
> S 10.4:
>
>    A P2P environment can do little to protect against an individual node
>    compromising the registrations it is responsible for.  Accordingly, a
>    UA cannot trust a response from a single node, whether it indicates a
>    successful search or an error.  In the absence of other methods of
>    verifying the response (such as having a certificate of the user
>    being searched for and a signed registration that can be verified
>    with the certificate) a UA should search for the primary registration
>    and at least one replica.  Because the locations the replicas are
>    stored are unrelated to the location of the primary registration, a
>    single attacker is unlikely to be able to compromise both entries.
>    As the overlay gains more nodes and more replicas are searched for,
>    the odds of a compromise are reduced.
>
> It's not clear to me why you think your replica model has superior
> security properties to the "next N successors" model that chord uses.
> In both cases, the attacker needs to compromise (or compromise
> routing to) N nodes. And since the preimages of the next N node IDs
> are randomly distributed in IP space, I don't see an advantage
> here. Please explain.

So here you see one of our open problems, listed at the end. We are
using IP address for hash to get IDs, but we want to allow users to
have multiple nodes behind NATs, so we have suggested in the open
questions appending a port number to the hashed ID of the NAT. That
makes it difficult for an attacker in Sybil mode to simply generate
tons of IDs from one IP address simply by changing the port, as would
occur if we hashed ip:port. But, the downside is that it is trivial to
get sequential ones in such a case. Thus, we suggest the replica key.

Now as you point out, this is still fatally flawed in IPv6. This draft
tries to avoid any central servers at all, and I am increasingly
convinced we need centrally issued certs with corresponding centrally
issued IDs to really make this work, but in here I tried to avoid
those all together.

I fully agree that security is still a very hard problem, one that
must be solved, and one we haven't addressed very well yet.

David

> S 10.7:
>
>    computer.  Cao et al, [9] describe an approach that separates the
>    storage of resource location information from the actual storage of
>    the offline research.
>
> Is "research" really the word you want here?
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From ekr at networkresonance.com  Mon Mar 13 00:39:07 2006
From: ekr at networkresonance.com (Eric Rescorla)
Date: Sun, 12 Mar 2006 21:39:07 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <8b2769930603122054x6f6b5b8ej85d67452f18300f9@mail.gmail.com>
	(David A. Bryan's message of "Sun, 12 Mar 2006 23:54:04 -0500")
References: <20060313024122.82E97222418@laser.networkresonance.com>
	<8b2769930603122054x6f6b5b8ej85d67452f18300f9@mail.gmail.com>
Message-ID: <86acburjic.fsf@raman.networkresonance.com>

"David A. Bryan" <bryan at ethernot.org> writes:
> Now if there is a standard protocol for DHT messages (not a library
> like OpenDHT, but a true protocol) I can definitely get on board with
> that and think we could (and should) use that. As far as I know, there
> isn't one approved by the IETF, and it seemed a task quite a bit
> beyond my expertise (and the scope of a draft intended for SIPPING) to
> try to suggest a generic DHT. So in summary, yep -- I agree if we had
> a really good DHT protocol defined this should definitely use it, but
> I don't know if one and I'm not sure designing it here is the right
> approach. Any suggestions on this one?

It's not clear to me why inventing an entirely new protocol is better
than simply standardizing a specific instance of (and documenting)
the protocol in a pre-existing DHT, like, say Bamboo... 


>> 3. It's really confusing.
>>
>
> I suppose I can't really argue that ;) Suggestions to make it cleaner?
> I sort of hoped that breaking it so things were done with register --
> essentially only adding and querying registrations -- it would be

Yes. Describe the system as generic operations rather than as
SIP operations and then describe a binding to SIP.


>>
>> Security
>> --------
>> There are a number of known security issues with DHTs. In
>> the specific design described here:
>>
>> 1. There is no authentication for names. The authors write:
>>
>>    "The lack of central authority to resolve name disputes in the
>>    namespace means that at some level this problem is unsolved.  The
>>    approach has tended to be to allow everyone to call themselves Wally
>>    then let the certificates sort them out.  Users with names that are
>>    often "stolen" by others will learn that theirs is a poor choice of
>>    name because it is too valuable, and they will select a less valuable
>>    name.  Equilibrium will prevail, or chaos."
>>
>> I don't really understand how this works. If the attacker can
>> insert itself at a single replica, it can create a systematic
>> inconsistency about the appropriate holder of a name. This
>> makes it very easy to mount an attack on a specific target.
>> This draft seems to assume that it's specific names that are
>> valuable, but in many cases it's not the name itself but rather
>> the identity of the name holder that's valuable, so changing
>> your name won't help, at last not for long.
>>
>> 2. You need some mechanism for resolving which replica is
>> correct. Otherwise, any inconsistency leads to what...?
>>
>> 3. The protection against Sybil attacks depends entirely on the
>> attacker not having access to a large block of routable
>> addresses. This is questionable at best with Pv4 and silly with
>> v6. Designing a protocol which is only secure on v4 seems unwise.
>> The discussion of this in 10.11 is clearly inadequate and I
>> don't really see how it can work unless you assume specific
>> address boundaries.
>>
>> It's not clear to me what constraints lead to this design.
>> The use cases documents talk a lot about not necessarily
>> having access to a global nameservice, but the kind of
>> non-real-time access required for secure naming (e.g.,
>> to get DNSSEC "certificates") is very different from requiring
>> a full-time service. Having any kind of distributed naming
>> service obviates many of the security problems listed above.
>>
>>
>> These aren't security problems that can be solved by gluing on
>> some crypto at the end of the day. Security for DHTs is a
>> major unsolved research problem, so this is rather worrisome.
>>
>>
>> DETAILED COMMENTS
>> S 7.2.1:
>>
>>    ipport         =  IPV4address ":" port
>>
>> And v6?
>>
>>
>> S 10.4:
>>
>>    A P2P environment can do little to protect against an individual node
>>    compromising the registrations it is responsible for.  Accordingly, a
>>    UA cannot trust a response from a single node, whether it indicates a
>>    successful search or an error.  In the absence of other methods of
>>    verifying the response (such as having a certificate of the user
>>    being searched for and a signed registration that can be verified
>>    with the certificate) a UA should search for the primary registration
>>    and at least one replica.  Because the locations the replicas are
>>    stored are unrelated to the location of the primary registration, a
>>    single attacker is unlikely to be able to compromise both entries.
>>    As the overlay gains more nodes and more replicas are searched for,
>>    the odds of a compromise are reduced.
>>
>> It's not clear to me why you think your replica model has superior
>> security properties to the "next N successors" model that chord uses.
>> In both cases, the attacker needs to compromise (or compromise
>> routing to) N nodes. And since the preimages of the next N node IDs
>> are randomly distributed in IP space, I don't see an advantage
>> here. Please explain.
>
> So here you see one of our open problems, listed at the end. We are
> using IP address for hash to get IDs, but we want to allow users to
> have multiple nodes behind NATs, so we have suggested in the open
> questions appending a port number to the hashed ID of the NAT. That
> makes it difficult for an attacker in Sybil mode to simply generate
> tons of IDs from one IP address simply by changing the port, as would
> occur if we hashed ip:port. But, the downside is that it is trivial to
> get sequential ones in such a case. Thus, we suggest the replica key.

OK. I think you need to be clearer that this is why you want to
use keyed replicas, since you only discuss this appending
idea in the Open Issues section rather than in 7.4.2 where
the algorithm includes only the hash of teh port.


> Now as you point out, this is still fatally flawed in IPv6.

It's not just IPv6. There are serious security problems even
if you ignore the Sybil attack, as indicated above in my review.


-Ekr

From ekr at networkresonance.com  Mon Mar 13 00:40:24 2006
From: ekr at networkresonance.com (Eric Rescorla)
Date: Sun, 12 Mar 2006 21:40:24 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <20060313043201.71229222418@laser.networkresonance.com> (David
	Barrett's message of "Sun, 12 Mar 2006 20:24:51 -0800")
References: <20060313043201.71229222418@laser.networkresonance.com>
Message-ID: <8664mirjg7.fsf@raman.networkresonance.com>

"David Barrett" <dbarrett at quinthar.com> writes:
>> -----Original Message-----
>> From: Eric Rescorla
>> These aren't security problems that can be solved by gluing on
>> some crypto at the end of the day. Security for DHTs is a
>> major unsolved research problem, so this is rather worrisome.
>
> Given that a secure DHT is an unsolved problem, and given the IETF only
> deals with solved problems, should we:
>
> 1) Standardize the best practices today, despite their insecurity.
>
> 	-- OR --
>
> 2) Wait until a secure DHT has been invented and reconvene.
>
> I vote for #1.  I certainly don't vote for the unstated #3: "Simultaneously
> invent and standardize a new variant of DHT that is secure (for some
> definition of 'secure')."

There's a point below which the security of the system is so bad
that it shouldn't be standardized. The question is whether
that's the case here.

-Ekr

From dbarrett at quinthar.com  Mon Mar 13 01:19:00 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Sun, 12 Mar 2006 22:19:00 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <8664mirjg7.fsf@raman.networkresonance.com>
Message-ID: <200603130619.k2D6JAOZ004347@cs.columbia.edu>

> -----Original Message-----
> From: Eric Rescorla [mailto:ekr at networkresonance.com]
> 
> "David Barrett" <dbarrett at quinthar.com> writes:
> >
> > Given that a secure DHT is an unsolved problem, and given the IETF only
> > deals with solved problems, should we:
> >
> > 1) Standardize the best practices today, despite their insecurity.
> >
> > 	-- OR --
> >
> > 2) Wait until a secure DHT has been invented and reconvene.
> >
> > I vote for #1.  I certainly don't vote for the unstated #3:
> "Simultaneously
> > invent and standardize a new variant of DHT that is secure (for some
> > definition of 'secure')."
> 
> There's a point below which the security of the system is so bad
> that it shouldn't be standardized. The question is whether
> that's the case here.

That's a fair question, how would you answer it?

If your answer is "yes, DHTs are so insecure that standardization is
pointless" then would you advise that we dissolve and reconvene later?

If your answer is "no, DHTs aren't so insecure so as to make standardization
pointless" then would you advise that we continue on our current path?

Or do you have another answer such as "yes, the DHT we're currently
discussing is so insecure that standardization is pointless, and therefore
we should <insert alternate plan>."

I'm in the second camp (keep on the current path, despite outstanding
security concerns).  I'd be eager to hear something in the third camp (solve
the security problems).  I'm not in favor of the first camp.

Which camp are you advocating?

-david


From jdrosen at cisco.com  Mon Mar 13 02:20:46 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 02:20:46 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4m1p4d$7d5mfd@sj-inbound-d.cisco.com>
References: <4m1p4d$7d5mfd@sj-inbound-d.cisco.com>
Message-ID: <44151D4E.1030304@cisco.com>

I think the burden of proof would be to demonstrate that ICE cannot and 
doesn't, for some reason, work for p2p SIP.

ICE, as you may know, is itself a p2p nat traversal technique. Its 
efficacy is predicated on both sides supporting it. Thus, if we want 
interop between normal SIP clients and p2p sip clients, NAT traversal 
for everyone needs to be ICE based for things to work well.

This, of course, applies to lots of other things that sip does, which 
require endpoint support. If you are going to start throwing all of that 
away and defining new variants of sip extensions and features for p2p 
clients, you should drop the 'sip' part, and recognize that you are 
defining a new non-sip protocol for p2p. That would be a terrible travesty.

-Jonathan R.

David Barrett wrote:
>>-----Original Message-----
>>From: Juergen Quittek
>>Sent: Wednesday, March 08, 2006 3:27 AM
>>To: Martin Stiemerling; Jonathan Rosenberg
>>Cc: P2P-SIP
>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>>
>>>>Secondly, the draft says its an analysis of existing nat traversal
>>>>techniques for sip. However, it omits the main SIP nat traversal
>>>>techinque in its entirety - ICE and sip-outbound. I dont know how an
>>>>analysis of the applicability of existing nat traversal techniques can
>>>>be considered complete when it has missed the main standards in
>>>>this area.
>>>
>>>Granted. We have missed ICE, even though it is well-known. This is not
>>>because of ignorance but due to a very fast approaching -00 deadline.
>>>As said above, the document's intention is not to open the box for a
>>>new protocols, but to explorer the existing NAT traversal space.
>>
>>Yes, the lack of ICE coverage is really a flaw.  But it should be no
>>problem to fix it for the next version.  Also, I do not see a big impact
>>on the identified issues that we should take care of when developing
>>P2P SIP.
> 
> 
> My impression was that we would define what we need from NAT traversal, and
> then evaluate all the options (ICE included).  Furthermore, once our
> requirements are specified, I thought we were going to remain option to new
> NAT traversal options, and not presuppose ICE as the final choice (otherwise
> we should just pick it now).
> 
> For example, would we be open to the following chain of events?
> 
> 1) We publish what P2P-SIP requires in the way of NAT traversal
> 2) Somebody invents a protocol that meets these requirements
> 3) We adopt this new protocol, despite it not being ICE
> 
> If we're not, then let's just say "P2P-SIP will use ICE" and be done with
> it.  Granted, I believe that'd be a mistake, but I respect that reasonable
> minds can disagree on this point.
> 
> -david
> 

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From dbarrett at quinthar.com  Mon Mar 13 02:31:44 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Sun, 12 Mar 2006 23:31:44 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <44151D4E.1030304@cisco.com>
Message-ID: <200603130731.k2D7VnOZ020879@cs.columbia.edu>

> -----Original Message-----
> From: Jonathan Rosenberg [mailto:jdrosen at cisco.com]
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> I think the burden of proof would be to demonstrate that ICE cannot and
> doesn't, for some reason, work for p2p SIP.

That's fair.  Let's first define what 'works for P2P-SIP' is, then we can
have a productive discussion on this point.  Specifically, I mean we should
define the NAT-traversal requirements of P2P-SIP without regard to a
solution (eg, without presupposing ICE), and then evaluate whether and to
what degree ICE meets these requirements.

Clearly, if ICE is the best solution, let's take it.  But if ICE isn't the
best solution, let's not take it, and instead find what is.  

I'm sorry, I didn't mean to present this as something new; I thought this
was already the consensus.

-david


From lowekamp at cs.wm.edu  Mon Mar 13 07:42:27 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Mon, 13 Mar 2006 07:42:27 -0500
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <86acburjic.fsf@raman.networkresonance.com>
References: <20060313024122.82E97222418@laser.networkresonance.com>
	<8b2769930603122054x6f6b5b8ej85d67452f18300f9@mail.gmail.com>
	<86acburjic.fsf@raman.networkresonance.com>
Message-ID: <e9132f820603130442hd0b6482i8c5891ba8c02738@mail.gmail.com>

On 3/13/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> >> 3. It's really confusing.
> >>
> >
> > I suppose I can't really argue that ;) Suggestions to make it cleaner?
> > I sort of hoped that breaking it so things were done with register --
> > essentially only adding and querying registrations -- it would be
>
> Yes. Describe the system as generic operations rather than as
> SIP operations and then describe a binding to SIP.

As the current draft is organized (and as I'm sure you're aware),
section 6 gives an overview of the generic operations, 7  does
synteax, and 8 describes specific semantics.  I disbelieve that it
would be possible to write something that would be totally generic,
then list a short set of SIP bindings.  While I realize you want to
find a "generic" DHT, there are a few problems with that.  One is that
no "generic" "standard" DHT exists, another is that to define one, we
would have to define all kinds of protocol syntax, headers, routing,
redirecting, etc.  My belief is that if we tried to write something
more generic and then said "hey, look, here are some SIP bindings that
happen to fit our protocol," someone who is more familiar with another
protocol would look at it and say that we failed to describe something
necessary (that we simply assume SIP provides, like tags), or we make
an assumption that is unique to SIP.

Fortunately, the SIP devices we are targetting have a SIP stack that
already handles what we need, so we build on top of it.  If we try to
define something more generic, we get into extensive issues of what
the minimum requirements are to support it, etc.

Bruce


From lowekamp at cs.wm.edu  Mon Mar 13 08:11:05 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Mon, 13 Mar 2006 08:11:05 -0500
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <20060313024122.82E97222418@laser.networkresonance.com>
References: <20060313024122.82E97222418@laser.networkresonance.com>
Message-ID: <e9132f820603130511j3bcf3d8dg23fa5c752bbba63c@mail.gmail.com>

On 3/12/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> Security
> --------
> There are a number of known security issues with DHTs. In
> the specific design described here:
>
> 1. There is no authentication for names. The authors write:
>
>    "The lack of central authority to resolve name disputes in the
>    namespace means that at some level this problem is unsolved.  The
>    approach has tended to be to allow everyone to call themselves Wally
>    then let the certificates sort them out.  Users with names that are
>    often "stolen" by others will learn that theirs is a poor choice of
>    name because it is too valuable, and they will select a less valuable
>    name.  Equilibrium will prevail, or chaos."
>
> I don't really understand how this works. If the attacker can
> insert itself at a single replica, it can create a systematic
> inconsistency about the appropriate holder of a name. This
> makes it very easy to mount an attack on a specific target.
> This draft seems to assume that it's specific names that are
> valuable, but in many cases it's not the name itself but rather
> the identity of the name holder that's valuable, so changing
> your name won't help, at last not for long.
>

Hmmm.  This may not be clear from the draft (and actually, I think
parts of the draft may not have been updated to reflect this thought,
which is entirely our fault), but resource name -> location is not a
one-to-one mapping, it's one-to-many (as conventional SIP supports
multiple Contacts in a REGISTER).  As defined right now, some of those
contacts could be forged.  Without further extension, clearly our
protocol defines no way for a UA to identify the "correct" mappings. 
Thus, without some sort of certificates or central authority, there is
no way to verify that the contact registration you are currently
examining is legitimate.   I don't think I would want to deploy this
current protocol on a public network without certificates.

The goal is to leave it easy to insert forged registrations, but
impossible to make someone believe they are real.

> 2. You need some mechanism for resolving which replica is
> correct. Otherwise, any inconsistency leads to what...?
>

If you have the above certificates, it's easy to verify.

> 3. The protection against Sybil attacks depends entirely on the
> attacker not having access to a large block of routable
> addresses. This is questionable at best with Pv4 and silly with
> v6. Designing a protocol which is only secure on v4 seems unwise.
> The discussion of this in 10.11 is clearly inadequate and I
> don't really see how it can work unless you assume specific
> address boundaries.
>
> It's not clear to me what constraints lead to this design.
> The use cases documents talk a lot about not necessarily
> having access to a global nameservice, but the kind of
> non-real-time access required for secure naming (e.g.,
> to get DNSSEC "certificates") is very different from requiring
> a full-time service. Having any kind of distributed naming
> service obviates many of the security problems listed above.
>
>
> These aren't security problems that can be solved by gluing on
> some crypto at the end of the day. Security for DHTs is a
> major unsolved research problem, so this is rather worrisome.
>
>
> DETAILED COMMENTS
> S 7.2.1:
>
>    ipport         =  IPV4address ":" port
>
> And v6?

You're absolutely right that there is a problem with this level of
security in v6.  First, though, let's address v4.  We've discussed
previously that the port number should probably be removed from the
hashed portion when generating the NodeID.  Given that, I think we can
safely assume that no attacker will have access to a large block of
addresses.  At least, I suspect that anyone who would have that much
control would have to also be in control of routing (by virtue of
either being an ISP or controlling a VLAN we're on), and can attack
the protocol in any way they see fit.

I wrote that BNF and left out v6 quite deliberately, for a couple of reasons:

- security, as you mention, is an interesting issue.  It's a lot more
challenging there, and I'm not personally familiar enough with how
addresses are being assigned in actual ipv6 deployments to address all
of the issues.  I guess the first question is whether we can identify
the portion of the address that is ISP-assigned and the portion where
the host has control.
- transition is difficult for hybrid v4/v6 overlays.  A node needs to
have a single, static NodeID (at least for the duration of each
session associated with the overlay).  If some nodes are v4 and some
are v6, how do we determine the naming domain of the overlay.  I
considered adding another parameter to the node-id header to describe
this, but I'm not quite sure what it should be.  One possible solution
is to adopt a STUN-like system where each node at startup queries one
of a set of servers for that overlay to determine their proper
"mapped" address (except here mapping into the native space of the
overlay).  That would probably have to be signed by the generating
server to make verifying it cheap.

I think there is obviously a lot of work to be done here.


>
>
> S 10.4:
>
>    A P2P environment can do little to protect against an individual node
>    compromising the registrations it is responsible for.  Accordingly, a
>    UA cannot trust a response from a single node, whether it indicates a
>    successful search or an error.  In the absence of other methods of
>    verifying the response (such as having a certificate of the user
>    being searched for and a signed registration that can be verified
>    with the certificate) a UA should search for the primary registration
>    and at least one replica.  Because the locations the replicas are
>    stored are unrelated to the location of the primary registration, a
>    single attacker is unlikely to be able to compromise both entries.
>    As the overlay gains more nodes and more replicas are searched for,
>    the odds of a compromise are reduced.
>
> It's not clear to me why you think your replica model has superior
> security properties to the "next N successors" model that chord uses.
> In both cases, the attacker needs to compromise (or compromise
> routing to) N nodes. And since the preimages of the next N node IDs
> are randomly distributed in IP space, I don't see an advantage
> here. Please explain.
>

The act of inserting a compromised single node gives you an advantage
in compromising the routing to the (non-compromised) successor.  Even
though they're randomly distributed in IP space, they're neighbors in
the overlay, so most other nodes in the network will learn about the
successor through the compromised node.  Therefore, by storing
replicas at successors, the replica is dependent on the possibly
compromised node.  Getting to the successor, even if you know about
the compromised node, also requires some amount of linear search.  By
keying the replicas and distributing them throughout the DHT, there is
no dependency between the N nodes that would be required to compromise
the system, so an attacker would have to be able to compromise N
indepdendent nodes.  Also, searching for each replica is a logarithmic
operation.


> S 10.7:
>
>    computer.  Cao et al, [9] describe an approach that separates the
>    storage of resource location information from the actual storage of
>    the offline research.
>
> Is "research" really the word you want here?
>

ugh.  resources.  fixed in the source...

Thanks,
Bruce


From ekr at networkresonance.com  Mon Mar 13 09:15:39 2006
From: ekr at networkresonance.com (Eric Rescorla)
Date: Mon, 13 Mar 2006 06:15:39 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <20060313062454.B0D54222418@laser.networkresonance.com> (David
	Barrett's message of "Sun, 12 Mar 2006 22:19:00 -0800")
References: <20060313062454.B0D54222418@laser.networkresonance.com>
Message-ID: <861wx6qvlg.fsf@raman.networkresonance.com>

"David Barrett" <dbarrett at quinthar.com> writes:

>> -----Original Message-----
>> From: Eric Rescorla [mailto:ekr at networkresonance.com]
>> 
>> "David Barrett" <dbarrett at quinthar.com> writes:
>> >
>> > Given that a secure DHT is an unsolved problem, and given the IETF only
>> > deals with solved problems, should we:
>> >
>> > 1) Standardize the best practices today, despite their insecurity.
>> >
>> > 	-- OR --
>> >
>> > 2) Wait until a secure DHT has been invented and reconvene.
>> >
>> > I vote for #1.  I certainly don't vote for the unstated #3:
>> "Simultaneously
>> > invent and standardize a new variant of DHT that is secure (for some
>> > definition of 'secure')."
>> 
>> There's a point below which the security of the system is so bad
>> that it shouldn't be standardized. The question is whether
>> that's the case here.
>
> That's a fair question, how would you answer it?
>
> If your answer is "yes, DHTs are so insecure that standardization is
> pointless" then would you advise that we dissolve and reconvene later?
>
> If your answer is "no, DHTs aren't so insecure so as to make standardization
> pointless" then would you advise that we continue on our current path?
>
> Or do you have another answer such as "yes, the DHT we're currently
> discussing is so insecure that standardization is pointless, and therefore
> we should <insert alternate plan>."

As far as I can tell, the combined system described in this document
is so insecure that it should not be standardized. There may be some
hybrid system (e.g., along the lines of that described by Cox et al
<http://www.pdos.lcs.mit.edu/chord/papers/ddns.pdf>), most likely with
externally verified authentication that would be worth standardizing.

-Ekr




   

From ekr at networkresonance.com  Mon Mar 13 09:17:56 2006
From: ekr at networkresonance.com (Eric Rescorla)
Date: Mon, 13 Mar 2006 06:17:56 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <e9132f820603130442hd0b6482i8c5891ba8c02738@mail.gmail.com>
	(Bruce Lowekamp's message of "Mon, 13 Mar 2006 07:42:27 -0500")
References: <20060313024122.82E97222418@laser.networkresonance.com>
	<8b2769930603122054x6f6b5b8ej85d67452f18300f9@mail.gmail.com>
	<86acburjic.fsf@raman.networkresonance.com>
	<e9132f820603130442hd0b6482i8c5891ba8c02738@mail.gmail.com>
Message-ID: <86wteypgx7.fsf@raman.networkresonance.com>

"Bruce Lowekamp" <lowekamp at cs.wm.edu> writes:

> On 3/13/06, Eric Rescorla <ekr at networkresonance.com> wrote:
>> >> 3. It's really confusing.
>> >>
>> >
>> > I suppose I can't really argue that ;) Suggestions to make it cleaner?
>> > I sort of hoped that breaking it so things were done with register --
>> > essentially only adding and querying registrations -- it would be
>>
>> Yes. Describe the system as generic operations rather than as
>> SIP operations and then describe a binding to SIP.
>
> As the current draft is organized (and as I'm sure you're aware),
> section 6 gives an overview of the generic operations, 7  does
> synteax, and 8 describes specific semantics.  I disbelieve that it
> would be possible to write something that would be totally generic,
> then list a short set of SIP bindings.  While I realize you want to
> find a "generic" DHT, there are a few problems with that.  One is that
> no "generic" "standard" DHT exists, another is that to define one, we
> would have to define all kinds of protocol syntax, headers, routing,
> redirecting, etc. My belief is that if we tried to write something
> more generic and then said "hey, look, here are some SIP bindings that
> happen to fit our protocol," someone who is more familiar with another
> protocol would look at it and say that we failed to describe something
> necessary (that we simply assume SIP provides, like tags), or we make
> an assumption that is unique to SIP.

I think we're talking about different things when we say "generic".
What I mean is to pick a DHT (e.g., Chord, Bamboo) and then 
describe your system in terms of the operations that that
DHT provides (INSERT, SUCCESSOR, etc.) Then when you've done that
describe how to implement those messages in SIP.

> Fortunately, the SIP devices we are targetting have a SIP stack that
> already handles what we need, so we build on top of it.  If we try to
> define something more generic, we get into extensive issues of what
> the minimum requirements are to support it, etc.

Yes, I think that kind of architectural division and clarity is
a feature, not a bug.

-Ekr


From ekr at networkresonance.com  Mon Mar 13 09:40:32 2006
From: ekr at networkresonance.com (Eric Rescorla)
Date: Mon, 13 Mar 2006 06:40:32 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <e9132f820603130511j3bcf3d8dg23fa5c752bbba63c@mail.gmail.com>
	(Bruce Lowekamp's message of "Mon, 13 Mar 2006 08:11:05 -0500")
References: <20060313024122.82E97222418@laser.networkresonance.com>
	<e9132f820603130511j3bcf3d8dg23fa5c752bbba63c@mail.gmail.com>
Message-ID: <86slpmpfvj.fsf@raman.networkresonance.com>

"Bruce Lowekamp" <lowekamp at cs.wm.edu> writes:
> On 3/12/06, Eric Rescorla <ekr at networkresonance.com> wrote:
>> These aren't security problems that can be solved by gluing on
>> some crypto at the end of the day. Security for DHTs is a
>> major unsolved research problem, so this is rather worrisome.
>>
>>
>> DETAILED COMMENTS
>> S 7.2.1:
>>
>>    ipport         =  IPV4address ":" port
>>
>> And v6?
>
> You're absolutely right that there is a problem with this level of
> security in v6.  First, though, let's address v4.  We've discussed
> previously that the port number should probably be removed from the
> hashed portion when generating the NodeID.  Given that, I think we can
> safely assume that no attacker will have access to a large block of
> addresses.  At least, I suspect that anyone who would have that much
> control would have to also be in control of routing (by virtue of
> either being an ISP or controlling a VLAN we're on), and can attack
> the protocol in any way they see fit.

No, I don't think that's a safe conclusion. 

1. Many people do have very large blocks of addresses. Xerox,
   for instance, has a /8. 

2. A large NATed network will occupy a large number of
   nodes in the ring--and legitimately--they'll just be
   contiguous and densely packed in your design. [0]
   I haven't thought through the implications of that
   yet.


> I wrote that BNF and left out v6 quite deliberately, for a couple of reasons:
>
> - security, as you mention, is an interesting issue.  It's a lot more
> challenging there, and I'm not personally familiar enough with how
> addresses are being assigned in actual ipv6 deployments to address all
> of the issues.  I guess the first question is whether we can identify
> the portion of the address that is ISP-assigned and the portion where
> the host has control.

In theory yes, in practice, who knows. The addressing architecture
could change.


>> It's not clear to me why you think your replica model has superior
>> security properties to the "next N successors" model that chord uses.
>> In both cases, the attacker needs to compromise (or compromise
>> routing to) N nodes. And since the preimages of the next N node IDs
>> are randomly distributed in IP space, I don't see an advantage
>> here. Please explain.
>>
>
> The act of inserting a compromised single node gives you an advantage
> in compromising the routing to the (non-compromised) successor.  Even
> though they're randomly distributed in IP space, they're neighbors in
> the overlay, so most other nodes in the network will learn about the
> successor through the compromised node.  Therefore, by storing
> replicas at successors, the replica is dependent on the possibly
> compromised node.  Getting to the successor, even if you know about
> the compromised node, also requires some amount of linear search.  By
> keying the replicas and distributing them throughout the DHT, there is
> no dependency between the N nodes that would be required to compromise
> the system, so an attacker would have to be able to compromise N
> indepdendent nodes.  Also, searching for each replica is a logarithmic
> operation.

Sure, but it's an "all your eggs in one basket" versus "many baskets"
things. In the design you describe, the chance that at least one
replica will be compromised by a compromised node is much higher
than in the design where they're all in the successor set. So,
you may always get one valid replica, but there's also a high
chance of inconsistencies. Unless there's some resolution procedure...

In any case, if your goal is to assume that someone gets a
a large subset of the replica set and can then work things out
for themselves, there's quite a large literature about how to protect
routing in the case of some set of compromised nodes. Why do
you think that this is the optimal technique as opposed to
that described, by e.g., Wallach?

-Ekr

[0] Note that this design tends to defeat the goal of having
    even distribution throughout the DHT. Imagine the
    degenerate case of a ring consisting of 1000 machines
    behind a NAT and then a single machine outside the
    NAT. That other node is responsible for 99% of the
    address space. This is also a problem with having
    a partial 

From bryan at ethernot.org  Mon Mar 13 09:51:27 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Mon, 13 Mar 2006 09:51:27 -0500
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <861wx6qvlg.fsf@raman.networkresonance.com>
References: <20060313062454.B0D54222418@laser.networkresonance.com>
	<861wx6qvlg.fsf@raman.networkresonance.com>
Message-ID: <8b2769930603130651x1fbcec02n97878bd5bddb2638@mail.gmail.com>

Just to make sure everyone remembers, this draft isn't really intended
to be standardized in any way as it stands anyway (we say that in the
intro, and I've mentioned that in all my talks), instead it was
proposed as a "hey, here is how we might make this work, this is what
we are experimenting with" strawman, and the current iteration is
changes as a result of lessons learned in implementation, so I will be
the first to agree with Eric that as it stands this ain't even
remotely close to something to consider for standardization. It was
put out as a strawman to shoot at to come up with something useful, so
looks like it is working!

The security section in particular is really old (most of it from the
-00 draft that Cullen and I wrote over a year ago), and hasn't been
updated much since we largely have been working with the message
passing in our coding and wanted to share some things we thought
improved that work. I've actually had some conversations (several with
Eric, actually) where using centrally issued certificates (possibly
rate limited somehow), and storing fully verifiable certificate chains
in the overlay was suggested, and in fact seem the only way to truly
secure something like this. We are likely to look toward something
like that for the next iteration. I realize only addresses *part* of
Eric's concerns anyway, but I just want everyone to be clear about
what this draft is (strawman) and what it isn't (something that in
anything resembling its current form is being suggested as a
standard).

Heck, I'm hoping we get a new WG which can define better requirements
and architecture from many more sources and then we can get some folks
working on a better (and far more secure) idea for how to get this
done.

--David

On 3/13/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> "David Barrett" <dbarrett at quinthar.com> writes:
>
> >> -----Original Message-----
> >> From: Eric Rescorla [mailto:ekr at networkresonance.com]
> >>
> >> "David Barrett" <dbarrett at quinthar.com> writes:
> >> >
> >> > Given that a secure DHT is an unsolved problem, and given the IETF only
> >> > deals with solved problems, should we:
> >> >
> >> > 1) Standardize the best practices today, despite their insecurity.
> >> >
> >> >    -- OR --
> >> >
> >> > 2) Wait until a secure DHT has been invented and reconvene.
> >> >
> >> > I vote for #1.  I certainly don't vote for the unstated #3:
> >> "Simultaneously
> >> > invent and standardize a new variant of DHT that is secure (for some
> >> > definition of 'secure')."
> >>
> >> There's a point below which the security of the system is so bad
> >> that it shouldn't be standardized. The question is whether
> >> that's the case here.
> >
> > That's a fair question, how would you answer it?
> >
> > If your answer is "yes, DHTs are so insecure that standardization is
> > pointless" then would you advise that we dissolve and reconvene later?
> >
> > If your answer is "no, DHTs aren't so insecure so as to make standardization
> > pointless" then would you advise that we continue on our current path?
> >
> > Or do you have another answer such as "yes, the DHT we're currently
> > discussing is so insecure that standardization is pointless, and therefore
> > we should <insert alternate plan>."
>
> As far as I can tell, the combined system described in this document
> is so insecure that it should not be standardized. There may be some
> hybrid system (e.g., along the lines of that described by Cox et al
> <http://www.pdos.lcs.mit.edu/chord/papers/ddns.pdf>), most likely with
> externally verified authentication that would be worth standardizing.
>
> -Ekr
>
>
>
>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From bryan at ethernot.org  Mon Mar 13 09:57:15 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Mon, 13 Mar 2006 09:57:15 -0500
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <86wteypgx7.fsf@raman.networkresonance.com>
References: <20060313024122.82E97222418@laser.networkresonance.com>
	<8b2769930603122054x6f6b5b8ej85d67452f18300f9@mail.gmail.com>
	<86acburjic.fsf@raman.networkresonance.com>
	<e9132f820603130442hd0b6482i8c5891ba8c02738@mail.gmail.com>
	<86wteypgx7.fsf@raman.networkresonance.com>
Message-ID: <8b2769930603130657v48941acbke679dbe8d8c3cbb1@mail.gmail.com>

> I think we're talking about different things when we say "generic".
> What I mean is to pick a DHT (e.g., Chord, Bamboo) and then
> describe your system in terms of the operations that that
> DHT provides (INSERT, SUCCESSOR, etc.) Then when you've done that
> describe how to implement those messages in SIP.

I'm fine with that, but most (all) of the DHTs I have seen so far have
some drawbacks -- chord is recursive for example, so intermediate
nodes pass twice as many messages as the sender and receiver, which
looks like a bad design and possible DOS issue to me, but perhaps not.

I'm a bit concerned that finding a perfect DHT, and simply lifting it
and standardizing it will gain us little -- all the same effort but
not quite right. I *don't* think we want to design one from scratch,
but if we are going to express it in terms of their operations and
have no room to tweak it at all, it does limit us a bit. Reworking the
paper to seperate the sections is certainly a good idea, however.

--David


>
> > Fortunately, the SIP devices we are targetting have a SIP stack that
> > already handles what we need, so we build on top of it.  If we try to
> > define something more generic, we get into extensive issues of what
> > the minimum requirements are to support it, etc.
>
> Yes, I think that kind of architectural division and clarity is
> a feature, not a bug.
>
> -Ekr
>
>


From dromasca at avaya.com  Mon Mar 13 10:06:30 2006
From: dromasca at avaya.com (Romascanu, Dan (Dan))
Date: Mon, 13 Mar 2006 17:06:30 +0200
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
Message-ID: <AAB4B3D3CF0F454F98272CBE187FDE2F0A2DE35F@is0004avexu1.global.avaya.com>

Yes, but this brings me to the charter proposal and to the fact that
security is very vaguely dealt with in the charter. 

In assumptions we say: 

5. The level of authenticity of identity will vary by the peer
   association group. However, the working group will define at least
   one mechanism that will provide assertion of identity having
   strength equivalent to that provided by the "SIP Identity
   Mechanism" RFC, perhaps by reusing some or all of the mechanism of
   that RFC.

And then, the scope of the work is defined 

While ensuring:

1. Security is addressed in these systems.

It is not clear where the security requirements are defined beyond what
is presented in the assumptions, and in the absence of requirements I am
afraid that we may continue this type of discussions at later phases. 

Regards,

Dan


 
 

> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu 
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David A. Bryan
> Sent: Monday, March 13, 2006 4:51 PM
> To: EKR
> Cc: p2p-sip at cs.columbia.edu
> Subject: Re: [p2p-sip] Review of draft-bryan-sipping-p2p-02
> 
> Just to make sure everyone remembers, this draft isn't really 
> intended to be standardized in any way as it stands anyway 
> (we say that in the intro, and I've mentioned that in all my 
> talks), instead it was proposed as a "hey, here is how we 
> might make this work, this is what we are experimenting with" 
> strawman, and the current iteration is changes as a result of 
> lessons learned in implementation, so I will be the first to 
> agree with Eric that as it stands this ain't even remotely 
> close to something to consider for standardization. It was 
> put out as a strawman to shoot at to come up with something 
> useful, so looks like it is working!
> 
> The security section in particular is really old (most of it 
> from the -00 draft that Cullen and I wrote over a year ago), 
> and hasn't been updated much since we largely have been 
> working with the message passing in our coding and wanted to 
> share some things we thought improved that work. I've 
> actually had some conversations (several with Eric, actually) 
> where using centrally issued certificates (possibly rate 
> limited somehow), and storing fully verifiable certificate 
> chains in the overlay was suggested, and in fact seem the 
> only way to truly secure something like this. We are likely 
> to look toward something like that for the next iteration. I 
> realize only addresses *part* of Eric's concerns anyway, but 
> I just want everyone to be clear about what this draft is 
> (strawman) and what it isn't (something that in anything 
> resembling its current form is being suggested as a standard).
> 
> Heck, I'm hoping we get a new WG which can define better 
> requirements and architecture from many more sources and then 
> we can get some folks working on a better (and far more 
> secure) idea for how to get this done.
> 
> --David
> 
> On 3/13/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> > "David Barrett" <dbarrett at quinthar.com> writes:
> >
> > >> -----Original Message-----
> > >> From: Eric Rescorla [mailto:ekr at networkresonance.com]
> > >>
> > >> "David Barrett" <dbarrett at quinthar.com> writes:
> > >> >
> > >> > Given that a secure DHT is an unsolved problem, and given the 
> > >> > IETF only deals with solved problems, should we:
> > >> >
> > >> > 1) Standardize the best practices today, despite their 
> insecurity.
> > >> >
> > >> >    -- OR --
> > >> >
> > >> > 2) Wait until a secure DHT has been invented and reconvene.
> > >> >
> > >> > I vote for #1.  I certainly don't vote for the unstated #3:
> > >> "Simultaneously
> > >> > invent and standardize a new variant of DHT that is 
> secure (for 
> > >> > some definition of 'secure')."
> > >>
> > >> There's a point below which the security of the system is so bad 
> > >> that it shouldn't be standardized. The question is 
> whether that's 
> > >> the case here.
> > >
> > > That's a fair question, how would you answer it?
> > >
> > > If your answer is "yes, DHTs are so insecure that 
> standardization is 
> > > pointless" then would you advise that we dissolve and 
> reconvene later?
> > >
> > > If your answer is "no, DHTs aren't so insecure so as to make 
> > > standardization pointless" then would you advise that we 
> continue on our current path?
> > >
> > > Or do you have another answer such as "yes, the DHT we're 
> currently 
> > > discussing is so insecure that standardization is pointless, and 
> > > therefore we should <insert alternate plan>."
> >
> > As far as I can tell, the combined system described in this 
> document 
> > is so insecure that it should not be standardized. There 
> may be some 
> > hybrid system (e.g., along the lines of that described by Cox et al 
> > <http://www.pdos.lcs.mit.edu/chord/papers/ddns.pdf>), most 
> likely with 
> > externally verified authentication that would be worth 
> standardizing.
> >
> > -Ekr
> >



From lowekamp at cs.wm.edu  Mon Mar 13 10:21:39 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Mon, 13 Mar 2006 10:21:39 -0500
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <86slpmpfvj.fsf@raman.networkresonance.com>
References: <20060313024122.82E97222418@laser.networkresonance.com>
	<e9132f820603130511j3bcf3d8dg23fa5c752bbba63c@mail.gmail.com>
	<86slpmpfvj.fsf@raman.networkresonance.com>
Message-ID: <e9132f820603130721w6069ce2eqe910e70c118ff359@mail.gmail.com>

On 3/13/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> "Bruce Lowekamp" <lowekamp at cs.wm.edu> writes:
> > On 3/12/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> >> These aren't security problems that can be solved by gluing on
> >> some crypto at the end of the day. Security for DHTs is a
> >> major unsolved research problem, so this is rather worrisome.
> >>
> >>
> >> DETAILED COMMENTS
> >> S 7.2.1:
> >>
> >>    ipport         =  IPV4address ":" port
> >>
> >> And v6?
> >
> > You're absolutely right that there is a problem with this level of
> > security in v6.  First, though, let's address v4.  We've discussed
> > previously that the port number should probably be removed from the
> > hashed portion when generating the NodeID.  Given that, I think we can
> > safely assume that no attacker will have access to a large block of
> > addresses.  At least, I suspect that anyone who would have that much
> > control would have to also be in control of routing (by virtue of
> > either being an ISP or controlling a VLAN we're on), and can attack
> > the protocol in any way they see fit.
>
> No, I don't think that's a safe conclusion.
>
> 1. Many people do have very large blocks of addresses. Xerox,
>    for instance, has a /8.
>

I think the heart of this issue goes to use-cases and architectural
goals.   On the one hand, it's very nice to have a simple protocol
that a few people can set up and use to talk with each other.  On the
other hand, if we want to develop the One True (dht) Ring, then we
have to defend against someone who conceivably could allocate a node
anywhere in a /8 network.   I'd like to see a protocol where the first
case can be done easily, and security extensions are available so that
the One True Ring works reliably.  For most applications I want to
think about right now, my feeling is that a lightweight protocol that
is resistant to attacks that do not assume someone has ISP-level
control over their attacks is sufficient.  I would like to see/work to
develop a protocol that can also be extended to be secure in the face
of more determined attacks.  One interesting thing about attacking a
DHT is that the ease with which attacks can be done depends greatly on
the size of the network.  An open overlay with only a few nodes is
trivially easy to disrupt.  Right now seeding the Identifier is a way
for a private overlay to be protected, but I agree that small open
overlays are a challenge.


> 2. A large NATed network will occupy a large number of
>    nodes in the ring--and legitimately--they'll just be
>    contiguous and densely packed in your design. [0]
>    I haven't thought through the implications of that
>    yet.


Large NATed companies are an interesting challenge.  As you mention,
the issue is that they can disrupt the uniform spacing assumption that
a lot of the analysis makes.  There is some practical limit to the
number of nodes that can be behind one NAT address, but if the number
of nodes is large compared to the total number in the overlay (or,
more accurately, if the number of nodes/number of NATs grows large for
the entire network) there are some issues.  The one I see is that as
you lose uniform spacing of nodes, but presumably not uniform spacing
of resource-IDs, you are going to wind up with some nodes that are
responsible for storing a much larger than normal share of the
registrations because they just happen to be responsible for a large
gap in the ring.  So maybe some support for further distributing
registrations would be appropriate?  I haven't thought about it.  I'm
sure some people would say that as soon as v6 is adopted NATs will
disappear, but I think we need a better answer for this question.


> > The act of inserting a compromised single node gives you an advantage
> > in compromising the routing to the (non-compromised) successor.  Even
> > though they're randomly distributed in IP space, they're neighbors in
> > the overlay, so most other nodes in the network will learn about the
> > successor through the compromised node.  Therefore, by storing
> > replicas at successors, the replica is dependent on the possibly
> > compromised node.  Getting to the successor, even if you know about
> > the compromised node, also requires some amount of linear search.  By
> > keying the replicas and distributing them throughout the DHT, there is
> > no dependency between the N nodes that would be required to compromise
> > the system, so an attacker would have to be able to compromise N
> > indepdendent nodes.  Also, searching for each replica is a logarithmic
> > operation.
>
> Sure, but it's an "all your eggs in one basket" versus "many baskets"
> things. In the design you describe, the chance that at least one
> replica will be compromised by a compromised node is much higher
> than in the design where they're all in the successor set. So,
> you may always get one valid replica, but there's also a high
> chance of inconsistencies. Unless there's some resolution procedure...
>
> In any case, if your goal is to assume that someone gets a
> a large subset of the replica set and can then work things out
> for themselves, there's quite a large literature about how to protect
> routing in the case of some set of compromised nodes. Why do
> you think that this is the optimal technique as opposed to
> that described, by e.g., Wallach?
>

I make no claims of optimality!!!  I'm trying to work on feasibility.

If we use certified NodeIDs, as Wallach does, then I think this is
great.  However, while it helps the One True Ring, it doesn't help the
bunch of friends setting up a network (except for those of us who can
set up a CA for recreational purposes).



I think the one common theme here is the tradeoffs between a protocol
that can be used easily in a small, closed environment versus one that
could be used to run a large, globally-open communications system. 
You're entirely right that this draft doesn't address nearly all of
the security issues for doing the later.  I think the main question
here is whether the ideas in here are the type that can be used to
build a simple protocol that works in the first case, but can have
security extensions such that it also works in the second case.

Bruce


From lowekamp at cs.wm.edu  Mon Mar 13 10:29:58 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Mon, 13 Mar 2006 10:29:58 -0500
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <86wteypgx7.fsf@raman.networkresonance.com>
References: <20060313024122.82E97222418@laser.networkresonance.com>
	<8b2769930603122054x6f6b5b8ej85d67452f18300f9@mail.gmail.com>
	<86acburjic.fsf@raman.networkresonance.com>
	<e9132f820603130442hd0b6482i8c5891ba8c02738@mail.gmail.com>
	<86wteypgx7.fsf@raman.networkresonance.com>
Message-ID: <e9132f820603130729l4d6ce36k7a571ea2e16f2a57@mail.gmail.com>

On 3/13/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> "Bruce Lowekamp" <lowekamp at cs.wm.edu> writes:
>
> > On 3/13/06, Eric Rescorla <ekr at networkresonance.com> wrote:
> >> >> 3. It's really confusing.
> >> >>
> >> >
> >> > I suppose I can't really argue that ;) Suggestions to make it cleaner?
> >> > I sort of hoped that breaking it so things were done with register --
> >> > essentially only adding and querying registrations -- it would be
> >>
> >> Yes. Describe the system as generic operations rather than as
> >> SIP operations and then describe a binding to SIP.
> >
> > As the current draft is organized (and as I'm sure you're aware),
> > section 6 gives an overview of the generic operations, 7  does
> > synteax, and 8 describes specific semantics.  I disbelieve that it
> > would be possible to write something that would be totally generic,
> > then list a short set of SIP bindings.  While I realize you want to
> > find a "generic" DHT, there are a few problems with that.  One is that
> > no "generic" "standard" DHT exists, another is that to define one, we
> > would have to define all kinds of protocol syntax, headers, routing,
> > redirecting, etc. My belief is that if we tried to write something
> > more generic and then said "hey, look, here are some SIP bindings that
> > happen to fit our protocol," someone who is more familiar with another
> > protocol would look at it and say that we failed to describe something
> > necessary (that we simply assume SIP provides, like tags), or we make
> > an assumption that is unique to SIP.
>
> I think we're talking about different things when we say "generic".
> What I mean is to pick a DHT (e.g., Chord, Bamboo) and then
> describe your system in terms of the operations that that
> DHT provides (INSERT, SUCCESSOR, etc.) Then when you've done that
> describe how to implement those messages in SIP.
>

Ah, I see what you're saying now.  I think a lot of the feedback that
this draft received last time was that integrating a specific overlay
algorithm (iterative Chord) was likely a mistake, as hopefully there
would eventually be more widely accepted (hopefully IETF) overlay
algorithms.  For this version of the draft, I tried to extract the
Chord-specific portions of the discussion and put them into a separate
section to make it clear that the specific algorithm isn't inherent to
the concept.  But that does remove the "this is SUCCESSOR" part from
the earlier section.  I guess it would be clearer if earlier there was
a "these are the generic DHT operations that we will implement."   I
still don't want to define the whole thing in terms of Chord,
specifically.

Bruce


From jdrosen at cisco.com  Mon Mar 13 12:50:18 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 12:50:18 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415A15E.5010508@research.panasonic.com>
References: <200603130731.k2D7VnOZ020879@cs.columbia.edu>
	<4415A15E.5010508@research.panasonic.com>
Message-ID: <4415B0DA.1010700@cisco.com>



Eunsoo Shim wrote:

> In principle, working out requirements and looking for a solution 
> without presupposing a solution are all fair. Nevertheless, I think we 
> should pay attention to existing solutions first.
> 
> As shown in our draft (draft-shim-sipping-p2p-arch-00.txt), I envision 
> that P2P SIP is composed of two layers: the P2P overlay layer and SIP. 
> In this architecture, the main difference between the P2P SIP and the 
> client server SIP is how locations of resources are identified. In a 
> sense, what we are touching in the SIP spec for P2P SIP is RFC 3263 as 
> mentioned by Henning, Kundan and others before. I don't see much 
> difference betwee the P2P SIP and the client server SIP regarding NAT 
> traversal.

Right - the main difference should be location of resources. In the case 
of ICE, the problem is how to locate STUN and TURN servers on the p2p 
network. Whats interesting with this is that ANY stun server will do; 
with a relay you'd ideally like one close by. Be interesting to see how 
that can best be done. But once you've got one, regular ICE and SIP 
should work.

-Jonathan R.
-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From dbarrett at quinthar.com  Mon Mar 13 13:06:26 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Mon, 13 Mar 2006 10:06:26 -0800
Subject: [p2p-sip] Review of draft-bryan-sipping-p2p-02
In-Reply-To: <861wx6qvlg.fsf@raman.networkresonance.com>
Message-ID: <200603131806.k2DI6WOZ013705@cs.columbia.edu>

> -----Original Message-----
> From: Eric Rescorla [mailto:ekr at networkresonance.com]
> Subject: Re: [p2p-sip] Review of draft-bryan-sipping-p2p-02
> 
> "David Barrett" <dbarrett at quinthar.com> writes:
> >
> > If your answer is "yes, DHTs are so insecure that standardization is
> > pointless" then would you advise that we dissolve and reconvene later?
> 
> As far as I can tell, the combined system described in this document
> is so insecure that it should not be standardized. There may be some
> hybrid system (e.g., along the lines of that described by Cox et al
> <http://www.pdos.lcs.mit.edu/chord/papers/ddns.pdf>), most likely with
> externally verified authentication that would be worth standardizing.

Ahh, I understand.  Thanks.

-david
 


From jdrosen at cisco.com  Mon Mar 13 13:09:36 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 13:09:36 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4m1q7l$7ob0rv@sj-inbound-e.cisco.com>
References: <4m1q7l$7ob0rv@sj-inbound-e.cisco.com>
Message-ID: <4415B560.6080707@cisco.com>

inline.

David Barrett wrote:

>>-----Original Message-----
>>From: Jonathan Rosenberg [mailto:jdrosen at cisco.com]
>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>>I think the burden of proof would be to demonstrate that ICE cannot and
>>doesn't, for some reason, work for p2p SIP.
> 
> 
> That's fair.  Let's first define what 'works for P2P-SIP' is, then we can
> have a productive discussion on this point.  Specifically, I mean we should
> define the NAT-traversal requirements of P2P-SIP without regard to a
> solution (eg, without presupposing ICE), and then evaluate whether and to
> what degree ICE meets these requirements.

This would make sense if you were desiging a protocol from scratch. But 
we're not. SIP p2p can be thought of as yet another sip extension. We've 
had lots of those. To my knowledge, none of them has gone off and 
decided to write requirements around SIP problems that already have a 
SIP solution. I think the assumption needs to be that existing SIP 
mechanisms are used, and if they break in some way, bring some 
requirements to get them fixed. AFAIK, no one has said that ICE will not 
work with p2p SIP. I think it will work swimmingly well, in fact.




> 
> Clearly, if ICE is the best solution, let's take it.  But if ICE isn't the
> best solution, let's not take it, and instead find what is.  

Again, I disagree. If your beef is with ICE itself, as I suspect it is, 
from this note at least:

David Barrett wrote:
> I rather like the idea of just extending the base SIP protocol to include
> the DHT, rather than rolling an entirely new protocol (*cough* ICE *cough*)
> from scratch, using its own parsers, own state machines, own eccentricities,
> etc.

then I suggest you take these comments to mmusic list for discussion 
there. I will note that state machines are an essential part of nearly 
any protocol definition, whether it be an extension or a new protocol. 
Many sip extensions have them. The same is true for new syntax. I think 
it is naive to think you can do what ICE is doing and end up with 
something that defines no new syntax or state machinery.


> 
> I'm sorry, I didn't mean to present this as something new; I thought this
> was already the consensus.

I'm not sure what the consensus is, but I cannot agree with a plan of 
action that has us revisiting all sorts of sip design decisions that 
have been made already, in the name of a new, non-interoperable p2p-sip.

-Jonathan R.

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From dbarrett at quinthar.com  Mon Mar 13 14:35:45 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Mon, 13 Mar 2006 11:35:45 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415A15E.5010508@research.panasonic.com>
Message-ID: <200603131935.k2DJZsOZ010998@cs.columbia.edu>

> -----Original Message-----
> From: Eunsoo Shim [mailto:eunsoo at research.panasonic.com]
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> In principle, working out requirements and looking for a solution
> without presupposing a solution are all fair. Nevertheless, I think we
> should pay attention to existing solutions first.

Certainly.  But there are many "existing solutions".  Even if we stick
within the bounds of IETF there are at least two serious contenders: ICE and
Teredo (RFC 4380).  And if we look further afield I know of at least two
others (and I'm sure there are more).  But unless we make a serious effort
to evaluate ICE alternatives, we're not making a serious effort to evaluate
ICE alternatives.  So the question is:

"Are we going to make a serious effort to evaluate alternatives to ICE?"

If the answer is no, then let's state as much.  It certainly simplifies our
work.  This seems to be Jonathan's suggestion, and it's an entirely
reasonable approach (though not the one I recommend).

But if the answer is yes, then let's do it.  Let's state what the criteria
for selection are, and take the best option.  If ICE comes out on top,
wonderful, we can defend our choice as not only convenient but optimal.  And
if we can do better than ICE, wonderful, who wouldn't want this?

I'd just like to know if we're even considering anything other than ICE, and
what the grounds for consideration are.  It was my impression that a
discussion on this topic on this list had already occurred (late Jan) and
this plan (defining the NAT-traversal requirements before choosing a
solution) was the result [1]. 

----------------------------------------------------------------------------
[1] https://lists.cs.columbia.edu/pipermail/p2p-sip/2006-January/000366.html

Specifically, the working group will:

...

5. Select firewall/NAT traversal technique(s) for P2P SIP and integrate
    them into the P2P SIP architecture or framework.  If necessary,
    define requirements for enhancements of existing techniques to be
    acted on by other working groups.

----------------------------------------------------------------------------

My contribution to these requirements (if we choose to have them) would be:

- Establishes direct connections with >90% effectiveness in today's internet
	. eg, include symmetric<->symmetric piercing
- Evaluates all routes in parallel and select the fastest
	. vs sequential evaluation
- Discovers the minimum keepalive frequency to maintain UDP NAT mappings
	. vs deferring to the application layer
- Integrates with core SIP syntax (INVITE, REGISTER, etc)
	. vs a standalone protocol with unique grammar

I'm not totally sure how ICE and Teredo stack up against this, nor am I sure
what the final stack of criteria should be.  Thus I'm just not sure ICE is
sufficient because "sufficient" hasn't been defined.

So in order to make headway in this discussion I'd like to know:

a) Are we going to make a serious effort to evaluate alternatives to ICE?
b) What are the criteria whereby we'll make our selection?

-david



From jdrosen at cisco.com  Mon Mar 13 14:52:02 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 14:52:02 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com>
Message-ID: <4415CD62.4020906@cisco.com>

inline.

David Barrett wrote:

>>-----Original Message-----
>>From: Eunsoo Shim [mailto:eunsoo at research.panasonic.com]
>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>>In principle, working out requirements and looking for a solution
>>without presupposing a solution are all fair. Nevertheless, I think we
>>should pay attention to existing solutions first.
> 
> 
> Certainly.  But there are many "existing solutions".  Even if we stick
> within the bounds of IETF there are at least two serious contenders: ICE and
> Teredo (RFC 4380). 

Teredo can be used with ICE, of course. Its even called out in the draft 
as such. It only helps with connecting to other v6 endpoints. YMMV on 
how important that is.

> And if we look further afield I know of at least two
> others (and I'm sure there are more).  But unless we make a serious effort
> to evaluate ICE alternatives, we're not making a serious effort to evaluate
> ICE alternatives.  So the question is:
> 
> "Are we going to make a serious effort to evaluate alternatives to ICE?"
> 
> If the answer is no, then let's state as much.  It certainly simplifies our
> work.  This seems to be Jonathan's suggestion, and it's an entirely
> reasonable approach (though not the one I recommend).

You are right, that is my suggestion.

> 
> But if the answer is yes, then let's do it.  Let's state what the criteria
> for selection are, and take the best option.  If ICE comes out on top,
> wonderful, we can defend our choice as not only convenient but optimal.  And
> if we can do better than ICE, wonderful, who wouldn't want this?

If there is something better than ICE, what makes that specific to p2p?

> 
> I'd just like to know if we're even considering anything other than ICE, and
> what the grounds for consideration are.  It was my impression that a
> discussion on this topic on this list had already occurred (late Jan) and
> this plan (defining the NAT-traversal requirements before choosing a
> solution) was the result [1]. 
> 
> ----------------------------------------------------------------------------
> [1] https://lists.cs.columbia.edu/pipermail/p2p-sip/2006-January/000366.html
> 
> Specifically, the working group will:
> 
> ...
> 
> 5. Select firewall/NAT traversal technique(s) for P2P SIP and integrate
>     them into the P2P SIP architecture or framework.  If necessary,
>     define requirements for enhancements of existing techniques to be
>     acted on by other working groups.
> 
> ----------------------------------------------------------------------------
> 
> My contribution to these requirements (if we choose to have them) would be:
> 
> - Establishes direct connections with >90% effectiveness in today's internet
> 	. eg, include symmetric<->symmetric piercing

I'll agree with that.

> - Evaluates all routes in parallel and select the fastest
> 	. vs sequential evaluation

You can't have your cake and eat it too. ICE originally worked this way 
and it was found to be hugely problematic from a congestion and nat 
overload perspective. That is why it was changed later on.

So, I strongly disagree with this requirement.

I also dont see what it has to do with p2p. Nothing about this 
requirement is unique to p2p sip.


> - Discovers the minimum keepalive frequency to maintain UDP NAT mappings
> 	. vs deferring to the application layer

Ideally I'd agree with you; but this is profoundly hard to do without 
more directly controlled nats.

I also don't see what this has to do with p2p.


> - Integrates with core SIP syntax (INVITE, REGISTER, etc)
> 	. vs a standalone protocol with unique grammar

That seems like a solution proposal and not a requirement. It also has 
nothing to do with p2p.

I find it odd that your suggested requirements miss what I think is the 
only new requirement here - that the solution not require a centralized 
server in order to operate. THAT is actually a reasonable and new 
requirement introduced by p2p.

> 
> I'm not totally sure how ICE and Teredo stack up against this,

Of course you are. Please don't dance around the point. You know very 
well from your other comments how ICE does and doesn't address these 
points. To date, your proposals on investigating SIP p2p nat traversal 
seem to be a thinly veiled attempt to design and/or select alternate nat 
traversal techniques because of some grudge against what has been 
specified already.

As I have suggested before, if you have beefs with the design of ICE, 
take them to mmusic and lets discuss there as part of ICE discussions. 
If you have an alternative proposal, I look forward to the draft. But, 
it is a travesty to the community to use p2p sip as an excuse to 
redesign parts of SIP with new requirements that have nothing to do with 
p2p.

-Jonathan R.
-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From slavitch at gmail.com  Mon Mar 13 15:54:11 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 13 Mar 2006 15:54:11 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415CD62.4020906@cisco.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> <4415CD62.4020906@cisco.com>
Message-ID: <4415DBF3.1090009@gmail.com>

Jonathan Rosenberg wrote:
>
> As I have suggested before, if you have beefs with the design of ICE, 
> take them to mmusic and lets discuss there as part of ICE discussions. 
> If you have an alternative proposal, I look forward to the draft. But, 
> it is a travesty to the community to use p2p sip as an excuse to 
> redesign parts of SIP with new requirements that have nothing to do with 
> p2p.
>
> -Jonathan R.
>   
The result of the process will be a set of requirements to solve the 
use-case needs of P2PSIP.

If it can be */demonstrated /*that ICE can meet these requirements ICE 
will be fine. If not a set of requirements will be sent to MMUSIC.
If MMUSIC cannot act on these requirements or if ICE cannot be seen as 
effectively meeting those requirements something else must be 
considered.  It is the onus of ICE to meet these requirements, not the 
other way around.

I say requirement #1 is:  P2PSIP NAT traversal must be in broad 
interoperable use.
Teredo may be such a beast.  I am not so sure about ICE.

It is the requirements that win, not a draft implementation.  ICE is 
still an Internet-Draft.  It is not yet RFC standard in wide adoption.

It is true that P2PSIP should borrow on existing standards for SIP.  SIP 
has been around for a decade.
ICE is still in draft form, version seven and counting.

When ICE is RFC level and in wide interoperable use the same will be 
said for it. 

Until then it is a draft, and drafts can still be fed requirements or 
suggestions, or rejected outright. 

How many iterations of ICE have we seen. How many YEARS has it taken? 
And it is still in draft form. 
I would respect "use ICE" if the ICE RFC was settled five years ago.  It 
was not.  ICE is still an open question.

Meanwhile,  Skype works right now and is gobbling up share.  If ICE was 
a simple RFC three years ago this would not be the case.

See you in Dallas

Regards

Michael


From dbarrett at quinthar.com  Mon Mar 13 15:59:01 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Mon, 13 Mar 2006 12:59:01 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415CD62.4020906@cisco.com>
Message-ID: <200603132059.k2DKxCOZ002902@cs.columbia.edu>

> -----Original Message-----
> From: Jonathan Rosenberg [mailto:jdrosen at cisco.com]
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> > - Establishes direct connections with >90% effectiveness in today's
> internet
> > 	. eg, include symmetric<->symmetric piercing
> 
> I'll agree with that.

Cool!  I didn't realize ICE supported this yet.


> > - Evaluates all routes in parallel and select the fastest
> > 	. vs sequential evaluation
> 
> You can't have your cake and eat it too. ICE originally worked this way
> and it was found to be hugely problematic from a congestion and nat
> overload perspective. That is why it was changed later on.

Extra cool!  Again, my understanding of ICE was out of date.


> > - Discovers the minimum keepalive frequency to maintain UDP NAT mappings
> > 	. vs deferring to the application layer
> 
> Ideally I'd agree with you; but this is profoundly hard to do without
> more directly controlled nats.

Well, I disagree here.  Just use a parallel "throwaway" connection whose
only purpose is to discover NAT refresh times.  I use this in iGlance and it
works great.  With it I've discovered that NAT refresh times are horribly
erratic, and not at all constant (I've experienced NATs that start at 1s and
grow up to 512s over time).  I believe this is a sufficiently advanced
problem that it should be handled by the NAT solution itself, rather than
left up to the application developer.


> I also don't see what this has to do with p2p.

If P2P is defined purely as DHT, then I'd agree.  But it's my understanding
our mandate takes a broader view, including the integration of NAT
penetration best practices.


> > - Integrates with core SIP syntax (INVITE, REGISTER, etc)
> > 	. vs a standalone protocol with unique grammar
> 
> That seems like a solution proposal and not a requirement. It also has
> nothing to do with p2p.

Well, yes, I'd suggest we not abstract NAT from SIP, just like I'd prefer we
not abstract DHT from SIP (another topic of ongoing debate).  I don't
dispute we *can* (ICE being a great example of the result when you do), but
I personally think SIP can and should be extended to accomplish both.  

As such, I'm offering this as a requirement.  I don't claim that it *is* a
requirement (because the list doesn't yet exist, and certainly hasn't been
formalized), but I'm happy to argue for its inclusion.


> I find it odd that your suggested requirements miss what I think is the
> only new requirement here - that the solution not require a centralized
> server in order to operate. THAT is actually a reasonable and new
> requirement introduced by p2p.

Again, I accept as reasonable the view that the *only* topic open for debate
is how to decentralize the SIP proxy.  Personally, I take a broader view
(and one that I believe is consistent with our mandate).


> > I'm not totally sure how ICE and Teredo stack up against this,
> 
> Of course you are. Please don't dance around the point. You know very
> well from your other comments how ICE does and doesn't address these
> points. To date, your proposals on investigating SIP p2p nat traversal
> seem to be a thinly veiled attempt to design and/or select alternate nat
> traversal techniques because of some grudge against what has been
> specified already.
> 
> As I have suggested before, if you have beefs with the design of ICE,
> take them to mmusic and lets discuss there as part of ICE discussions.
> If you have an alternative proposal, I look forward to the draft. But,
> it is a travesty to the community to use p2p sip as an excuse to
> redesign parts of SIP with new requirements that have nothing to do with
> p2p.

Well, given that I was wrong on at least two points about ICE, I can say
with certainty that I didn't know how they'd stack up.  ICE has undergone
(and, I think, is undergoing) rapid change, and it's hard to keep up.

As for my ulterior motive, I think it's not so thinly veiled: I believe
there are superior options to ICE and I'd like us to investigate them.
You're correct, this has been my explicit goal from the start.  So far as I
can tell, it's within our mandate to consider alternatives to ICE, and I'm
just trying to push along that path.

That's a fair suggestion that I take my concerns to the ICE group, but given
that I fundamentally disagree that ICE is needed at all (my point is that
SIP should handle it implicitly), this seems like a futile gesture.

And I appreciate your request for a drafted alternative.  If the P2P-SIP
group lists the criteria for acceptance and commits to a fair evaluation
based on those criteria, that's just what I (and I'm sure others) will do.
 
Finally, let's keep this focused on the merits and combinations of ICE, SIP,
P2P, DHTs, etc.  This isn't about me.


Anyway, ultimately the P2P-SIP group will recommend best practices and
standards for deploying SIP in internet-wide, (mostly?) serverless
scenarios.  One component of this is NAT penetration.  One proposal is to
just say "we recommend ICE" and be done with it.  My proposal is to define
what criteria we'll use to evaluate NAT penetration proposals, and then
judge the options accordingly.

Which proposal do you (ie, everyone else) prefer?

-david


From jdrosen at cisco.com  Mon Mar 13 16:18:27 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 16:18:27 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415DBF3.1090009@gmail.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com>
	<4415CD62.4020906@cisco.com> <4415DBF3.1090009@gmail.com>
Message-ID: <4415E1A3.7010502@cisco.com>



Michael Slavitch wrote:

> Jonathan Rosenberg wrote:
> 
>>
>> As I have suggested before, if you have beefs with the design of ICE, 
>> take them to mmusic and lets discuss there as part of ICE discussions. 
>> If you have an alternative proposal, I look forward to the draft. But, 
>> it is a travesty to the community to use p2p sip as an excuse to 
>> redesign parts of SIP with new requirements that have nothing to do 
>> with p2p.
>>
>> -Jonathan R.
>>   
> 
> The result of the process will be a set of requirements to solve the 
> use-case needs of P2PSIP.
> 
> If it can be */demonstrated /*that ICE can meet these requirements ICE 
> will be fine. If not a set of requirements will be sent to MMUSIC.
> If MMUSIC cannot act on these requirements or if ICE cannot be seen as 
> effectively meeting those requirements something else must be 
> considered.  It is the onus of ICE to meet these requirements, not the 
> other way around.

I'm afraid I have to disagree with that.

It is not the job of p2p sip to redefine the requirements for all of the 
SIP extensions that our out there, so that you can choose to retain or 
discard various ones as you see fit. If this were the case, "p2p sip" is 
a poor name for this group. It sounds like some folks are looking for 
'p2p voip', and you'd like to define a protocol that is not 
interoperable with sip to do that. Well, that is one possible charter, 
but not the one I'd hope for.

> 
> I say requirement #1 is:  P2PSIP NAT traversal must be in broad 
> interoperable use.
> Teredo may be such a beast.  I am not so sure about ICE.

Well, we can debate all of that. ICE is actually in use quite a bit in 
various softphones and voip deployments in service providers today. 
Teredo is for v6, as you know. As I said, a v6 only solution for nat 
traversal is .... interesting. YMMV as to whether that meets the needs 
of the community.

In any case, all of this is really just sounding like the 'we hate 
certain sip extensions and plan on doing our own' WG. Now, I can point 
to other SIP documents in various stages of deployment. So, I guess, by 
your metric, p2p sip shall be able to discard anything it likes that has 
not reached some magic deployment threshold? Absurd.


> 
> It is the requirements that win, not a draft implementation.  ICE is 
> still an Internet-Draft.  It is not yet RFC standard in wide adoption.

Its not an RFC, no. WIDE usage? Depends on your definition. Implemented? 
Absolutely. Deployed? Yes.

I'll note I found this criteria (technique must be in widespread 
deployment) laughable. p2p SIP itself is not even yet a working group. 
We've got threads going on debating fundamental issues like, should the 
DHT be done within SIP or through a layer separate? And can it be 
secured? p2p SIP is itself only just getting started. And you are going 
to throw away a fundamental piece of SIP architecture because its not 
done yet? Now that is ironic.


> 
> It is true that P2PSIP should borrow on existing standards for SIP.  SIP 
> has been around for a decade.
> ICE is still in draft form, version seven and counting.
> 
> When ICE is RFC level and in wide interoperable use the same will be 
> said for it.
> Until then it is a draft, and drafts can still be fed requirements or 
> suggestions, or rejected outright.

Sure it can be changed or improved; we've been doing that for a while. 
This is a hard problem as you know, with many constraints from different 
communities.

But rejected outright? Well, its a WG document, it has broad support 
from the WG as such. There are no plans for its rejection that I know of.


> How many iterations of ICE have we seen. How many YEARS has it taken? 
> And it is still in draft form. I would respect "use ICE" if the ICE RFC 
> was settled five years ago.  It was not.  ICE is still an open question.
> 
> Meanwhile,  Skype works right now and is gobbling up share.  If ICE was 
> a simple RFC three years ago this would not be the case.

Ah. So, by your metric (wide deployment), our only possible solution for 
p2p SIP, nat traversal and all, would be Skype?

-Jonathan R.

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From jdrosen at cisco.com  Mon Mar 13 16:41:28 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 16:41:28 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4m6448$72jfsj@sj-inbound-c.cisco.com>
References: <4m6448$72jfsj@sj-inbound-c.cisco.com>
Message-ID: <4415E708.6050106@cisco.com>



David Barrett wrote:

>>-----Original Message-----
>>From: Jonathan Rosenberg [mailto:jdrosen at cisco.com]
>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>>
>>>- Establishes direct connections with >90% effectiveness in today's
>>
>>internet
>>
>>>	. eg, include symmetric<->symmetric piercing
>>
>>I'll agree with that.
> 
> 
> Cool!  I didn't realize ICE supported this yet.

It does it with relays, of course. Thats what turn is for.

> 
> 
> 
>>>- Evaluates all routes in parallel and select the fastest
>>>	. vs sequential evaluation
>>
>>You can't have your cake and eat it too. ICE originally worked this way
>>and it was found to be hugely problematic from a congestion and nat
>>overload perspective. That is why it was changed later on.
> 
> 
> Extra cool!  Again, my understanding of ICE was out of date.

To be clear - ICE use a sequential mode. The parallel mode was 
previously in ICE and was discarded. Your requirement was for paralell. 
I am saying that I disagree with parallel because of the problems that 
were found. This had extensive discussion in mmusic.

> 
> 
> 
>>>- Discovers the minimum keepalive frequency to maintain UDP NAT mappings
>>>	. vs deferring to the application layer
>>
>>Ideally I'd agree with you; but this is profoundly hard to do without
>>more directly controlled nats.
> 
> 
> Well, I disagree here.  Just use a parallel "throwaway" connection whose
> only purpose is to discover NAT refresh times.

These are unreliable, as they change as the NAT becomes loaded, for 
example. See the behave specs for discussions on this. STUN used to 
attempt to do this timing discovery (its in rfc3489), but that whole 
nat-discovery and black-box measurement thing has been thrown away.

   I use this in iGlance and it
> works great.  With it I've discovered that NAT refresh times are horribly
> erratic, and not at all constant (I've experienced NATs that start at 1s and
> grow up to 512s over time).

Exactly. Thats why its a bad idea to try and discover it. Its going to 
change.

> I believe this is a sufficiently advanced
> problem that it should be handled by the NAT solution itself, rather than
> left up to the application developer.

The solution is to use a minimum value that works with high reliability.

> 
> 
> 
>>I also don't see what this has to do with p2p.
> 
> 
> If P2P is defined purely as DHT, then I'd agree.  But it's my understanding
> our mandate takes a broader view, including the integration of NAT
> penetration best practices.

Well, the purpose of a bof is to debate the charter. I am proposing that 
the objective of p2p sip is not a free pass to change anything in sip it 
  likes.

> 
> 
> 
>>>- Integrates with core SIP syntax (INVITE, REGISTER, etc)
>>>	. vs a standalone protocol with unique grammar
>>
>>That seems like a solution proposal and not a requirement. It also has
>>nothing to do with p2p.
> 
> 
> Well, yes, I'd suggest we not abstract NAT from SIP, just like I'd prefer we
> not abstract DHT from SIP (another topic of ongoing debate).  I don't
> dispute we *can* (ICE being a great example of the result when you do), but
> I personally think SIP can and should be extended to accomplish both.  

SIP has been extended to deal with nat traversal. That extension is ICE 
for media and sip-outbound for signaling.

> 
> As such, I'm offering this as a requirement.  I don't claim that it *is* a
> requirement (because the list doesn't yet exist, and certainly hasn't been
> formalized), but I'm happy to argue for its inclusion.
> 
> 
> 
>>I find it odd that your suggested requirements miss what I think is the
>>only new requirement here - that the solution not require a centralized
>>server in order to operate. THAT is actually a reasonable and new
>>requirement introduced by p2p.
> 
> 
> Again, I accept as reasonable the view that the *only* topic open for debate
> is how to decentralize the SIP proxy.  Personally, I take a broader view
> (and one that I believe is consistent with our mandate).

Well, like I said, the charter is up for discussion still.

You have not yet commented on the interoperability issue. I take it 
then, you have no interest in interoperability with regular sip endpoints?



> 
> 
> 
>>>I'm not totally sure how ICE and Teredo stack up against this,
>>
>>Of course you are. Please don't dance around the point. You know very
>>well from your other comments how ICE does and doesn't address these
>>points. To date, your proposals on investigating SIP p2p nat traversal
>>seem to be a thinly veiled attempt to design and/or select alternate nat
>>traversal techniques because of some grudge against what has been
>>specified already.
>>
>>As I have suggested before, if you have beefs with the design of ICE,
>>take them to mmusic and lets discuss there as part of ICE discussions.
>>If you have an alternative proposal, I look forward to the draft. But,
>>it is a travesty to the community to use p2p sip as an excuse to
>>redesign parts of SIP with new requirements that have nothing to do with
>>p2p.
> 
> 
> Well, given that I was wrong on at least two points about ICE, I can say
> with certainty that I didn't know how they'd stack up.  ICE has undergone
> (and, I think, is undergoing) rapid change, and it's hard to keep up.
> 
> As for my ulterior motive, I think it's not so thinly veiled: I believe
> there are superior options to ICE and I'd like us to investigate them.

Hmm, I couldn't find your draft in the archives... must've missed it.


> You're correct, this has been my explicit goal from the start.  So far as I
> can tell, it's within our mandate to consider alternatives to ICE, and I'm
> just trying to push along that path.

I disagree, and I thank you for making it clear that you believe it is 
in the charter to reinvent whatever pieces of sip you like, including 
nat traversal, even if what we have works for p2p - just because you 
don't like them. No doubt I'll have comments on that part of the charter 
text.


> 
> That's a fair suggestion that I take my concerns to the ICE group, but given
> that I fundamentally disagree that ICE is needed at all (my point is that
> SIP should handle it implicitly), this seems like a futile gesture.


So, if I get this correctly, you are saying that, you disagree with 
ICE's design, but instead of taking that to the group where its being 
developed, you would rather start a new group to do something in 
parallel - something different - using p2p as an excuse.

Help me understand how that is in the best interests of the IETF 
community or in the best interests of interoperability.



> 
> And I appreciate your request for a drafted alternative.  If the P2P-SIP
> group lists the criteria for acceptance and commits to a fair evaluation
> based on those criteria, that's just what I (and I'm sure others) will do.

Rather than hiding behind p2p, if you've got beefs with SIPs nat 
traversal soltuions, bring em up and let the community working on them 
have a chance to review and comment.



>  
> Finally, let's keep this focused on the merits and combinations of ICE, SIP,
> P2P, DHTs, etc.  This isn't about me.

The merits of ICE belong in the mmusic list, where that community can 
provide comment.

-Jonathan R.

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From slavitch at gmail.com  Mon Mar 13 16:42:08 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 13 Mar 2006 16:42:08 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415E1A3.7010502@cisco.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> <4415CD62.4020906@cisco.com>
	<4415DBF3.1090009@gmail.com> <4415E1A3.7010502@cisco.com>
Message-ID: <1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>

On 3/13/06, Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>
>
>
> Well, we can debate all of that. ICE is actually in use quite a bit in
> various softphones and voip deployments in service providers today.
> Teredo is for v6, as you know. As I said, a v6 only solution for nat
> traversal is .... interesting. YMMV as to whether that meets the needs
> of the community.


I know, we use an implementation of ICE in our softphones.  It works if you
have a systems
engineer handy.

However our softphones don't work with Vendor B softphones or service
provider C.
Vendor D doesn't know we exist, and Vendor E claims that they support ICE
but lie, and
there is no way to debunk that without extensive testing, which doesn't
scale.   Provider
F won't bother until his vendor supplies ICE and his vendor is waiting for
an RFC but\
pushing proprietary phones.

Without an RFC, an interoperability test,  and comparison code it's not an
open standard,
it's a recommendation.

If you want a bunch of interoperable silos, ICE is great.  Maintains the
CPE/Carrier silos fine.
That's great if there was no alternative, but there is. Skype just works.
My mother uses it.
Her ancient relatives in Ukraine use it.   No systems engineers at all.
Their IT departments
consist of twelve years olds, no NOC involved, and the equipment needed is
available at
the nearby shop.   Click, call, done.

I agree with you, there are implementations of NAT traversal that read on
ICE.  Your point?
When I can get a state diagram that can be fed to coders and to testers
without handholding
it's a protocol. Until then it's a recommendation.


In any case, all of this is really just sounding like the 'we hate
> certain sip extensions and plan on doing our own' WG. Now, I can point
> to other SIP documents in various stages of deployment. So, I guess, by
> your metric, p2p sip shall be able to discard anything it likes that has
> not reached some magic deployment threshold? Absurd.
>

I don't hate ICE.  I want ICE to work interoperably, simply and
effectively.  I want vovida code.
If that is true then there is no problem.   If a spec goes  to 100 vendors
and the result is 100
implementations that interoperate ICE is great.  I've wanted ICE for a long
time.
I'm still waiting.  Meanwhile ICE exists, no waiting.


>
> > It is the requirements that win, not a draft implementation.  ICE is
> > still an Internet-Draft.  It is not yet RFC standard in wide adoption.



Sure it can be changed or improved; we've been doing that for a while.
> This is a hard problem as you know, with many constraints from different
> communities.



Skype did it several years ago. It's done.  It borrowed from file sharing.
Done.

But rejected outright? Well, its a WG document, it has broad support
> from the WG as such. There are no plans for its rejection that I know of.


It's not rejection by the IETF that matters.  It's rejection by the market
at large.

  |  Ah. So, by your metric (wide deployment), our only possible solution
for

> p2p SIP, nat traversal and all, would be Skype?


Not just that.   Of course not.  But it works and has done so for  a while.
No systems engineering, no TURN setup, no STUN setup.  It just works.

The #1 requirement of P2PSIP: It just works.  No excuses.

If ICE "just works"  we're in total agreement and we have no issue.  Let's
find out.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060313/16c18d14/attachment.html

From slavitch at gmail.com  Mon Mar 13 16:44:24 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 13 Mar 2006 16:44:24 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> <4415CD62.4020906@cisco.com>
	<4415DBF3.1090009@gmail.com> <4415E1A3.7010502@cisco.com>
	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
Message-ID: <1fc32c8f0603131344w7141020ct56434dc92d3c0a44@mail.gmail.com>

One sentance above should state:


"Meanwhile ICE exists, no waiting."

Should be:

"Meanwhile Skype exists, no waiting."


That is all,

Michael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060313/d7dfb59b/attachment-0001.html

From eunsoo at research.panasonic.com  Mon Mar 13 16:53:32 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Mon, 13 Mar 2006 16:53:32 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415E1A3.7010502@cisco.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> 
	<4415CD62.4020906@cisco.com> <4415DBF3.1090009@gmail.com> 
	<4415E1A3.7010502@cisco.com>
Message-ID: <4415E9DC.2090808@research.panasonic.com>

Jonathan Rosenberg wrote:

>Michael Slavitch wrote:
>
>  
>
>>Jonathan Rosenberg wrote:
>>
>>    
>>
>>>As I have suggested before, if you have beefs with the design of ICE, 
>>>take them to mmusic and lets discuss there as part of ICE discussions. 
>>>If you have an alternative proposal, I look forward to the draft. But, 
>>>it is a travesty to the community to use p2p sip as an excuse to 
>>>redesign parts of SIP with new requirements that have nothing to do 
>>>with p2p.
>>>
>>>-Jonathan R.
>>>  
>>>      
>>>
>>The result of the process will be a set of requirements to solve the 
>>use-case needs of P2PSIP.
>>
>>If it can be */demonstrated /*that ICE can meet these requirements ICE 
>>will be fine. If not a set of requirements will be sent to MMUSIC.
>>If MMUSIC cannot act on these requirements or if ICE cannot be seen as 
>>effectively meeting those requirements something else must be 
>>considered.  It is the onus of ICE to meet these requirements, not the 
>>other way around.
>>    
>>
>
>I'm afraid I have to disagree with that.
>
>It is not the job of p2p sip to redefine the requirements for all of the 
>SIP extensions that our out there, so that you can choose to retain or 
>discard various ones as you see fit. 
>
I agree with Jonathan. Reviewing the requirements of various SIP 
extensions MUST be performed in WGs like SIP, SIPPING, SIMPLE and so on 
according to each WG's charter.

>If this were the case, "p2p sip" is 
>a poor name for this group. It sounds like some folks are looking for 
>'p2p voip', and you'd like to define a protocol that is not 
>interoperable with sip to do that. Well, that is one possible charter, 
>but not the one I'd hope for.
>
>  
>
To repeat myself, I strongly believe that interoperability with the 
client server SIP is important for P2P SIP.
We are not trying to reinvent SIP. We should not.
As far as I remember, the separation of SIP and the overlay layer was 
supported widely in the P2P SIP ad-hoc meeting in Vancouver.
One of the main advantages of the separation was "no change in SIP 
almost or at all".

>>I say requirement #1 is:  P2PSIP NAT traversal must be in broad 
>>interoperable use.
>>Teredo may be such a beast.  I am not so sure about ICE.
>>    
>>
>
>Well, we can debate all of that. ICE is actually in use quite a bit in 
>various softphones and voip deployments in service providers today. 
>Teredo is for v6, as you know. As I said, a v6 only solution for nat 
>traversal is .... interesting. YMMV as to whether that meets the needs 
>of the community.
>
>In any case, all of this is really just sounding like the 'we hate 
>certain sip extensions and plan on doing our own' WG. Now, I can point 
>to other SIP documents in various stages of deployment. So, I guess, by 
>your metric, p2p sip shall be able to discard anything it likes that has 
>not reached some magic deployment threshold? Absurd.
>
>
>  
>
>>It is the requirements that win, not a draft implementation.  ICE is 
>>still an Internet-Draft.  It is not yet RFC standard in wide adoption.
>>    
>>
>
>Its not an RFC, no. WIDE usage? Depends on your definition. Implemented? 
>Absolutely. Deployed? Yes.
>
>I'll note I found this criteria (technique must be in widespread 
>deployment) laughable. p2p SIP itself is not even yet a working group. 
>We've got threads going on debating fundamental issues like, should the 
>DHT be done within SIP or through a layer separate? And can it be 
>secured? p2p SIP is itself only just getting started. And you are going 
>to throw away a fundamental piece of SIP architecture because its not 
>done yet? Now that is ironic.
>
>
>  
>
>>It is true that P2PSIP should borrow on existing standards for SIP.  SIP 
>>has been around for a decade.
>>ICE is still in draft form, version seven and counting.
>>
>>When ICE is RFC level and in wide interoperable use the same will be 
>>said for it.
>>Until then it is a draft, and drafts can still be fed requirements or 
>>suggestions, or rejected outright.
>>    
>>
>
>Sure it can be changed or improved; we've been doing that for a while. 
>This is a hard problem as you know, with many constraints from different 
>communities.
>
>But rejected outright? Well, its a WG document, it has broad support 
>from the WG as such. There are no plans for its rejection that I know of.
>
>
>  
>
>>How many iterations of ICE have we seen. How many YEARS has it taken? 
>>And it is still in draft form. I would respect "use ICE" if the ICE RFC 
>>was settled five years ago.  It was not.  ICE is still an open question.
>>
>>Meanwhile,  Skype works right now and is gobbling up share.  If ICE was 
>>a simple RFC three years ago this would not be the case.
>>    
>>
>
>Ah. So, by your metric (wide deployment), our only possible solution for 
>p2p SIP, nat traversal and all, would be Skype?
>
>  
>

Thanks.

Eunsoo

From jdrosen at cisco.com  Mon Mar 13 16:58:09 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 16:58:09 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com>
	<4415CD62.4020906@cisco.com>	 <4415DBF3.1090009@gmail.com>
	<4415E1A3.7010502@cisco.com>
	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
Message-ID: <4415EAF1.6050308@cisco.com>

Interop is easy when the protocol and all implementations come from a 
single company, as they do with Skype.

Interop is harder when you want to introduce standards so that multiple 
vendors and multiple providers can interoperate. Thats what IETF does. 
Thats why we are here. Its a difficult process that takes longer, 
because it requires multiple participants with differing interests and 
perspectives, and broader requirements covering more diverse needs.

If you want p2p-sip to 'just work', I suggest that time is better spent 
figuring out how to eliminate the need to configure the servers that ICE 
depends on - turn and stun servers - and figure out how to dynamically 
find ones amongst peers in the mesh. When you do that, ICE will just 
work and do so in a p2p environment. I think that is a better service to 
the community than rants about how ICE is taking so long along with bold 
claims about how you can do something better, faster, and with better 
interoperability results.

-Jonathan R.


Michael Slavitch wrote:

> On 3/13/06, *Jonathan Rosenberg* <jdrosen at cisco.com 
> <mailto:jdrosen at cisco.com>> wrote:
> 
> 
> 
>     Well, we can debate all of that. ICE is actually in use quite a bit in
>     various softphones and voip deployments in service providers today.
>     Teredo is for v6, as you know. As I said, a v6 only solution for nat
>     traversal is .... interesting. YMMV as to whether that meets the needs
>     of the community.
> 
> 
> I know, we use an implementation of ICE in our softphones.  It works if 
> you have a systems
> engineer handy.
> 
> However our softphones don't work with Vendor B softphones or service 
> provider C.
> Vendor D doesn't know we exist, and Vendor E claims that they support 
> ICE but lie, and
> there is no way to debunk that without extensive testing, which doesn't 
> scale.   Provider
> F won't bother until his vendor supplies ICE and his vendor is waiting 
> for an RFC but\
> pushing proprietary phones. 
> 
> Without an RFC, an interoperability test,  and comparison code it's not 
> an open standard,
> it's a recommendation.
> 
> If you want a bunch of interoperable silos, ICE is great.  Maintains the 
> CPE/Carrier silos fine.
> That's great if there was no alternative, but there is. Skype just 
> works.  My mother uses it. 
> Her ancient relatives in Ukraine use it.   No systems engineers at all. 
> Their IT departments
> consist of twelve years olds, no NOC involved, and the equipment needed 
> is available at
> the nearby shop.   Click, call, done. 
> 
> I agree with you, there are implementations of NAT traversal that read 
> on ICE.  Your point?
> When I can get a state diagram that can be fed to coders and to testers 
> without handholding
> it's a protocol. Until then it's a recommendation. 
> 
> 
>     In any case, all of this is really just sounding like the 'we hate
>     certain sip extensions and plan on doing our own' WG. Now, I can point
>     to other SIP documents in various stages of deployment. So, I guess, by
>     your metric, p2p sip shall be able to discard anything it likes that has
>     not reached some magic deployment threshold? Absurd.
> 
> 
> I don't hate ICE.  I want ICE to work interoperably, simply and 
> effectively.  I want vovida code.
> If that is true then there is no problem.   If a spec goes  to 100 
> vendors and the result is 100
> implementations that interoperate ICE is great.  I've wanted ICE for a 
> long time.
> I'm still waiting.  Meanwhile ICE exists, no waiting.
> 
> 
>      >
>      > It is the requirements that win, not a draft implementation.  ICE is
>      > still an Internet-Draft.  It is not yet RFC standard in wide
>     adoption.
> 
> 
> 
>     Sure it can be changed or improved; we've been doing that for a while.
>     This is a hard problem as you know, with many constraints from different
>     communities.
> 
> 
> 
> Skype did it several years ago. It's done.  It borrowed from file 
> sharing. Done.
> 
>     But rejected outright? Well, its a WG document, it has broad support
>     from the WG as such. There are no plans for its rejection that I
>     know of. 
> 
> 
> It's not rejection by the IETF that matters.  It's rejection by the 
> market at large. 
> 
>   |  Ah. So, by your metric (wide deployment), our only possible 
> solution for
> 
>     p2p SIP, nat traversal and all, would be Skype?
> 
> 
> Not just that.   Of course not.  But it works and has done so for  a while.
> No systems engineering, no TURN setup, no STUN setup.  It just works.
> 
> The #1 requirement of P2PSIP: It just works.  No excuses. 
> 
> If ICE "just works"  we're in total agreement and we have no issue.  
> Let's find out. 
> 
> 
> 

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From dean.willis at softarmor.com  Mon Mar 13 16:58:52 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Mon, 13 Mar 2006 15:58:52 -0600
Subject: [p2p-sip] Draft Agenda for P2P-SIP at IETF 65 posted to IETF site
In-Reply-To: <7D131673-D71A-4D97-90B7-B747E3A354CE@greycouncil.com>
References: <7D131673-D71A-4D97-90B7-B747E3A354CE@greycouncil.com>
Message-ID: <7A2C1BD6-1A8C-4196-B000-4BA51FD9FDE7@softarmor.com>


see:

http://www3.ietf.org/proceedings/06mar/agenda/p2p-sip.html


--
Dean



From salman at cs.columbia.edu  Mon Mar 13 17:05:09 2006
From: salman at cs.columbia.edu (Salman Abdul Baset)
Date: Mon, 13 Mar 2006 17:05:09 -0500 (EST)
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603131344w7141020ct56434dc92d3c0a44@mail.gmail.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> <4415CD62.4020906@cisco.com>
	<4415DBF3.1090009@gmail.com> <4415E1A3.7010502@cisco.com>
	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
	<1fc32c8f0603131344w7141020ct56434dc92d3c0a44@mail.gmail.com>
Message-ID: <Pine.GSO.4.58.0603131648050.1548@dynasty.cs.columbia.edu>

Skype works because they are not constrained by any interop issues. And
there is possibly no magic which can make a protocol interoperable.
Developers are dictated by deadlines and they make mistakes. I am sure the
Skype developers must have made mistakes. Their protocol suit has been
regularly updated to correct those mistakes (check for stack version
number in skype system call dumps)

Also, the core protocol developer community is very small in Skype as
compared to SIP/STUN/ICE which certainly has hundreds and possibly a few
thousand developers worldwide. They all have different needs and
requirements vs possibly a few requirements for Skype.

While a good majority of the academic world is busy in churning papers after
papers about making an algorithm 'purely' distributed, Skype combines the
benefits of distributed and central paradigms in a very decent way.

An interesting question for me is: how many internet-drafts/RFCs does a
developer needs to read to implement a Skype like network? From some of
the posts on this list, it appears that people want to have one document
to read and to be able to implement a Skype like system. After reading
a RFC/draft, they find it difficult to decide what to implement and what to
leave.

Regards,
Salman


On Mon, 13 Mar 2006, Michael Slavitch wrote:

> One sentance above should state:
>
>
> "Meanwhile ICE exists, no waiting."
>
> Should be:
>
> "Meanwhile Skype exists, no waiting."
>
>
> That is all,
>
> Michael
>

From ag at ag-projects.com  Mon Mar 13 17:06:15 2006
From: ag at ag-projects.com (Adrian Georgescu)
Date: Mon, 13 Mar 2006 14:06:15 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415EAF1.6050308@cisco.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com>
	<4415CD62.4020906@cisco.com>	 <4415DBF3.1090009@gmail.com>
	<4415E1A3.7010502@cisco.com>
	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
	<4415EAF1.6050308@cisco.com>
Message-ID: <0FFD87D0-AC1D-4D2A-A74F-BB7C2B05C565@ag-projects.com>

Jonathan is absolutely right, P2P SIP should only concentrate on how  
to allocate and find resources. The way the signaling and nat  
traversal work have already been sorted out, this to-be WG should  
just find a way to put the pieces together to achieve scalability.

Adrian

On Mar 13, 2006, at 1:58 PM, Jonathan Rosenberg wrote:

> Interop is easy when the protocol and all implementations come from a
> single company, as they do with Skype.
>
> Interop is harder when you want to introduce standards so that  
> multiple
> vendors and multiple providers can interoperate. Thats what IETF does.
> Thats why we are here. Its a difficult process that takes longer,
> because it requires multiple participants with differing interests and
> perspectives, and broader requirements covering more diverse needs.
>
> If you want p2p-sip to 'just work', I suggest that time is better  
> spent
> figuring out how to eliminate the need to configure the servers  
> that ICE
> depends on - turn and stun servers - and figure out how to dynamically
> find ones amongst peers in the mesh. When you do that, ICE will just
> work and do so in a p2p environment. I think that is a better  
> service to
> the community than rants about how ICE is taking so long along with  
> bold
> claims about how you can do something better, faster, and with better
> interoperability results.
>
> -Jonathan R.
>
>
> Michael Slavitch wrote:
>
>> On 3/13/06, *Jonathan Rosenberg* <jdrosen at cisco.com
>> <mailto:jdrosen at cisco.com>> wrote:
>>
>>
>>
>>     Well, we can debate all of that. ICE is actually in use quite  
>> a bit in
>>     various softphones and voip deployments in service providers  
>> today.
>>     Teredo is for v6, as you know. As I said, a v6 only solution  
>> for nat
>>     traversal is .... interesting. YMMV as to whether that meets  
>> the needs
>>     of the community.
>>
>>
>> I know, we use an implementation of ICE in our softphones.  It  
>> works if
>> you have a systems
>> engineer handy.
>>
>> However our softphones don't work with Vendor B softphones or service
>> provider C.
>> Vendor D doesn't know we exist, and Vendor E claims that they support
>> ICE but lie, and
>> there is no way to debunk that without extensive testing, which  
>> doesn't
>> scale.   Provider
>> F won't bother until his vendor supplies ICE and his vendor is  
>> waiting
>> for an RFC but\
>> pushing proprietary phones.
>>
>> Without an RFC, an interoperability test,  and comparison code  
>> it's not
>> an open standard,
>> it's a recommendation.
>>
>> If you want a bunch of interoperable silos, ICE is great.   
>> Maintains the
>> CPE/Carrier silos fine.
>> That's great if there was no alternative, but there is. Skype just
>> works.  My mother uses it.
>> Her ancient relatives in Ukraine use it.   No systems engineers at  
>> all.
>> Their IT departments
>> consist of twelve years olds, no NOC involved, and the equipment  
>> needed
>> is available at
>> the nearby shop.   Click, call, done.
>>
>> I agree with you, there are implementations of NAT traversal that  
>> read
>> on ICE.  Your point?
>> When I can get a state diagram that can be fed to coders and to  
>> testers
>> without handholding
>> it's a protocol. Until then it's a recommendation.
>>
>>
>>     In any case, all of this is really just sounding like the 'we  
>> hate
>>     certain sip extensions and plan on doing our own' WG. Now, I  
>> can point
>>     to other SIP documents in various stages of deployment. So, I  
>> guess, by
>>     your metric, p2p sip shall be able to discard anything it  
>> likes that has
>>     not reached some magic deployment threshold? Absurd.
>>
>>
>> I don't hate ICE.  I want ICE to work interoperably, simply and
>> effectively.  I want vovida code.
>> If that is true then there is no problem.   If a spec goes  to 100
>> vendors and the result is 100
>> implementations that interoperate ICE is great.  I've wanted ICE  
>> for a
>> long time.
>> I'm still waiting.  Meanwhile ICE exists, no waiting.
>>
>>
>>>
>>> It is the requirements that win, not a draft implementation.  ICE is
>>> still an Internet-Draft.  It is not yet RFC standard in wide
>>     adoption.
>>
>>
>>
>>     Sure it can be changed or improved; we've been doing that for  
>> a while.
>>     This is a hard problem as you know, with many constraints from  
>> different
>>     communities.
>>
>>
>>
>> Skype did it several years ago. It's done.  It borrowed from file
>> sharing. Done.
>>
>>     But rejected outright? Well, its a WG document, it has broad  
>> support
>>     from the WG as such. There are no plans for its rejection that I
>>     know of.
>>
>>
>> It's not rejection by the IETF that matters.  It's rejection by the
>> market at large.
>>
>>   |  Ah. So, by your metric (wide deployment), our only possible
>> solution for
>>
>>     p2p SIP, nat traversal and all, would be Skype?
>>
>>
>> Not just that.   Of course not.  But it works and has done so for   
>> a while.
>> No systems engineering, no TURN setup, no STUN setup.  It just works.
>>
>> The #1 requirement of P2PSIP: It just works.  No excuses.
>>
>> If ICE "just works"  we're in total agreement and we have no issue.
>> Let's find out.
>>
>>
>>
>
> -- 
> Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
> Cisco Fellow                                   Parsippany, NJ  
> 07054-2711
> Cisco Systems
> jdrosen at cisco.com                              FAX:   (973) 952-5050
> http://www.jdrosen.net                         PHONE: (973) 952-5000
> http://www.cisco.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From slavitch at gmail.com  Mon Mar 13 17:09:25 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 13 Mar 2006 17:09:25 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415E9DC.2090808@research.panasonic.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> <4415CD62.4020906@cisco.com>
	<4415DBF3.1090009@gmail.com> <4415E1A3.7010502@cisco.com>
	<4415E9DC.2090808@research.panasonic.com>
Message-ID: <1fc32c8f0603131409l1c9133a2x61e7b818e3e89ee9@mail.gmail.com>

'To repeat myself, I strongly believe that interoperability with the client
server SIP is important for P2P SIP.
We are not trying to reinvent SIP. We should not. As far as I remember, the
separation of SIP and the overlay layer was
supported widely in the P2P SIP ad-hoc meeting in Vancouver. One of the main
advantages of the separation was "no change in SIP
almost or at all".'

I fully agree with this.   The separation of SIP and the overlay is what
makes SIP reuse possible.   The biggest change I foresee in this
model involves the overloading of names used to identify nodes, agents and
addresses.   In other words,  "words" and "names".

SIP syntax could allow:  DNS name, IPV4, IPV6,  or "P2PNAME".  There are
lots of people working on smart ways to do 'P2PNAME'.

In our experience if SIP/RTP/SDP allow p2pnames to overload addresses in
descriptions  of entities (much like DNS, but possibly not DNS, perhaps
something else)  then those protocols can continue merrily along.  No more
changes to SIP or SDP needed.  Minor changes that don't break
interoperability.

Given the nature of these protocols this many not require any changes to the
existing state machines.
It only adds one additional descriptor to the valid set of possible
descriptors.

Changes involving syntax are not major, and they are backward compatible.
Changes involving state behavior are major and break things.

In code-speak it becomes a recompile with another accepted data-type, and
some testing, using existing test suites.

It becomes easy.

If  the SIP agenst on top of the overlay uses ICE, great.

If ICE meets the requirements for the underlying overlay, better.

Let's find out.

Regards

Michael





--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060313/b37198cd/attachment-0001.html

From sathya at research.panasonic.com  Mon Mar 13 17:09:27 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Mon, 13 Mar 2006 17:09:27 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <200603132059.k2DKxCOZ002902@cs.columbia.edu>
References: <200603132059.k2DKxCOZ002902@cs.columbia.edu>
Message-ID: <4415ED97.4080002@Research.Panasonic.COM>

David -

> <snip>
>
>As for my ulterior motive, I think it's not so thinly veiled: I believe
>there are superior options to ICE and I'd like us to investigate them.
>You're correct, this has been my explicit goal from the start.  So far as I
>can tell, it's within our mandate to consider alternatives to ICE, and I'm
>just trying to push along that path.
>  
>
I don't understand how you came to the conclusion that "it's within our
mandate to consider alternatives to ICE". The relevant sentence in the
charter reads,
"

Select firewall/NAT traversal technique(s) for P2P SIP and integrate
   them into the P2P SIP architecture or framework.  If necessary,
   define requirements for enhancements of existing techniques to be
   acted on by other working groups.

"

Please note "If necessary" - the necessity MUST be demonstrated before
suggesting new requirements and it MUST be acted on by other
(appropriate) working groups.
We had a debate on this earlier and agreed to the sentence as it is
written now.

thanks,
Sathya

>That's a fair suggestion that I take my concerns to the ICE group, but given
>that I fundamentally disagree that ICE is needed at all (my point is that
>SIP should handle it implicitly), this seems like a futile gesture.
>
>And I appreciate your request for a drafted alternative.  If the P2P-SIP
>group lists the criteria for acceptance and commits to a fair evaluation
>based on those criteria, that's just what I (and I'm sure others) will do.
> 
>Finally, let's keep this focused on the merits and combinations of ICE, SIP,
>P2P, DHTs, etc.  This isn't about me.
>
>
>Anyway, ultimately the P2P-SIP group will recommend best practices and
>standards for deploying SIP in internet-wide, (mostly?) serverless
>scenarios.  One component of this is NAT penetration.  One proposal is to
>just say "we recommend ICE" and be done with it.  My proposal is to define
>what criteria we'll use to evaluate NAT penetration proposals, and then
>judge the options accordingly.
>
>Which proposal do you (ie, everyone else) prefer?
>
>-david
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From eunsoo at research.panasonic.com  Mon Mar 13 17:15:34 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Mon, 13 Mar 2006 17:15:34 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> 
	<4415CD62.4020906@cisco.com> <4415DBF3.1090009@gmail.com> 
	<4415E1A3.7010502@cisco.com> 
	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
Message-ID: <4415EF06.1080301@research.panasonic.com>

I think any open standard protocol has an issue of interoperability 
because there will be various implementations.
A simple mistake of a developer is enough to cause a problem in 
interoperability.
The interoperability issue cannot be cited as a weakness in comparison 
with Skype's software since it is an implementation by just one company.
Whatever we design in P2P SIP WG will have the same interoperability issue.
Thanks.

Eunsoo

Michael Slavitch wrote:

> On 3/13/06, *Jonathan Rosenberg* <jdrosen at cisco.com 
> <mailto:jdrosen at cisco.com>> wrote:
>
>
>
>     Well, we can debate all of that. ICE is actually in use quite a bit in
>     various softphones and voip deployments in service providers today.
>     Teredo is for v6, as you know. As I said, a v6 only solution for nat
>     traversal is .... interesting. YMMV as to whether that meets the needs
>     of the community.
>
>
> I know, we use an implementation of ICE in our softphones.  It works 
> if you have a systems
> engineer handy.
>
> However our softphones don't work with Vendor B softphones or service 
> provider C.
> Vendor D doesn't know we exist, and Vendor E claims that they support 
> ICE but lie, and
> there is no way to debunk that without extensive testing, which 
> doesn't scale.   Provider
> F won't bother until his vendor supplies ICE and his vendor is waiting 
> for an RFC but\
> pushing proprietary phones. 
>
> Without an RFC, an interoperability test,  and comparison code it's 
> not an open standard,
> it's a recommendation.
>
> If you want a bunch of interoperable silos, ICE is great.  Maintains 
> the CPE/Carrier silos fine.
> That's great if there was no alternative, but there is. Skype just 
> works.  My mother uses it. 
> Her ancient relatives in Ukraine use it.   No systems engineers at 
> all. Their IT departments
> consist of twelve years olds, no NOC involved, and the equipment 
> needed is available at
> the nearby shop.   Click, call, done. 
>
> I agree with you, there are implementations of NAT traversal that read 
> on ICE.  Your point?
> When I can get a state diagram that can be fed to coders and to 
> testers without handholding
> it's a protocol. Until then it's a recommendation. 
>
>
>     In any case, all of this is really just sounding like the 'we hate
>     certain sip extensions and plan on doing our own' WG. Now, I can
>     point
>     to other SIP documents in various stages of deployment. So, I
>     guess, by
>     your metric, p2p sip shall be able to discard anything it likes
>     that has
>     not reached some magic deployment threshold? Absurd.
>
>
> I don't hate ICE.  I want ICE to work interoperably, simply and 
> effectively.  I want vovida code.
> If that is true then there is no problem.   If a spec goes  to 100 
> vendors and the result is 100
> implementations that interoperate ICE is great.  I've wanted ICE for a 
> long time.
> I'm still waiting.  Meanwhile ICE exists, no waiting.
>
>
>     >
>     > It is the requirements that win, not a draft
>     implementation.  ICE is
>     > still an Internet-Draft.  It is not yet RFC standard in wide
>     adoption.
>
>
>
>     Sure it can be changed or improved; we've been doing that for a while.
>     This is a hard problem as you know, with many constraints from
>     different
>     communities.
>
>
>
> Skype did it several years ago. It's done.  It borrowed from file 
> sharing. Done.
>
>     But rejected outright? Well, its a WG document, it has broad support
>     from the WG as such. There are no plans for its rejection that I
>     know of. 
>
>
> It's not rejection by the IETF that matters.  It's rejection by the 
> market at large. 
>
>   |  Ah. So, by your metric (wide deployment), our only possible 
> solution for
>
>     p2p SIP, nat traversal and all, would be Skype?
>
>
> Not just that.   Of course not.  But it works and has done so for  a 
> while.
> No systems engineering, no TURN setup, no STUN setup.  It just works.
>
> The #1 requirement of P2PSIP: It just works.  No excuses. 
>
> If ICE "just works"  we're in total agreement and we have no issue.  
> Let's find out. 
>
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From slavitch at gmail.com  Mon Mar 13 17:30:55 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 13 Mar 2006 17:30:55 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415EF06.1080301@research.panasonic.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> <4415CD62.4020906@cisco.com>
	<4415DBF3.1090009@gmail.com> <4415E1A3.7010502@cisco.com>
	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
	<4415EF06.1080301@research.panasonic.com>
Message-ID: <1fc32c8f0603131430k7e6f870cl30d28383084606bd@mail.gmail.com>

"The interoperability issue cannot be cited as a weakness in comparison"

Interoperability becomes a non-issue if there is a reference design to draw
from,
as well as an interoperability test plan, and a clear design.

If a spec is too complex to be implemented in software the problem may not
be the software.

"with Skype's software since it is an implementation by just one company.
Whatever we design in P2P SIP WG will have the same interoperability issue."

Which is why demonstratable code always wins.  I agree with you.

Skype's software is software.   Code always wins.

It's always easier to build a spec from code than code from a spec.

An internet-draft  should have code alongside to test the design.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060313/a6e5b9e2/attachment.html

From huang-ming.pan at comcast.net  Mon Mar 13 17:32:06 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Mon, 13 Mar 2006 14:32:06 -0800
Subject: [p2p-sip] p2p-sip Digest, Vol 8, Issue 19
References: <mailman.30.1142269227.14849.p2p-sip@cs.columbia.edu>
Message-ID: <002b01c646ed$f5f6e630$a10010ac@pi2000>

Hi,

A few stupid questions came to me these days.

Q1: Will a open-source based p2p-sip network be secure?

Most open-source p2p file sharing network seem not secure. Sometimes files
downloaded for days are found to be incomplete or contain contents not as
titled. Anyone can share malicious craps.

Skype says itself as secure. It is proprietary and close-source. Nobody can
join the network without Skype assistance.

Q2: When networks are implemented on p2p-sip standards but individually
operated and not open source, are they better (to end users) than those in
proprietary protocols?

As for most well-known commercial voip networks nowaday, serious hackers
should easily know how to bombard (transaction or call-stateful) server
infrastructure of their networks where sip msgs seem not encrypted.

If one of primary purposes of p2p-sip standardization is to formalize the
protocols of one single p2p sip "universe' where all nodes are based on
p2p-sip standards and made from an open-source reference implementation so
that anyone can contribute node products to the universe, will it be secure
if supernodes or STUN/TURN servers can be contributed irresponsibly or
maliciously?

If a p2p-sip universe is never in the plan and eventually all service
providers will operate their own networks (or certificate authorities), is a
close-source proprietary approach like Skype's (coupled with protocol
bridges among networks if interoperability is necessary) be faster to
achieve than first spending years on p2p-sip standardization and finally
find derived networks to be insecure (w/ a known security association) or
not fully interoperable (w/ with different overlays and private security
schemes)?

Don't hit me...I'm on your side.

Peter

----- Original Message -----
From: <p2p-sip-request at cs.columbia.edu>
To: <p2p-sip at cs.columbia.edu>
Sent: Monday, March 13, 2006 9:00 AM
Subject: p2p-sip Digest, Vol 8, Issue 19


> Send p2p-sip mailing list submissions to
> p2p-sip at cs.columbia.edu
>
> To subscribe or unsubscribe via the World Wide Web, visit
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> or, via email, send a message with subject or body 'help' to
> p2p-sip-request at cs.columbia.edu
>
> You can reach the person managing the list at
> p2p-sip-owner at cs.columbia.edu
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of p2p-sip digest..."
>


----------------------------------------------------------------------------
----


> Today's Topics:
>
>    1. Re: Review of draft-bryan-sipping-p2p-02 (Bruce Lowekamp)
>


----------------------------------------------------------------------------
----


> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From dean.willis at softarmor.com  Mon Mar 13 17:33:11 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Mon, 13 Mar 2006 16:33:11 -0600
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415ED97.4080002@Research.Panasonic.COM>
References: <200603132059.k2DKxCOZ002902@cs.columbia.edu>
	<4415ED97.4080002@Research.Panasonic.COM>
Message-ID: <ECC0741B-DCC2-4E4E-AB85-7E3268E6AADF@softarmor.com>



On Mar 13, 2006, at 4:09 PM, Sathya Narayanan wrote:

> David -
>
>> <snip>
>>
>> As for my ulterior motive, I think it's not so thinly veiled: I  
>> believe
>> there are superior options to ICE and I'd like us to investigate  
>> them.
>> You're correct, this has been my explicit goal from the start.  So  
>> far as I
>> can tell, it's within our mandate to consider alternatives to ICE,  
>> and I'm
>> just trying to push along that path.
>>
>>
> I don't understand how you came to the conclusion that "it's within  
> our
> mandate to consider alternatives to ICE". The relevant sentence in the
> charter reads,
> "
>
> Select firewall/NAT traversal technique(s) for P2P SIP and integrate
>    them into the P2P SIP architecture or framework.  If necessary,
>    define requirements for enhancements of existing techniques to be
>    acted on by other working groups.
>
> "
>
> Please note "If necessary" - the necessity MUST be demonstrated before
> suggesting new requirements and it MUST be acted on by other
> (appropriate) working groups.
> We had a debate on this earlier and agreed to the sentence as it is
> written now.
>

I'm thinking I agree with Sathya here. IETF-wise, we may have a  
difficult time chartering work that overlaps with MMUSIC's work.

--
Dean

From RADHIKA.R.ROY at saic.com  Mon Mar 13 17:33:26 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Mon, 13 Mar 2006 17:33:26 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31D9DB9@0591-ITS-EXMP02.us.saic.com>

Mike:

 

OK, the question is this: Where do you like see this work should be done?
SIP/SIPPING or P2P-SIP?

 

Radhika

 

  _____  

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Michael Slavitch
Sent: Monday, March 13, 2006 5:09 PM
To: Eunsoo Shim
Cc: P2P-SIP
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal

 

'To repeat myself, I strongly believe that interoperability with the client
server SIP is important for P2P SIP.
We are not trying to reinvent SIP. We should not. As far as I remember, the
separation of SIP and the overlay layer was 
supported widely in the P2P SIP ad-hoc meeting in Vancouver. One of the main
advantages of the separation was "no change in SIP
almost or at all".'

I fully agree with this.   The separation of SIP and the overlay is what
makes SIP reuse possible.   The biggest change I foresee in this 
model involves the overloading of names used to identify nodes, agents and
addresses.   In other words,  "words" and "names".

SIP syntax could allow:  DNS name, IPV4, IPV6,  or "P2PNAME".  There are
lots of people working on smart ways to do 'P2PNAME'. 

In our experience if SIP/RTP/SDP allow p2pnames to overload addresses in
descriptions  of entities (much like DNS, but possibly not DNS, perhaps
something else)  then those protocols can continue merrily along.  No more
changes to SIP or SDP needed.  Minor changes that don't break
interoperability. 

Given the nature of these protocols this many not require any changes to the
existing state machines.  
It only adds one additional descriptor to the valid set of possible
descriptors. 

Changes involving syntax are not major, and they are backward compatible.
Changes involving state behavior are major and break things.

In code-speak it becomes a recompile with another accepted data-type, and
some testing, using existing test suites.

It becomes easy.

If  the SIP agenst on top of the overlay uses ICE, great.   

If ICE meets the requirements for the underlying overlay, better. 

Let's find out.

Regards

Michael





-- 
Michael Slavitch
Ottawa, Ontario Canada 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060313/49a04955/attachment-0001.html

From jdrosen at cisco.com  Mon Mar 13 17:41:33 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Mon, 13 Mar 2006 17:41:33 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415ED97.4080002@Research.Panasonic.COM>
References: <200603132059.k2DKxCOZ002902@cs.columbia.edu>
	<4415ED97.4080002@Research.Panasonic.COM>
Message-ID: <4415F51D.5010205@cisco.com>

So, we clearly still have some disconnects on the interpretation of the 
charter text here, since my read was the same as Sathya's. Let me suggest:

5. The group shall determine how to apply the existing SIP NAT traversal 
techniques (ICE/STUN/TURN for media and SIP-outbound for signaling) in a 
p2p environment. It is anticipated that this work will primarily involve 
discovery of STUN and TURN server resources in the p2p mesh. The 
resulting mechanism must be interoperable with non-p2p clients. If the 
group discovers that the existing techniques do not work, it shall 
develop requirements into the relevant working groups (MMUSIC, BEHAVE, 
SIP as needed) for changes needed to make them work.


-Jonathan R.

Sathya Narayanan wrote:

> David -
> 
> 
>><snip>
>>
>>As for my ulterior motive, I think it's not so thinly veiled: I believe
>>there are superior options to ICE and I'd like us to investigate them.
>>You're correct, this has been my explicit goal from the start.  So far as I
>>can tell, it's within our mandate to consider alternatives to ICE, and I'm
>>just trying to push along that path.
>> 
>>
> 
> I don't understand how you came to the conclusion that "it's within our
> mandate to consider alternatives to ICE". The relevant sentence in the
> charter reads,
> "
> 
> Select firewall/NAT traversal technique(s) for P2P SIP and integrate
>    them into the P2P SIP architecture or framework.  If necessary,
>    define requirements for enhancements of existing techniques to be
>    acted on by other working groups.
> 
> "
> 
> Please note "If necessary" - the necessity MUST be demonstrated before
> suggesting new requirements and it MUST be acted on by other
> (appropriate) working groups.
> We had a debate on this earlier and agreed to the sentence as it is
> written now.
> 
> thanks,
> Sathya
> 
> 
>>That's a fair suggestion that I take my concerns to the ICE group, but given
>>that I fundamentally disagree that ICE is needed at all (my point is that
>>SIP should handle it implicitly), this seems like a futile gesture.
>>
>>And I appreciate your request for a drafted alternative.  If the P2P-SIP
>>group lists the criteria for acceptance and commits to a fair evaluation
>>based on those criteria, that's just what I (and I'm sure others) will do.
>>
>>Finally, let's keep this focused on the merits and combinations of ICE, SIP,
>>P2P, DHTs, etc.  This isn't about me.
>>
>>
>>Anyway, ultimately the P2P-SIP group will recommend best practices and
>>standards for deploying SIP in internet-wide, (mostly?) serverless
>>scenarios.  One component of this is NAT penetration.  One proposal is to
>>just say "we recommend ICE" and be done with it.  My proposal is to define
>>what criteria we'll use to evaluate NAT penetration proposals, and then
>>judge the options accordingly.
>>
>>Which proposal do you (ie, everyone else) prefer?
>>
>>-david
>>
>>_______________________________________________
>>p2p-sip mailing list
>>p2p-sip at cs.columbia.edu
>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>> 
>>
> 
> 

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From huang-ming.pan at comcast.net  Mon Mar 13 17:57:56 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Mon, 13 Mar 2006 14:57:56 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com><4415CD62.4020906@cisco.com>	
	<4415DBF3.1090009@gmail.com><4415E1A3.7010502@cisco.com><1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
	<4415EAF1.6050308@cisco.com>
Message-ID: <007401c646f1$919a50b0$a10010ac@pi2000>

I forgot to ask another few stupid questions:

Q3: By "interop" does it mean "p2ply" or bridged?

Q4: Is it possible to "mesh" peers in a Chord overlay interop with peers in
a Kademlia?

Thanks.

Peter

----- Original Message -----
From: "Jonathan Rosenberg" <jdrosen at cisco.com>
To: "Michael Slavitch" <slavitch at gmail.com>
Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
Sent: Monday, March 13, 2006 1:58 PM
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal


> Interop is easy when the protocol and all implementations come from a
> single company, as they do with Skype.
>
> Interop is harder when you want to introduce standards so that multiple
> vendors and multiple providers can interoperate. Thats what IETF does.
> Thats why we are here. Its a difficult process that takes longer,
> because it requires multiple participants with differing interests and
> perspectives, and broader requirements covering more diverse needs.
>
> If you want p2p-sip to 'just work', I suggest that time is better spent
> figuring out how to eliminate the need to configure the servers that ICE
> depends on - turn and stun servers - and figure out how to dynamically
> find ones amongst peers in the mesh. When you do that, ICE will just
> work and do so in a p2p environment. I think that is a better service to
> the community than rants about how ICE is taking so long along with bold
> claims about how you can do something better, faster, and with better
> interoperability results.
>
> -Jonathan R.
>
>
> Michael Slavitch wrote:
>
> > On 3/13/06, *Jonathan Rosenberg* <jdrosen at cisco.com
> > <mailto:jdrosen at cisco.com>> wrote:
> >
> >
> >
> >     Well, we can debate all of that. ICE is actually in use quite a bit
in
> >     various softphones and voip deployments in service providers today.
> >     Teredo is for v6, as you know. As I said, a v6 only solution for nat
> >     traversal is .... interesting. YMMV as to whether that meets the
needs
> >     of the community.
> >
> >
> > I know, we use an implementation of ICE in our softphones.  It works if
> > you have a systems
> > engineer handy.
> >
> > However our softphones don't work with Vendor B softphones or service
> > provider C.
> > Vendor D doesn't know we exist, and Vendor E claims that they support
> > ICE but lie, and
> > there is no way to debunk that without extensive testing, which doesn't
> > scale.   Provider
> > F won't bother until his vendor supplies ICE and his vendor is waiting
> > for an RFC but\
> > pushing proprietary phones.
> >
> > Without an RFC, an interoperability test,  and comparison code it's not
> > an open standard,
> > it's a recommendation.
> >
> > If you want a bunch of interoperable silos, ICE is great.  Maintains the
> > CPE/Carrier silos fine.
> > That's great if there was no alternative, but there is. Skype just
> > works.  My mother uses it.
> > Her ancient relatives in Ukraine use it.   No systems engineers at all.
> > Their IT departments
> > consist of twelve years olds, no NOC involved, and the equipment needed
> > is available at
> > the nearby shop.   Click, call, done.
> >
> > I agree with you, there are implementations of NAT traversal that read
> > on ICE.  Your point?
> > When I can get a state diagram that can be fed to coders and to testers
> > without handholding
> > it's a protocol. Until then it's a recommendation.
> >
> >
> >     In any case, all of this is really just sounding like the 'we hate
> >     certain sip extensions and plan on doing our own' WG. Now, I can
point
> >     to other SIP documents in various stages of deployment. So, I guess,
by
> >     your metric, p2p sip shall be able to discard anything it likes that
has
> >     not reached some magic deployment threshold? Absurd.
> >
> >
> > I don't hate ICE.  I want ICE to work interoperably, simply and
> > effectively.  I want vovida code.
> > If that is true then there is no problem.   If a spec goes  to 100
> > vendors and the result is 100
> > implementations that interoperate ICE is great.  I've wanted ICE for a
> > long time.
> > I'm still waiting.  Meanwhile ICE exists, no waiting.
> >
> >
> >      >
> >      > It is the requirements that win, not a draft implementation.  ICE
is
> >      > still an Internet-Draft.  It is not yet RFC standard in wide
> >     adoption.
> >
> >
> >
> >     Sure it can be changed or improved; we've been doing that for a
while.
> >     This is a hard problem as you know, with many constraints from
different
> >     communities.
> >
> >
> >
> > Skype did it several years ago. It's done.  It borrowed from file
> > sharing. Done.
> >
> >     But rejected outright? Well, its a WG document, it has broad support
> >     from the WG as such. There are no plans for its rejection that I
> >     know of.
> >
> >
> > It's not rejection by the IETF that matters.  It's rejection by the
> > market at large.
> >
> >   |  Ah. So, by your metric (wide deployment), our only possible
> > solution for
> >
> >     p2p SIP, nat traversal and all, would be Skype?
> >
> >
> > Not just that.   Of course not.  But it works and has done so for  a
while.
> > No systems engineering, no TURN setup, no STUN setup.  It just works.
> >
> > The #1 requirement of P2PSIP: It just works.  No excuses.
> >
> > If ICE "just works"  we're in total agreement and we have no issue.
> > Let's find out.
> >
> >
> >
>
> --
> Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
> Cisco Fellow                                   Parsippany, NJ 07054-2711
> Cisco Systems
> jdrosen at cisco.com                              FAX:   (973) 952-5050
> http://www.jdrosen.net                         PHONE: (973) 952-5000
> http://www.cisco.com
>
>


From eunsoo at research.panasonic.com  Mon Mar 13 18:10:49 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Mon, 13 Mar 2006 18:10:49 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <200603132058.AEB41689@redwood.research.panasonic.com>
References: <200603132058.AEB41689@redwood.research.panasonic.com>
Message-ID: <4415FBF9.7010301@research.panasonic.com>

I appreciate the importance of NAT traversal for P2P SIP and respect all 
the discussions about what ICE can, cannot, should or should not do.
Whatever P2P SIP is, if the software does not do NAT traversal very 
well, it won't be adopted for a large scale use over the Internet.

But this group is not the right community to talk about all the aspects 
of ICE since ICE is a work at MMUSIC.
As all understand, we cannot simply start redesigning things other WGs 
have been working on.

However, if P2P SIP brings about any new requirement for NAT traversal 
or for ICE (,STUN,TURN, sip-outbound, etc.), I think it is appropriate 
to talk about it with the purpose of submitting the requirement to the 
appropriate WG. Certainly such requirement is not something like "we 
need a standard right now" or "a particular solution does not work 
regardless of P2P SIP."

If we see major problems with things under the charters of other WGs and 
we need fix for the problems even though the problems are not specific 
to P2P (SIP), then we should join the appropriate WGs and work out 
solutions there. I would be happy to join such efforts.

Thanks.

Eunsoo


David Barrett wrote:

>>-----Original Message-----
>>From: Jonathan Rosenberg [mailto:jdrosen at cisco.com]
>>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>>
>>    
>>
>>>- Establishes direct connections with >90% effectiveness in today's
>>>      
>>>
>>internet
>>    
>>
>>>	. eg, include symmetric<->symmetric piercing
>>>      
>>>
>>I'll agree with that.
>>    
>>
>
>Cool!  I didn't realize ICE supported this yet.
>
>
>  
>
>>>- Evaluates all routes in parallel and select the fastest
>>>	. vs sequential evaluation
>>>      
>>>
>>You can't have your cake and eat it too. ICE originally worked this way
>>and it was found to be hugely problematic from a congestion and nat
>>overload perspective. That is why it was changed later on.
>>    
>>
>
>Extra cool!  Again, my understanding of ICE was out of date.
>
>
>  
>
>>>- Discovers the minimum keepalive frequency to maintain UDP NAT mappings
>>>	. vs deferring to the application layer
>>>      
>>>
>>Ideally I'd agree with you; but this is profoundly hard to do without
>>more directly controlled nats.
>>    
>>
>
>Well, I disagree here.  Just use a parallel "throwaway" connection whose
>only purpose is to discover NAT refresh times.  I use this in iGlance and it
>works great.  With it I've discovered that NAT refresh times are horribly
>erratic, and not at all constant (I've experienced NATs that start at 1s and
>grow up to 512s over time).  I believe this is a sufficiently advanced
>problem that it should be handled by the NAT solution itself, rather than
>left up to the application developer.
>
>
>  
>
>>I also don't see what this has to do with p2p.
>>    
>>
>
>If P2P is defined purely as DHT, then I'd agree.  But it's my understanding
>our mandate takes a broader view, including the integration of NAT
>penetration best practices.
>
>
>  
>
>>>- Integrates with core SIP syntax (INVITE, REGISTER, etc)
>>>	. vs a standalone protocol with unique grammar
>>>      
>>>
>>That seems like a solution proposal and not a requirement. It also has
>>nothing to do with p2p.
>>    
>>
>
>Well, yes, I'd suggest we not abstract NAT from SIP, just like I'd prefer we
>not abstract DHT from SIP (another topic of ongoing debate).  I don't
>dispute we *can* (ICE being a great example of the result when you do), but
>I personally think SIP can and should be extended to accomplish both.  
>
>As such, I'm offering this as a requirement.  I don't claim that it *is* a
>requirement (because the list doesn't yet exist, and certainly hasn't been
>formalized), but I'm happy to argue for its inclusion.
>
>
>  
>
>>I find it odd that your suggested requirements miss what I think is the
>>only new requirement here - that the solution not require a centralized
>>server in order to operate. THAT is actually a reasonable and new
>>requirement introduced by p2p.
>>    
>>
>
>Again, I accept as reasonable the view that the *only* topic open for debate
>is how to decentralize the SIP proxy.  Personally, I take a broader view
>(and one that I believe is consistent with our mandate).
>
>
>  
>
>>>I'm not totally sure how ICE and Teredo stack up against this,
>>>      
>>>
>>Of course you are. Please don't dance around the point. You know very
>>well from your other comments how ICE does and doesn't address these
>>points. To date, your proposals on investigating SIP p2p nat traversal
>>seem to be a thinly veiled attempt to design and/or select alternate nat
>>traversal techniques because of some grudge against what has been
>>specified already.
>>
>>As I have suggested before, if you have beefs with the design of ICE,
>>take them to mmusic and lets discuss there as part of ICE discussions.
>>If you have an alternative proposal, I look forward to the draft. But,
>>it is a travesty to the community to use p2p sip as an excuse to
>>redesign parts of SIP with new requirements that have nothing to do with
>>p2p.
>>    
>>
>
>Well, given that I was wrong on at least two points about ICE, I can say
>with certainty that I didn't know how they'd stack up.  ICE has undergone
>(and, I think, is undergoing) rapid change, and it's hard to keep up.
>
>As for my ulterior motive, I think it's not so thinly veiled: I believe
>there are superior options to ICE and I'd like us to investigate them.
>You're correct, this has been my explicit goal from the start.  So far as I
>can tell, it's within our mandate to consider alternatives to ICE, and I'm
>just trying to push along that path.
>
>That's a fair suggestion that I take my concerns to the ICE group, but given
>that I fundamentally disagree that ICE is needed at all (my point is that
>SIP should handle it implicitly), this seems like a futile gesture.
>
>And I appreciate your request for a drafted alternative.  If the P2P-SIP
>group lists the criteria for acceptance and commits to a fair evaluation
>based on those criteria, that's just what I (and I'm sure others) will do.
> 
>Finally, let's keep this focused on the merits and combinations of ICE, SIP,
>P2P, DHTs, etc.  This isn't about me.
>
>
>Anyway, ultimately the P2P-SIP group will recommend best practices and
>standards for deploying SIP in internet-wide, (mostly?) serverless
>scenarios.  One component of this is NAT penetration.  One proposal is to
>just say "we recommend ICE" and be done with it.  My proposal is to define
>what criteria we'll use to evaluate NAT penetration proposals, and then
>judge the options accordingly.
>
>Which proposal do you (ie, everyone else) prefer?
>
>-david
>
>  
>


From eunsoo at research.panasonic.com  Mon Mar 13 18:27:24 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Mon, 13 Mar 2006 18:27:24 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603131430k7e6f870cl30d28383084606bd@mail.gmail.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com> 
	<4415CD62.4020906@cisco.com> <4415DBF3.1090009@gmail.com> 
	<4415E1A3.7010502@cisco.com> 
	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com> 
	<4415EF06.1080301@research.panasonic.com> 
	<1fc32c8f0603131430k7e6f870cl30d28383084606bd@mail.gmail.com>
Message-ID: <4415FFDC.3080709@research.panasonic.com>

Michael Slavitch wrote:

> "The interoperability issue cannot be cited as a weakness in comparison"
>
> Interoperability becomes a non-issue if there is a reference design to 
> draw from,
> as well as an interoperability test plan, and a clear design. 
>
> If a spec is too complex to be implemented in software the problem may 
> not be the software.
>
> "with Skype's software since it is an implementation by just one company.
> Whatever we design in P2P SIP WG will have the same interoperability 
> issue. "
>
> Which is why demonstratable code always wins.  I agree with you.
>
> Skype's software is software.   Code always wins.
>
> It's always easier to build a spec from code than code from a spec. 
>
> An internet-draft  should have code alongside to test the design.
>
>
I appreciate your statement about the importance of running code.
"Rough consensus and running code" has been one of the most important 
operation principles of IETF in my understanding.
We will hear the speech about it by David Clark during the IETF social 
event in Dallas.
Actually I feel more urgency for running code of P2P SIP than as a 
matter of principle.
It is great that this group has many folks who are eager to implement 
the (future) P2P SIP spec and see running code.
I am one of them.
I hope we have active discussions about implementation while we work out 
all the details of the spec in parallel.
Thanks.

Eunsoo


From dbarrett at quinthar.com  Mon Mar 13 19:31:25 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Mon, 13 Mar 2006 16:31:25 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415F51D.5010205@cisco.com>
Message-ID: <200603140031.k2E0VXOZ022281@cs.columbia.edu>

> -----Original Message-----
> From: Jonathan Rosenberg [mailto:jdrosen at cisco.com]
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> So, we clearly still have some disconnects on the interpretation of the
> charter text here, since my read was the same as Sathya's. Let me suggest:
> 
> 5. The group shall determine how to apply the existing SIP NAT traversal
> techniques (ICE/STUN/TURN for media and SIP-outbound for signaling) in a
> p2p environment. It is anticipated that this work will primarily involve
> discovery of STUN and TURN server resources in the p2p mesh. The
> resulting mechanism must be interoperable with non-p2p clients. If the
> group discovers that the existing techniques do not work, it shall
> develop requirements into the relevant working groups (MMUSIC, BEHAVE,
> SIP as needed) for changes needed to make them work.

Excellent clarification.  If that's what the group wants, then I agree with
this wording.

However, the first line of the original wording was "Select firewall/NAT
traversal technique(s) for P2P SIP" -- I presumed the word 'select' implied
there would be a selection of some kind from multiple options, with ICE
being one of them.  If this isn't the case, and if ICE is the forgone
conclusion, then let's use Jonathan's text above.

But NAT penetration is a "make or break" feature on which we shouldn't
compromise.  Assigning a "no-bid contract" to ICE without investigating the
alternatives seems extremely convenient, but hardly ideal.

If ICE is so clearly the right choice, what's the harm in proving it by
pitting against a field of challengers?

-david



From RADHIKA.R.ROY at saic.com  Mon Mar 13 20:45:45 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Mon, 13 Mar 2006 20:45:45 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31E9621@0591-ITS-EXMP02.us.saic.com>

Inline [RRR]

  _____  

From: p2p-sip-bounces at cs.columbia.edu on behalf of Eunsoo Shim
Sent: Mon 3/13/2006 6:10 PM
To: David Barrett
Cc: P2P-SIP
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal



If we see major problems with things under the charters of other WGs and 
we need fix for the problems even though the problems are not specific 
to P2P (SIP), then we should join the appropriate WGs and work out 
solutions there. I would be happy to join such efforts. 

[RRR] A universal solution for FW/NAT traversal did NOT work in the past
(e.g., couple of WGs were formed with this expectation). Is not possible to
understand what p2p-SIP wants for FW/NAT traversal first? I guess that this
has been main focus for this discussion initially.

[RRR] Later on, I believe, ICE and others on-going works of FW/NAT traversal
by MMUSIC and other WGs were broght into discussions as a part of the
pointers for solutions even for p2p-SIP.

[RRR] If FW/NAT traversal becomes make or break for p2p-SIP, I guess, we
need to investigate further to have final decision as others (e.g., David)
also pointed out: ICE/STUN and alternatives to ICE/STUN .

Thanks. 

Eunsoo 


From slavitch at gmail.com  Mon Mar 13 21:39:15 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 13 Mar 2006 21:39:15 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31E9621@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9621@0591-ITS-EXMP02.us.saic.com>
Message-ID: <1fc32c8f0603131839i32e282d5tf7d43a8ad69dce36@mail.gmail.com>

P2PSIP may be the wrong name for the WG.  I don't have a better one but it's
more P2P than it is pure SIP.

In fact what I keep on hearing is that changing SIP is verboten. So, let's
not do that any more than we have to.  If the complaint is "don't mess with
SIP", then "don't mess with SIP".   Message heard.  Live within what can be
construed as legitimate constraints.

Let me come up with yet another new buzzword/acronym. If you're not fond of
it I don't mind.

Peer To Peer Overlay With Address Overloading:  P2POAO

Advantages:

1.   Overlay is abstracted from application and protocol. SIP et al  sit on
top of OAO.
2.   Changes to SIP/SDP/RTP involve only AO  (recognizing overlay
addresses), which is
      nomenclature. One example of an overlay address could be a CERT-like
numeric string.
      In fact it is probably the best overlay address. Globally unique
identifiers with real
      certified identity.
3.   These changes do not alter state nor existing functionality of existing
standards.
4.   They are additions to, not changes to, existing protocols. Old still
works with new.
5.   Overlay NAT traversal is still needed, but need not be offer/answer.
6.   If offer/answer is needed ICE would be the IETF preference..
7.   If ICE can't support the overlay,  then request changes to ICE.
8.   If ICE can't or won't be changed to support the overlay, don't use
offer/answer.
9.   Offer/answer is not really needed if bootstrap can be seeded out of
band.
10. Once bootstrap is taken care of there is a wealth of choices.

For offer answer the answer that is insisted upon looks to be ICE.
But if it isn't offer-answer there are a wealth of things to draw upon,

The advantage is that there are a whole number of overlay schemes that would
work well in such an OAO environment. I do not know which one is best for
P2PSIP, but I have my personal preferences that I can argue later.  But that
preference is not religious, it just has to work. First we have to decide if
the overlay and SIP should be separate entities.

Given "don't change SIP"  there is no other way to go forward. There is no
other choice, so the choice is easy.

Now:

STUN-like things may still be needed, as a protocol that assists discovery.

TURN-like things (might as well be TURN-derived) could supply super node
services.
(In fact, they could be only self-organizing self-advertising
overlay-addressed versions of STUN and TURN)

ICE would still be used by SIP over P2POAO  if the overlay translates the
addresses transparently. This may well be the only requirement change for
ICE:

1.  to recognize overlay addressing as legitimate.
2.  to accept substitutions from the overlay, make it the canonical source
of real addresses.
3.  to move from 'overlay' addressing to 'real' and vice-versa using the
overlay as a reference.

That's about it.  Everything else in ICE would remain the same.  We may be
able to live with that.

Nothing apart from some basic housekeeping would change for MMUSIC, SIP,
SIPING amongst others. (an investigation would be needed to make sure this
is true).

The rest  of the work would belong in P2PSIP (or P2POAO or whatever it
really is).
Perhaps it's the "SIP" part of the name that causes kerfuffles, so if so,
name it something else.  Perhaps.

So, if the insistence is "don't change SIP"  and "for offer/answer NAT
traversal use ICE" the answer well may be:

1. Don't change SIP (apart from name support).
2. If ICE doesn't work for P2PSIP overlay offer/answer don't use
offer/answer in the overlay.
3. Given that offer/answer would bot be used by the overlay look for
alternatives and their traversal methods that are hence wise appropriate.
4. Allow non-ICE SIP UA systems to use the overlay as a tunnel in the place
of ICE.

I am not sure how this could step on toes. If ICE is an algorithm for
offer/answer protocols then not using ICE for protocols that are not based
on offer-answer should not be a problem.

If it still is, we're talking religion, not technology, and that means there
is no desire for a solution to the problem.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060313/af7881d2/attachment.html

From RADHIKA.R.ROY at saic.com  Mon Mar 13 22:34:43 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Mon, 13 Mar 2006 22:34:43 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31E9626@0591-ITS-EXMP02.us.saic.com>

Mike:
 
Just some word-smithing. How about: Peer-to-Peer Overlay Networking Protocol
(P2P-ONP)?
 
BR/Radhika

  _____  

From: Michael Slavitch [mailto:slavitch at gmail.com]
Sent: Mon 3/13/2006 9:39 PM
To: Roy, Radhika R.
Cc: Eunsoo Shim; David Barrett; P2P-SIP
Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal


P2PSIP may be the wrong name for the WG.  I don't have a better one but it's
more P2P than it is pure SIP. 

In fact what I keep on hearing is that changing SIP is verboten. So, let's
not do that any more than we have to.  If the complaint is "don't mess with
SIP", then "don't mess with SIP".   Message heard.  Live within what can be
construed as legitimate constraints. 

Let me come up with yet another new buzzword/acronym. If you're not fond of
it I don't mind.

Peer To Peer Overlay With Address Overloading:  P2POAO

Advantages: 

1.   Overlay is abstracted from application and protocol. SIP et al  sit on
top of OAO.
2.   Changes to SIP/SDP/RTP involve only AO  (recognizing overlay
addresses), which is 
      nomenclature. One example of an overlay address could be a CERT-like
numeric string. 
      In fact it is probably the best overlay address. Globally unique
identifiers with real   
      certified identity.
3.   These changes do not alter state nor existing functionality of existing
standards. 
4.   They are additions to, not changes to, existing protocols. Old still
works with new.
5.   Overlay NAT traversal is still needed, but need not be offer/answer.
6.   If offer/answer is needed ICE would be the IETF preference.. 
7.   If ICE can't support the overlay,  then request changes to ICE.
8.   If ICE can't or won't be changed to support the overlay, don't use
offer/answer.
9.   Offer/answer is not really needed if bootstrap can be seeded out of
band. 
10. Once bootstrap is taken care of there is a wealth of choices.

For offer answer the answer that is insisted upon looks to be ICE. 
But if it isn't offer-answer there are a wealth of things to draw upon, 

The advantage is that there are a whole number of overlay schemes that would
work well in such an OAO environment. I do not know which one is best for
P2PSIP, but I have my personal preferences that I can argue later.  But that
preference is not religious, it just has to work. First we have to decide if
the overlay and SIP should be separate entities. 

Given "don't change SIP"  there is no other way to go forward. There is no
other choice, so the choice is easy.

Now:

STUN-like things may still be needed, as a protocol that assists discovery. 

TURN-like things (might as well be TURN-derived) could supply super node
services.
(In fact, they could be only self-organizing self-advertising
overlay-addressed versions of STUN and TURN) 

ICE would still be used by SIP over P2POAO  if the overlay translates the
addresses transparently. This may well be the only requirement change for
ICE: 

1.  to recognize overlay addressing as legitimate.
2.  to accept substitutions from the overlay, make it the canonical source
of real addresses.
3.  to move from 'overlay' addressing to 'real' and vice-versa using the
overlay as a reference. 

That's about it.  Everything else in ICE would remain the same.  We may be
able to live with that.

Nothing apart from some basic housekeeping would change for MMUSIC, SIP,
SIPING amongst others. (an investigation would be needed to make sure this
is true). 

The rest  of the work would belong in P2PSIP (or P2POAO or whatever it
really is).
Perhaps it's the "SIP" part of the name that causes kerfuffles, so if so,
name it something else.  Perhaps.

So, if the insistence is "don't change SIP"  and "for offer/answer NAT
traversal use ICE" the answer well may be: 

1. Don't change SIP (apart from name support).
2. If ICE doesn't work for P2PSIP overlay offer/answer don't use
offer/answer in the overlay.
3. Given that offer/answer would bot be used by the overlay look for
alternatives and their traversal methods that are hence wise appropriate. 
4. Allow non-ICE SIP UA systems to use the overlay as a tunnel in the place
of ICE.

I am not sure how this could step on toes. If ICE is an algorithm for
offer/answer protocols then not using ICE for protocols that are not based
on offer-answer should not be a problem. 

If it still is, we're talking religion, not technology, and that means there
is no desire for a solution to the problem.



From slavitch at gmail.com  Mon Mar 13 22:51:05 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Mon, 13 Mar 2006 22:51:05 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31E9626@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9626@0591-ITS-EXMP02.us.saic.com>
Message-ID: <1fc32c8f0603131951y6efe446cg67568f4901908cd5@mail.gmail.com>

Sure.  That sounds more IETF. Fits with the religion.  Thank you.

We're naming a baby.  From where my parents come from this is a serious
business:

1. Don't piss anyone off.
2. Don't confuse them.
3. It must be pronounced in the native dialect and not seem too exotic.
4. It must be accepted by the rabbis or priests or shamans.
5. It must respect community elders.

Requirements of a name?

1. Say what it is.
2. Do what it says.

Are we doing SIP? Probably not.
Are we doing overlays? Yup.
Are we going to address them somehow that allows us to traverse networks and
protocols? If we want to build something more than a toy: yes.

Is it peer to peer? Yup.

P2P is the surname, which, like in certain traditions, goes first.

P2PONP?  Good.  Let's poll for alternatives, but I like it. Personally I
like mine, but P2PONP is more IETF-ish. I don't care as long as it meets the
requirements above.

OAO would be a nickname, then. Which, when you think a little, is exactly
what OAO is, a very pedantic and accurate nickname mechanism, globally
unique and assured.

What kind of nickname will be the next interesting discussion.  If we get to
that point we've entered  a process of problem reduction, not process
thrash.

M


On 3/13/06, Roy, Radhika R. <RADHIKA.R.ROY at saic.com> wrote:
>
> Mike:
>
> Just some word-smithing. How about: Peer-to-Peer Overlay Networking
> Protocol
> (P2P-ONP)?
>
> BR/Radhika
>
>   _____
>
> From: Michael Slavitch [mailto:slavitch at gmail.com]
> Sent: Mon 3/13/2006 9:39 PM
> To: Roy, Radhika R.
> Cc: Eunsoo Shim; David Barrett; P2P-SIP
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>
>
> P2PSIP may be the wrong name for the WG.  I don't have a better one but
> it's
> more P2P than it is pure SIP.
>
> In fact what I keep on hearing is that changing SIP is verboten. So, let's
> not do that any more than we have to.  If the complaint is "don't mess
> with
> SIP", then "don't mess with SIP".   Message heard.  Live within what can
> be
> construed as legitimate constraints.
>
> Let me come up with yet another new buzzword/acronym. If you're not fond
> of
> it I don't mind.
>
> Peer To Peer Overlay With Address Overloading:  P2POAO
>
> Advantages:
>
> 1.   Overlay is abstracted from application and protocol. SIP et al  sit
> on
> top of OAO.
> 2.   Changes to SIP/SDP/RTP involve only AO  (recognizing overlay
> addresses), which is
>       nomenclature. One example of an overlay address could be a CERT-like
> numeric string.
>       In fact it is probably the best overlay address. Globally unique
> identifiers with real
>       certified identity.
> 3.   These changes do not alter state nor existing functionality of
> existing
> standards.
> 4.   They are additions to, not changes to, existing protocols. Old still
> works with new.
> 5.   Overlay NAT traversal is still needed, but need not be offer/answer.
> 6.   If offer/answer is needed ICE would be the IETF preference..
> 7.   If ICE can't support the overlay,  then request changes to ICE.
> 8.   If ICE can't or won't be changed to support the overlay, don't use
> offer/answer.
> 9.   Offer/answer is not really needed if bootstrap can be seeded out of
> band.
> 10. Once bootstrap is taken care of there is a wealth of choices.
>
> For offer answer the answer that is insisted upon looks to be ICE.
> But if it isn't offer-answer there are a wealth of things to draw upon,
>
> The advantage is that there are a whole number of overlay schemes that
> would
> work well in such an OAO environment. I do not know which one is best for
> P2PSIP, but I have my personal preferences that I can argue later.  But
> that
> preference is not religious, it just has to work. First we have to decide
> if
> the overlay and SIP should be separate entities.
>
> Given "don't change SIP"  there is no other way to go forward. There is no
> other choice, so the choice is easy.
>
> Now:
>
> STUN-like things may still be needed, as a protocol that assists
> discovery.
>
> TURN-like things (might as well be TURN-derived) could supply super node
> services.
> (In fact, they could be only self-organizing self-advertising
> overlay-addressed versions of STUN and TURN)
>
> ICE would still be used by SIP over P2POAO  if the overlay translates the
> addresses transparently. This may well be the only requirement change for
> ICE:
>
> 1.  to recognize overlay addressing as legitimate.
> 2.  to accept substitutions from the overlay, make it the canonical source
> of real addresses.
> 3.  to move from 'overlay' addressing to 'real' and vice-versa using the
> overlay as a reference.
>
> That's about it.  Everything else in ICE would remain the same.  We may be
> able to live with that.
>
> Nothing apart from some basic housekeeping would change for MMUSIC, SIP,
> SIPING amongst others. (an investigation would be needed to make sure this
> is true).
>
> The rest  of the work would belong in P2PSIP (or P2POAO or whatever it
> really is).
> Perhaps it's the "SIP" part of the name that causes kerfuffles, so if so,
> name it something else.  Perhaps.
>
> So, if the insistence is "don't change SIP"  and "for offer/answer NAT
> traversal use ICE" the answer well may be:
>
> 1. Don't change SIP (apart from name support).
> 2. If ICE doesn't work for P2PSIP overlay offer/answer don't use
> offer/answer in the overlay.
> 3. Given that offer/answer would bot be used by the overlay look for
> alternatives and their traversal methods that are hence wise appropriate.
> 4. Allow non-ICE SIP UA systems to use the overlay as a tunnel in the
> place
> of ICE.
>
> I am not sure how this could step on toes. If ICE is an algorithm for
> offer/answer protocols then not using ICE for protocols that are not based
> on offer-answer should not be a problem.
>
> If it still is, we're talking religion, not technology, and that means
> there
> is no desire for a solution to the problem.
>
>
>


--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060313/9b280fe8/attachment-0001.html

From fluffy at cisco.com  Tue Mar 14 00:36:11 2006
From: fluffy at cisco.com (Cullen Jennings)
Date: Mon, 13 Mar 2006 21:36:11 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <1fc32c8f0603131839i32e282d5tf7d43a8ad69dce36@mail.gmail.com>
Message-ID: <C03B964B.7BA08%fluffy@cisco.com>


One of the items that was brought at some of the previous ad-hoc meetings
was that general overlay network problem seemed to be a very wide scope. In
fact it felt more like a research problem to many people than an engineering
problem. Overlay networks for file sharing clearly have very different
requirements than the overlay networks that have been discussed here. Doing
a general overlay network that worked for many application sounds hard. I'm
not sure if you were suggesting a wider scope or not but probably anything
we can do to keep the scope small is a good way to start.

Cullen


On 3/13/06 6:39 PM, "Michael Slavitch" <slavitch at gmail.com> wrote:

> P2PSIP may be the wrong name for the WG.  I don't have a better one but it's
> more P2P than it is pure SIP.
> 
> In fact what I keep on hearing is that changing SIP is verboten. So, let's not
> do that any more than we have to.  If the complaint is "don't mess with SIP",
> then "don't mess with SIP".   Message heard.  Live within what can be
> construed as legitimate constraints.
> 
> Let me come up with yet another new buzzword/acronym. If you're not fond of it
> I don't mind.
> 
> Peer To Peer Overlay With Address Overloading:  P2POAO
> 
> Advantages: 
> 
> 1.   Overlay is abstracted from application and protocol. SIP et al  sit on
> top of OAO.
> 2.   Changes to SIP/SDP/RTP involve only AO  (recognizing overlay addresses),
> which is 
>       nomenclature. One example of an overlay address could be a CERT-like
> numeric string. 
>       In fact it is probably the best overlay address. Globally unique
> identifiers with real
>       certified identity.
> 3.   These changes do not alter state nor existing functionality of existing
> standards. 
> 4.   They are additions to, not changes to, existing protocols. Old still
> works with new.
> 5.   Overlay NAT traversal is still needed, but need not be offer/answer.
> 6.   If offer/answer is needed ICE would be the IETF preference..
> 7.   If ICE can't support the overlay,  then request changes to ICE.
> 8.   If ICE can't or won't be changed to support the overlay, don't use
> offer/answer.
> 9.   Offer/answer is not really needed if bootstrap can be seeded out of band.
> 10. Once bootstrap is taken care of there is a wealth of choices.
> 
> For offer answer the answer that is insisted upon looks to be ICE.
> But if it isn't offer-answer there are a wealth of things to draw upon,
> 
> The advantage is that there are a whole number of overlay schemes that would
> work well in such an OAO environment. I do not know which one is best for
> P2PSIP, but I have my personal preferences that I can argue later.  But that
> preference is not religious, it just has to work. First we have to decide if
> the overlay and SIP should be separate entities.
> 
> Given "don't change SIP"  there is no other way to go forward. There is no
> other choice, so the choice is easy.
> 
> Now:
> 
> STUN-like things may still be needed, as a protocol that assists discovery.
> 
> TURN-like things (might as well be TURN-derived) could supply super node
> services.
> (In fact, they could be only self-organizing self-advertising
> overlay-addressed versions of STUN and TURN)
> 
> ICE would still be used by SIP over P2POAO  if the overlay translates the
> addresses transparently. This may well be the only requirement change for ICE:
> 
> 1.  to recognize overlay addressing as legitimate.
> 2.  to accept substitutions from the overlay, make it the canonical source of
> real addresses.
> 3.  to move from 'overlay' addressing to 'real' and vice-versa using the
> overlay as a reference.
> 
> That's about it.  Everything else in ICE would remain the same.  We may be
> able to live with that.
> 
> Nothing apart from some basic housekeeping would change for MMUSIC, SIP,
> SIPING amongst others. (an investigation would be needed to make sure this is
> true). 
> 
> The rest  of the work would belong in P2PSIP (or P2POAO or whatever it really
> is).
> Perhaps it's the "SIP" part of the name that causes kerfuffles, so if so, name
> it something else.  Perhaps.
> 
> So, if the insistence is "don't change SIP"  and "for offer/answer NAT
> traversal use ICE" the answer well may be:
> 
> 1. Don't change SIP (apart from name support).
> 2. If ICE doesn't work for P2PSIP overlay offer/answer don't use offer/answer
> in the overlay.
> 3. Given that offer/answer would bot be used by the overlay look for
> alternatives and their traversal methods that are hence wise appropriate.
> 4. Allow non-ICE SIP UA systems to use the overlay as a tunnel in the place of
> ICE.
> 
> I am not sure how this could step on toes. If ICE is an algorithm for
> offer/answer protocols then not using ICE for protocols that are not based on
> offer-answer should not be a problem.
> 
> If it still is, we're talking religion, not technology, and that means there
> is no desire for a solution to the problem.
> 
> 
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060314/4bee848b/attachment.html

From fluffy at cisco.com  Tue Mar 14 00:52:20 2006
From: fluffy at cisco.com (Cullen Jennings)
Date: Mon, 13 Mar 2006 21:52:20 -0800
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <ECC0741B-DCC2-4E4E-AB85-7E3268E6AADF@softarmor.com>
Message-ID: <C03B9A14.7BA21%fluffy@cisco.com>

On 3/13/06 2:33 PM, "Dean Willis" <dean.willis at softarmor.com> wrote:

> IETF-wise, we may have a
> difficult time chartering work that overlaps with MMUSIC's work.

It's very good bet that you will have difficulties chartering works that
overlaps with any existing WG.

From dean.willis at softarmor.com  Tue Mar 14 01:30:38 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 14 Mar 2006 00:30:38 -0600
Subject: [p2p-sip] Names (was Re: Problem statement for P2P SIP NAT
	traversal)
In-Reply-To: <1fc32c8f0603131951y6efe446cg67568f4901908cd5@mail.gmail.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9626@0591-ITS-EXMP02.us.saic.com>
	<1fc32c8f0603131951y6efe446cg67568f4901908cd5@mail.gmail.com>
Message-ID: <3DDBD882-A2A0-4BD0-9316-133AB5895C2D@softarmor.com>


On Mar 13, 2006, at 9:51 PM, Michael Slavitch wrote:

>
> P2PONP?  Good.  Let's poll for alternatives, but I like it.  
> Personally I like mine, but P2PONP is more IETF-ish. I don't care  
> as long as it meets the requirements above.
>

I'm struggling to find something that could use the acronym P2PONY.  
How about P2PONI? "Peer 2 Peer Overlay Naming for Internet"  . . .  
Maybe "Peer Overlay Naming for Internet Element Signaling", PONIES.  
It has, well, spirit.

A good catchy acronym nets one a world of marketing power.

But we do have to be very careful.

If we stay laser-focused on applying P2P techniques to solve the  
rendezvous problem for SIP, we have a solvable problem. If we  
generalize, we could be in a world of hurt. The lesson here -- don't  
generalize.

SIP does Offer/Answer. I think that narrows our focus for media  
negotiation to O/A. But that's not the same thing as correspondent  
discovery, which does not have to be an O/A problem.

Without generalizing we still have a couple of Big Questions:

1) Is the Insert/Search method SIP or some other protocol, and if  
some other, then what?

2) If we make our supernodes do STUN/TURN, does ICE solve our NAT  
traversal problem?

3) How do we solve routing topology security in a way that makes EKR  
happy and still works? I suppose one solution is to put all the  
supernodes into the trust domain and have all the user functions on  
ordinary nodes, but I suspect that isn't an optimal model for  
distribution of labor. Might work OK for the "self organizing proxy  
farm" use case, however.

--
Dean


From drage at lucent.com  Tue Mar 14 05:14:34 2006
From: drage at lucent.com (Drage, Keith (Keith))
Date: Tue, 14 Mar 2006 10:14:34 -0000
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
Message-ID: <475FF955A05DD411980D00508B6D5FB00C290829@en0033exch001u.uk.lucent.com>

I'll strongly support Jonathan on this one.

We have essentially already had this discussion on a wider level that just NATs.

The direction that the draft charter envisages is that p2p reuses SIP and primarily is meant to deal with the discovery issues and so on that relate to the centralised behaviour of existing SIP. It is not designing a total replacement for SIP. That charter is still obviously to be discussed at the BOF and then by the IESG, but I think we have to use what exists for the moment.

Anything else must be addressed as a set of requirements to the group concerned with the general issues as it applies to SIP, and in my view such requirements must indicate why this is a special issue for p2p rather than a general issue for SIP, otherwise we descend into a war between two working groups on the best way to meet an issue for SIP generally.

And as for the contention that any solution must be widely deployed and RFC status. We why stop there, surely to stack up to Michael's criteria it must at the very least be a draft standard - because we can only prove it is widely deployed and interoperable by taking the RFC to draft standard. Oh, dear! SIP is still only proposed standard and the roadmap doesn't even exist yet to take it to draft standard.

regards

Keith

> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu]On Behalf Of Jonathan 
> Rosenberg
> Sent: 13 March 2006 21:18
> To: Michael Slavitch
> Cc: 'P2P-SIP'
> Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
> 
> 
> 
> 
> Michael Slavitch wrote:
> 
> > Jonathan Rosenberg wrote:
> > 
> >>
> >> As I have suggested before, if you have beefs with the 
> design of ICE, 
> >> take them to mmusic and lets discuss there as part of ICE 
> discussions. 
> >> If you have an alternative proposal, I look forward to the 
> draft. But, 
> >> it is a travesty to the community to use p2p sip as an excuse to 
> >> redesign parts of SIP with new requirements that have 
> nothing to do 
> >> with p2p.
> >>
> >> -Jonathan R.
> >>   
> > 
> > The result of the process will be a set of requirements to 
> solve the 
> > use-case needs of P2PSIP.
> > 
> > If it can be */demonstrated /*that ICE can meet these 
> requirements ICE 
> > will be fine. If not a set of requirements will be sent to MMUSIC.
> > If MMUSIC cannot act on these requirements or if ICE cannot 
> be seen as 
> > effectively meeting those requirements something else must be 
> > considered.  It is the onus of ICE to meet these 
> requirements, not the 
> > other way around.
> 
> I'm afraid I have to disagree with that.
> 
> It is not the job of p2p sip to redefine the requirements for 
> all of the 
> SIP extensions that our out there, so that you can choose to 
> retain or 
> discard various ones as you see fit. If this were the case, 
> "p2p sip" is 
> a poor name for this group. It sounds like some folks are looking for 
> 'p2p voip', and you'd like to define a protocol that is not 
> interoperable with sip to do that. Well, that is one possible 
> charter, 
> but not the one I'd hope for.
> 
> > 
> > I say requirement #1 is:  P2PSIP NAT traversal must be in broad 
> > interoperable use.
> > Teredo may be such a beast.  I am not so sure about ICE.
> 
> Well, we can debate all of that. ICE is actually in use quite 
> a bit in 
> various softphones and voip deployments in service providers today. 
> Teredo is for v6, as you know. As I said, a v6 only solution for nat 
> traversal is .... interesting. YMMV as to whether that meets 
> the needs 
> of the community.
> 
> In any case, all of this is really just sounding like the 'we hate 
> certain sip extensions and plan on doing our own' WG. Now, I 
> can point 
> to other SIP documents in various stages of deployment. So, I 
> guess, by 
> your metric, p2p sip shall be able to discard anything it 
> likes that has 
> not reached some magic deployment threshold? Absurd.
> 
> 
> > 
> > It is the requirements that win, not a draft 
> implementation.  ICE is 
> > still an Internet-Draft.  It is not yet RFC standard in 
> wide adoption.
> 
> Its not an RFC, no. WIDE usage? Depends on your definition. 
> Implemented? 
> Absolutely. Deployed? Yes.
> 
> I'll note I found this criteria (technique must be in widespread 
> deployment) laughable. p2p SIP itself is not even yet a 
> working group. 
> We've got threads going on debating fundamental issues like, 
> should the 
> DHT be done within SIP or through a layer separate? And can it be 
> secured? p2p SIP is itself only just getting started. And you 
> are going 
> to throw away a fundamental piece of SIP architecture because its not 
> done yet? Now that is ironic.
> 
> 
> > 
> > It is true that P2PSIP should borrow on existing standards 
> for SIP.  SIP 
> > has been around for a decade.
> > ICE is still in draft form, version seven and counting.
> > 
> > When ICE is RFC level and in wide interoperable use the 
> same will be 
> > said for it.
> > Until then it is a draft, and drafts can still be fed 
> requirements or 
> > suggestions, or rejected outright.
> 
> Sure it can be changed or improved; we've been doing that for 
> a while. 
> This is a hard problem as you know, with many constraints 
> from different 
> communities.
> 
> But rejected outright? Well, its a WG document, it has broad support 
> from the WG as such. There are no plans for its rejection 
> that I know of.
> 
> 
> > How many iterations of ICE have we seen. How many YEARS has 
> it taken? 
> > And it is still in draft form. I would respect "use ICE" if 
> the ICE RFC 
> > was settled five years ago.  It was not.  ICE is still an 
> open question.
> > 
> > Meanwhile,  Skype works right now and is gobbling up share. 
>  If ICE was 
> > a simple RFC three years ago this would not be the case.
> 
> Ah. So, by your metric (wide deployment), our only possible 
> solution for 
> p2p SIP, nat traversal and all, would be Skype?
> 
> -Jonathan R.
> 
> -- 
> Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
> Cisco Fellow                                   Parsippany, NJ 
> 07054-2711
> Cisco Systems
> jdrosen at cisco.com                              FAX:   (973) 952-5050
> http://www.jdrosen.net                         PHONE: (973) 952-5000
> http://www.cisco.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 

From slavitch at gmail.com  Tue Mar 14 05:50:25 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Tue, 14 Mar 2006 05:50:25 -0500
Subject: [p2p-sip] Names (was Re: Problem statement for P2P SIP NAT
	traversal)
In-Reply-To: <3DDBD882-A2A0-4BD0-9316-133AB5895C2D@softarmor.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9626@0591-ITS-EXMP02.us.saic.com>
	<1fc32c8f0603131951y6efe446cg67568f4901908cd5@mail.gmail.com>
	<3DDBD882-A2A0-4BD0-9316-133AB5895C2D@softarmor.com>
Message-ID: <44169FF1.6050904@gmail.com>

The IETF must remember it's in a marketing war not of its choosing.  It 
is in an asymmetric war and it is losing.

Dean, I very much like your questions. They are good questions that 
encourage a solution.

It is 5:45 AM here, my dogs need a walk, I need a coffee and get ready 
to catch a plane.

I cannot yet offer an intelligent answer. (I also should never pun.)

I agree with this point:   Stay focused on SIP. Don't overgeneralize.  
But don't drop potential.  Do not over-restrict.

As a WG requirement this would be translated as:

"The focus is on solving problems associated with SIP."

In other words.  Fix things for SIP.  If it works for other things LATER 
that is a bonus.  First things first.

I'm sure we're all smart enough to handle 'other things' later.  If the 
IETF can spend years jamming class-5 switching into SIP we can do 'other 
things' with p2p-oriented SIP.  Amazing things.  First make it work for SIP.

Michael

Dean Willis wrote:
>
> On Mar 13, 2006, at 9:51 PM, Michael Slavitch wrote:
>
>>
>> P2PONP?  Good.  Let's poll for alternatives, but I like it. 
>> Personally I like mine, but P2PONP is more IETF-ish. I don't care as 
>> long as it meets the requirements above.
>>
>
> I'm struggling to find something that could use the acronym P2PONY. 
> How about P2PONI? "Peer 2 Peer Overlay Naming for Internet"  . . . 
> Maybe "Peer Overlay Naming for Internet Element Signaling", PONIES. It 
> has, well, spirit.
>
> A good catchy acronym nets one a world of marketing power.
>
> But we do have to be very careful.
>
> If we stay laser-focused on applying P2P techniques to solve the 
> rendezvous problem for SIP, we have a solvable problem. If we 
> generalize, we could be in a world of hurt. The lesson here -- don't 
> generalize.
>
> SIP does Offer/Answer. I think that narrows our focus for media 
> negotiation to O/A. But that's not the same thing as correspondent 
> discovery, which does not have to be an O/A problem.
>
> Without generalizing we still have a couple of Big Questions:
>
> 1) Is the Insert/Search method SIP or some other protocol, and if some 
> other, then what?
>
> 2) If we make our supernodes do STUN/TURN, does ICE solve our NAT 
> traversal problem?
>
> 3) How do we solve routing topology security in a way that makes EKR 
> happy and still works? I suppose one solution is to put all the 
> supernodes into the trust domain and have all the user functions on 
> ordinary nodes, but I suspect that isn't an optimal model for 
> distribution of labor. Might work OK for the "self organizing proxy 
> farm" use case, however.
>
> -- 
> Dean
>
>


From bryan at ethernot.org  Tue Mar 14 09:58:28 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Tue, 14 Mar 2006 09:58:28 -0500
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <C03B9A14.7BA21%fluffy@cisco.com>
References: <ECC0741B-DCC2-4E4E-AB85-7E3268E6AADF@softarmor.com>
	<C03B9A14.7BA21%fluffy@cisco.com>
Message-ID: <8b2769930603140658g3956248dv14d911f9eeb25d63@mail.gmail.com>

I'll chime in late here...on planes all day yesterday.

I think there are several motivations for reusing existing stuff
unless we really, really believe it won't work. That was certainly how
I intended the wording to read when I first started hacking up an
attempt at the charter:

1) Interoperability. If regular SIP endpoints use method x to do
function y, we should try as hard as possible to do that. If not, we
have just created another island like Skype. Maybe it is an open
island, but if it can't talk to regular SIP devices properly or only
in very limited cases, it is still an island.

2) Time to finish. The more of the (extensive) work that has been done
on SIP extensions we can reuse, the more quickly we can develop a
version of SIP with distributed resource location, and perhaps later
other distributed aspects (voice mail or what have you). If we
redesign something for the sake of redesigning it, we delay that.

Now of course the above statements don't mean every SIP extension
designed with client-server in mind will prove to be perfect in a p2p
environment, but I really think the burden of proof should be on does
it not work. Otherwise, we are, as some people pointed out, looking at
a different communications system, not really SIP. I don't think it
is, but if that is what folks really want, our straw man charter
really needs a whole lot of rewording ;)

Just my 2 cents.

David Bryan

On 3/14/06, Cullen Jennings <fluffy at cisco.com> wrote:
> On 3/13/06 2:33 PM, "Dean Willis" <dean.willis at softarmor.com> wrote:
>
> > IETF-wise, we may have a
> > difficult time chartering work that overlaps with MMUSIC's work.
>
> It's very good bet that you will have difficulties chartering works that
> overlaps with any existing WG.
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From RADHIKA.R.ROY at saic.com  Tue Mar 14 10:09:24 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Tue, 14 Mar 2006 10:09:24 -0500
Subject: [p2p-sip] Names (was Re: Problem statement for P2P SIP NAT
	tr	aversal)
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31D9DC0@0591-ITS-EXMP02.us.saic.com>

Dean:

The only problem that I have with your suggestion is the word: "NAMING."

NAMING seems to be very "limited" in scope. I believe that it has also
another aspect that we might be looking for: "PROTOCOL."

So, still my vote will be as follows:

1. P2P-ONP (P2P Overlay Networking Protocol)
2. Others

BR/Radhika

-----Original Message-----
From: Dean Willis [mailto:dean.willis at softarmor.com] 
Sent: Tuesday, March 14, 2006 1:31 AM
To: Michael Slavitch
Cc: Roy, Radhika R.; P2P-SIP
Subject: Names (was Re: [p2p-sip] Problem statement for P2P SIP NAT
traversal)


On Mar 13, 2006, at 9:51 PM, Michael Slavitch wrote:

>
> P2PONP?  Good.  Let's poll for alternatives, but I like it.  
> Personally I like mine, but P2PONP is more IETF-ish. I don't care  
> as long as it meets the requirements above.
>

I'm struggling to find something that could use the acronym P2PONY.  
How about P2PONI? "Peer 2 Peer Overlay Naming for Internet"  . . .  
Maybe "Peer Overlay Naming for Internet Element Signaling", PONIES.  
It has, well, spirit.

A good catchy acronym nets one a world of marketing power.

But we do have to be very careful.

If we stay laser-focused on applying P2P techniques to solve the  
rendezvous problem for SIP, we have a solvable problem. If we  
generalize, we could be in a world of hurt. The lesson here -- don't  
generalize.

SIP does Offer/Answer. I think that narrows our focus for media  
negotiation to O/A. But that's not the same thing as correspondent  
discovery, which does not have to be an O/A problem.

Without generalizing we still have a couple of Big Questions:

1) Is the Insert/Search method SIP or some other protocol, and if  
some other, then what?

2) If we make our supernodes do STUN/TURN, does ICE solve our NAT  
traversal problem?

3) How do we solve routing topology security in a way that makes EKR  
happy and still works? I suppose one solution is to put all the  
supernodes into the trust domain and have all the user functions on  
ordinary nodes, but I suspect that isn't an optimal model for  
distribution of labor. Might work OK for the "self organizing proxy  
farm" use case, however.

--
Dean


From gregd at research.panasonic.com  Tue Mar 14 10:16:08 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Tue, 14 Mar 2006 10:16:08 -0500
Subject: [p2p-sip] Names (was Re: Problem statement for P2P SIP
	NAT	traversal)
In-Reply-To: <44169FF1.6050904@gmail.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9626@0591-ITS-EXMP02.us.saic.com>
	<1fc32c8f0603131951y6efe446cg67568f4901908cd5@mail.gmail.com>
	<3DDBD882-A2A0-4BD0-9316-133AB5895C2D@softarmor.com>
	<44169FF1.6050904@gmail.com>
Message-ID: <4416DE38.6020801@research.panasonic.com>

Hi Michael,

I'm agreeing with you here.

Michael Slavitch wrote:
> The IETF must remember it's in a marketing war not of its choosing.  It 
> is in an asymmetric war and it is losing.

I understand where you're coming from, but the idealist in me
sees the current phase as an opportunity rather than a threat.

> Dean, I very much like your questions. They are good questions that 
> encourage a solution.
> 
> It is 5:45 AM here, my dogs need a walk, I need a coffee and get ready 
> to catch a plane.
> 
> I cannot yet offer an intelligent answer. (I also should never pun.)

Indeed.

> I agree with this point:   Stay focused on SIP. Don't overgeneralize.  
> But don't drop potential.  Do not over-restrict.

Right.  We've got a "distributed naming and lookup" problem to
solve.

This doesn't mean the ideas aren't more general,
but we're not tasked with making a file-sharing network
nor with replacing DNS.

We're here to establish end-to-end real time sessions between
hosts, and to connect Client/Server SIP systems to P2P systems,
even through NATs.

That's the whole box and dice.

We'll certainly have enough on our plate that even with a concerted
effort to work together and choose an acceptable (rather than best)
solution, we've still got a long, hard (but rewarding) road ahead.

Greg

From gregd at research.panasonic.com  Tue Mar 14 10:27:33 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Tue, 14 Mar 2006 10:27:33 -0500
Subject: [p2p-sip] Names (was Re: Problem statement for P2P SIP
	NAT	traversal)
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31D9DC0@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31D9DC0@0591-ITS-EXMP02.us.saic.com>
Message-ID: <4416E0E5.4040908@research.panasonic.com>

Hi Radhika,

Let's not change the name.

We're here to work on SIP related P2P only.

Even if we don't change the SIP protocol,
we still have to define operations for devices
which perform both P2P and CS SIP.

Certainly, we have presence work to do, and that's
not just a naming issue, it's a state and protocol
issue too (and one which has been simply, but neatly
described by Kundan and Henning in their technical
report.), but I don't think it needs to modify the name
of the WG.

If other P2P operations need to be performed,
let another WG worry about it: they can use our
protocols and tools if need be.

Greg

Roy, Radhika R. wrote:
> Dean:
> 
> The only problem that I have with your suggestion is the word: "NAMING."
> 
> NAMING seems to be very "limited" in scope. I believe that it has also
> another aspect that we might be looking for: "PROTOCOL."
> 
> So, still my vote will be as follows:
> 
> 1. P2P-ONP (P2P Overlay Networking Protocol)
> 2. Others
> 
> BR/Radhika
> 
> -----Original Message-----
> From: Dean Willis [mailto:dean.willis at softarmor.com] 
> Sent: Tuesday, March 14, 2006 1:31 AM
> To: Michael Slavitch
> Cc: Roy, Radhika R.; P2P-SIP
> Subject: Names (was Re: [p2p-sip] Problem statement for P2P SIP NAT
> traversal)
> 
> 
> On Mar 13, 2006, at 9:51 PM, Michael Slavitch wrote:
> 
>> P2PONP?  Good.  Let's poll for alternatives, but I like it.  
>> Personally I like mine, but P2PONP is more IETF-ish. I don't care  
>> as long as it meets the requirements above.
>>
> 
> I'm struggling to find something that could use the acronym P2PONY.  
> How about P2PONI? "Peer 2 Peer Overlay Naming for Internet"  . . .  
> Maybe "Peer Overlay Naming for Internet Element Signaling", PONIES.  
> It has, well, spirit.
> 
> A good catchy acronym nets one a world of marketing power.
> 
> But we do have to be very careful.
> 
> If we stay laser-focused on applying P2P techniques to solve the  
> rendezvous problem for SIP, we have a solvable problem. If we  
> generalize, we could be in a world of hurt. The lesson here -- don't  
> generalize.
> 
> SIP does Offer/Answer. I think that narrows our focus for media  
> negotiation to O/A. But that's not the same thing as correspondent  
> discovery, which does not have to be an O/A problem.
> 
> Without generalizing we still have a couple of Big Questions:
> 
> 1) Is the Insert/Search method SIP or some other protocol, and if  
> some other, then what?
> 
> 2) If we make our supernodes do STUN/TURN, does ICE solve our NAT  
> traversal problem?
> 
> 3) How do we solve routing topology security in a way that makes EKR  
> happy and still works? I suppose one solution is to put all the  
> supernodes into the trust domain and have all the user functions on  
> ordinary nodes, but I suspect that isn't an optimal model for  
> distribution of labor. Might work OK for the "self organizing proxy  
> farm" use case, however.
> 
> --
> Dean
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From aki.niemi at nokia.com  Tue Mar 14 10:30:02 2006
From: aki.niemi at nokia.com (Aki Niemi)
Date: Tue, 14 Mar 2006 17:30:02 +0200
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4415EAF1.6050308@cisco.com>
References: <4m1t2v$bvcp4d@sj-inbound-a.cisco.com>	<4415CD62.4020906@cisco.com>	
	<4415DBF3.1090009@gmail.com>	<4415E1A3.7010502@cisco.com>	<1fc32c8f0603131342x1d397ef0y6ca74b60bc276ae5@mail.gmail.com>
	<4415EAF1.6050308@cisco.com>
Message-ID: <4416E17A.1010800@nokia.com>



ext Jonathan Rosenberg wrote:
> If you want p2p-sip to 'just work', I suggest that time is better spent 
> figuring out how to eliminate the need to configure the servers that ICE 
> depends on - turn and stun servers - and figure out how to dynamically 
> find ones amongst peers in the mesh. 

Exactly.

However, last time I looked, this was buried in the charter text as
something to work on "later". I think it should be first priority. In
fact, I'd hazard a guess that running media relays (and not the SIP
rendezvous) is the highest costing resource that most desperately needs
to be distributed in this system.

Cheers,
Aki

From enrico.marocco at telecomitalia.it  Tue Mar 14 10:57:41 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Tue, 14 Mar 2006 16:57:41 +0100
Subject: [p2p-sip] Names (was Re: Problem statement for P2P SIP NAT	tr
	aversal)
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31D9DC0@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31D9DC0@0591-ITS-EXMP02.us.saic.com>
Message-ID: <4416E7F5.8010708@telecomitalia.it>

Roy, Radhika R. wrote:
> So, still my vote will be as follows:
> 
> 1. P2P-ONP (P2P Overlay Networking Protocol)

It sounds good, it ends in "P", but it doesn't meet the first 
requirement for any *successful* protocol: it MUST be 4-letter-long (not 
3: 4!).
Moreover, it has the following drowbacks:
+ it shows a number in the middle;
+ it has an hyphen;
+ it has nothing to do with my favourite text editor.

Thereby my suggestion would be "Serverless Kill Yank Protocol": SKYP, 
and you've got what you need!

Think about it...

Enrico
--------------------------------------------------------------------

CONFIDENTIALITY NOTICE

This message and its attachments are addressed solely to the persons above and may contain confidential information. If you have received the message in error, be informed that any use of the content hereof is prohibited. Please return it immediately to the sender and delete the message. Should you have any questions, please contact us by replying to webmaster at telecomitalia.it.

        Thank you

                                        www.telecomitalia.it

--------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060314/f0ed6b64/attachment.html

From bryan at ethernot.org  Tue Mar 14 12:36:21 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Tue, 14 Mar 2006 12:36:21 -0500
Subject: [p2p-sip] Names (was Re: Problem statement for P2P SIP NAT
	traversal)
In-Reply-To: <4416E0E5.4040908@research.panasonic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31D9DC0@0591-ITS-EXMP02.us.saic.com>
	<4416E0E5.4040908@research.panasonic.com>
Message-ID: <8b2769930603140936o314d0b84x8181864063812d56@mail.gmail.com>

I'll throw my support here too. I really would like to see the focus
stay narrow and limited to P2P as it applies to SIP, and keeping SIP
in the name makes that a good bit easier, from my perspective.

David

On 3/14/06, Greg Daley <gregd at research.panasonic.com> wrote:
> Hi Radhika,
>
> Let's not change the name.
>
> We're here to work on SIP related P2P only.
>
> Even if we don't change the SIP protocol,
> we still have to define operations for devices
> which perform both P2P and CS SIP.
>
> Certainly, we have presence work to do, and that's
> not just a naming issue, it's a state and protocol
> issue too (and one which has been simply, but neatly
> described by Kundan and Henning in their technical
> report.), but I don't think it needs to modify the name
> of the WG.
>
> If other P2P operations need to be performed,
> let another WG worry about it: they can use our
> protocols and tools if need be.
>
> Greg
>
> Roy, Radhika R. wrote:
> > Dean:
> >
> > The only problem that I have with your suggestion is the word: "NAMING."
> >
> > NAMING seems to be very "limited" in scope. I believe that it has also
> > another aspect that we might be looking for: "PROTOCOL."
> >
> > So, still my vote will be as follows:
> >
> > 1. P2P-ONP (P2P Overlay Networking Protocol)
> > 2. Others
> >
> > BR/Radhika
> >
> > -----Original Message-----
> > From: Dean Willis [mailto:dean.willis at softarmor.com]
> > Sent: Tuesday, March 14, 2006 1:31 AM
> > To: Michael Slavitch
> > Cc: Roy, Radhika R.; P2P-SIP
> > Subject: Names (was Re: [p2p-sip] Problem statement for P2P SIP NAT
> > traversal)
> >
> >
> > On Mar 13, 2006, at 9:51 PM, Michael Slavitch wrote:
> >
> >> P2PONP?  Good.  Let's poll for alternatives, but I like it.
> >> Personally I like mine, but P2PONP is more IETF-ish. I don't care
> >> as long as it meets the requirements above.
> >>
> >
> > I'm struggling to find something that could use the acronym P2PONY.
> > How about P2PONI? "Peer 2 Peer Overlay Naming for Internet"  . . .
> > Maybe "Peer Overlay Naming for Internet Element Signaling", PONIES.
> > It has, well, spirit.
> >
> > A good catchy acronym nets one a world of marketing power.
> >
> > But we do have to be very careful.
> >
> > If we stay laser-focused on applying P2P techniques to solve the
> > rendezvous problem for SIP, we have a solvable problem. If we
> > generalize, we could be in a world of hurt. The lesson here -- don't
> > generalize.
> >
> > SIP does Offer/Answer. I think that narrows our focus for media
> > negotiation to O/A. But that's not the same thing as correspondent
> > discovery, which does not have to be an O/A problem.
> >
> > Without generalizing we still have a couple of Big Questions:
> >
> > 1) Is the Insert/Search method SIP or some other protocol, and if
> > some other, then what?
> >
> > 2) If we make our supernodes do STUN/TURN, does ICE solve our NAT
> > traversal problem?
> >
> > 3) How do we solve routing topology security in a way that makes EKR
> > happy and still works? I suppose one solution is to put all the
> > supernodes into the trust domain and have all the user functions on
> > ordinary nodes, but I suspect that isn't an optimal model for
> > distribution of labor. Might work OK for the "self organizing proxy
> > farm" use case, however.
> >
> > --
> > Dean
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From dbarrett at quinthar.com  Tue Mar 14 16:20:13 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Tue, 14 Mar 2006 13:20:13 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <200603142120.k2ELKUOZ001289@cs.columbia.edu>

Ok, so if I understand this discussion:

1) The only problems we're trying to solve are decentralizing the SIP proxy
and STUN/TURN services.  Anything that doesn't relate to these problems is
out of scope.  (Specifically, the whole topic of NAT traversal between UAs
for media streams is entirely out of scope.)

2) While solving the problems in (1), we can make no significant change to
SIP itself.  In other words, our solution must seamlessly interoperate with
non-P2P SIP clients (for some definition of 'seamless').

3) We must craft our solution using the best of breeds techniques available
from the P2P community.  Thus we're focused on the *engineering* problem,
not the *research* problem.  (Specifically, problems that haven't been
clearly solved by the research community won't be tackled by us, including
many desirable security topics.)

Is this about right?

If so, I like the proposal of just focusing all our energy on creating a
top-notch overlay layer, and then documenting the "best practices" and
bindings in order to use this layer for SIP addressing, ICE resource
location, etc.  Thus we have a clean separation between our overlay service,
and the SIP service.

Unfortunately, because we're barred from modifying existing protocols
(beyond specific bindings, such as address types), that leaves us no choice
but to roll our own entirely new protocol from scratch.  Clearly we can
choose a model, or even standardize one that is already in use, but it's
still a bigger task than just piggypacking atop of SIP.  

But on the upside, it gives us a lot of flexibility, and simplifies the
problem space.  It nicely cleaves the work into:
- Creating an overlay service with a clean external interface
- Documenting how to use this overlay service by SIP and ICE

Does this sound about right?

-david


From RBrown at talkswitch.com  Tue Mar 14 16:43:46 2006
From: RBrown at talkswitch.com (Rob Brown)
Date: Tue, 14 Mar 2006 16:43:46 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <FA4B9088F1833A4FAAAA5BBB3A20A43B1BFC60@omega.ctpdomain.ctp>

Good summary.  The scope and goals are useful and achievable.

Regards,
Robert Brown

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David Barrett
Sent: Tuesday, March 14, 2006 4:20 PM
To: 'P2P-SIP'
Subject: [p2p-sip] SIP, ICE, and an Overlay Network

Ok, so if I understand this discussion:

1) The only problems we're trying to solve are decentralizing the SIP
proxy and STUN/TURN services.  Anything that doesn't relate to these
problems is out of scope.  (Specifically, the whole topic of NAT
traversal between UAs for media streams is entirely out of scope.)

2) While solving the problems in (1), we can make no significant change
to SIP itself.  In other words, our solution must seamlessly
interoperate with non-P2P SIP clients (for some definition of
'seamless').

3) We must craft our solution using the best of breeds techniques
available from the P2P community.  Thus we're focused on the
*engineering* problem, not the *research* problem.  (Specifically,
problems that haven't been clearly solved by the research community
won't be tackled by us, including many desirable security topics.)

Is this about right?

If so, I like the proposal of just focusing all our energy on creating a
top-notch overlay layer, and then documenting the "best practices" and
bindings in order to use this layer for SIP addressing, ICE resource
location, etc.  Thus we have a clean separation between our overlay
service, and the SIP service.

Unfortunately, because we're barred from modifying existing protocols
(beyond specific bindings, such as address types), that leaves us no
choice but to roll our own entirely new protocol from scratch.  Clearly
we can choose a model, or even standardize one that is already in use,
but it's still a bigger task than just piggypacking atop of SIP.  

But on the upside, it gives us a lot of flexibility, and simplifies the
problem space.  It nicely cleaves the work into:
- Creating an overlay service with a clean external interface
- Documenting how to use this overlay service by SIP and ICE

Does this sound about right?

-david

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From dean.willis at softarmor.com  Tue Mar 14 16:45:19 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 14 Mar 2006 15:45:19 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603142120.k2ELKUOZ001289@cs.columbia.edu>
References: <200603142120.k2ELKUOZ001289@cs.columbia.edu>
Message-ID: <E9887523-03B8-408B-9303-E6AA1CD8A81C@softarmor.com>


On Mar 14, 2006, at 3:20 PM, David Barrett wrote:

> 1) The only problems we're trying to solve are decentralizing the  
> SIP proxy
> and STUN/TURN services.  Anything that doesn't relate to these  
> problems is
> out of scope.  (Specifically, the whole topic of NAT traversal  
> between UAs
> for media streams is entirely out of scope.)

maybe I'm confused, but I see "STUN/TURN services" as key components  
of "NAT traversal between UAs for media streams".

--
Dean


From dbarrett at quinthar.com  Tue Mar 14 18:11:06 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Tue, 14 Mar 2006 15:11:06 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <E9887523-03B8-408B-9303-E6AA1CD8A81C@softarmor.com>
Message-ID: <200603142314.k2ENBCOZ027407@cs.columbia.edu>

> -----Original Message-----
> From: Dean Willis [mailto:dean.willis at softarmor.com]
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> On Mar 14, 2006, at 3:20 PM, David Barrett wrote:
> 
> > 1) The only problems we're trying to solve are decentralizing the
> > SIP proxy
> > and STUN/TURN services.  Anything that doesn't relate to these
> > problems is
> > out of scope.  (Specifically, the whole topic of NAT traversal
> > between UAs
> > for media streams is entirely out of scope.)
> 
> maybe I'm confused, but I see "STUN/TURN services" as key components
> of "NAT traversal between UAs for media streams".

Ah, sorry, I merely meant to mean we'd provide a decentralized directory of
STUN/TURN servers.  Thus our overlay network doesn't "understand" STUN any
more than your hard drive "understands" a Word doc.  Both are just storage
mechanisms that are used by other services.

-david



From dbarrett at quinthar.com  Tue Mar 14 19:35:08 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Tue, 14 Mar 2006 16:35:08 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603142314.k2ENBCOZ027407@cs.columbia.edu>
Message-ID: <200603150035.k2F0ZHOZ013079@cs.columbia.edu>

> -----Original Message-----
> From: David Barrett
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> > -----Original Message-----
> > From: Dean Willis [mailto:dean.willis at softarmor.com]
> > Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> >
> > On Mar 14, 2006, at 3:20 PM, David Barrett wrote:
> >
> > > 1) The only problems we're trying to solve are decentralizing the
> > > SIP proxy
> > > and STUN/TURN services.  Anything that doesn't relate to these
> > > problems is
> > > out of scope.  (Specifically, the whole topic of NAT traversal
> > > between UAs
> > > for media streams is entirely out of scope.)
> >
> > maybe I'm confused, but I see "STUN/TURN services" as key components
> > of "NAT traversal between UAs for media streams".
> 
> Ah, sorry, I merely meant to mean we'd provide a decentralized directory
> of
> STUN/TURN servers.  Thus our overlay network doesn't "understand" STUN any
> more than your hard drive "understands" a Word doc.  Both are just storage
> mechanisms that are used by other services.

Oh, I should also add that in this proposal, the overlay network doesn't
"understand" SIP either.  It's just a service that happens to work really
well with SIP.  Furthermore, SIP doesn't "understand" the overlay, it just
happens to work really well with it.

The hairy junction would be in what I'm calling the 'bindings', which is the
glue code that converts SIP requests/responses into a series of overlay
requests/responses, and back.  For example, the SIP binding would define how
a REGISTER command is converted into a series of GETs and PUTs into the
overlay, and vice versa.  The ICE binding would specify how a node uses the
overlay to both register and locate TURN and STUN servers.

In theory there could be other bindings as well (presence information, for
example), but those aren't the focus.

Is this what we have in mind?

-david



From lowekamp at cs.wm.edu  Tue Mar 14 19:52:33 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 14 Mar 2006 19:52:33 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603150035.k2F0ZHOZ013079@cs.columbia.edu>
References: <200603142314.k2ENBCOZ027407@cs.columbia.edu>
	<200603150035.k2F0ZHOZ013079@cs.columbia.edu>
Message-ID: <e9132f820603141652k1b0342ak100c173964ceae49@mail.gmail.com>

On 3/14/06, David Barrett <dbarrett at quinthar.com> wrote:
> Oh, I should also add that in this proposal, the overlay network doesn't
> "understand" SIP either.  It's just a service that happens to work really
> well with SIP.  Furthermore, SIP doesn't "understand" the overlay, it just
> happens to work really well with it.
>
> The hairy junction would be in what I'm calling the 'bindings', which is the
> glue code that converts SIP requests/responses into a series of overlay
> requests/responses, and back.  For example, the SIP binding would define how
> a REGISTER command is converted into a series of GETs and PUTs into the
> overlay, and vice versa.  The ICE binding would specify how a node uses the
> overlay to both register and locate TURN and STUN servers.
>
> In theory there could be other bindings as well (presence information, for
> example), but those aren't the focus.
>
> Is this what we have in mind?

I think there are a number of competing ideas on this point.  It's one
of the major issues we need to evaluate and address in a WG.  My
personal opinion (as you can probably guess from
draft-bryan-sipping-p2p) is that the overlay maintenance can, in fact,
benefit from being based on SIP.   The primary reason for is that if
we're targeting SIP devices, they already have a SIP stack, with all
the associated benefits of the work that's gone into that addressing
various challenges.  If we build an entirely new protocol for overlay
maintenance from scratch, that's a lot of work.  It will take time and
will have a lot of problems that SIP has already addressed  Our SIP
devices already have that protocol, so it makes sense to build on top
of that.  Our draft tries to do that with as few modifications to SIP
as possible (a few new headers and parameters, essentially),
essentially trying to see what can be done with this approach.  The
new version adds a cleaner separation between "p2p" operations and
"cs" operations (so that one never jumps to the other type in a
redirect of the same transaction, for example).

Some people are bothered by our SIP extensions and would prefer to
introduce an entirely new method rather than use REGISTER. 
Personally, I think REGISTER makes sense here, but it's an interesting
question.  Other people don't want to see any new headers introduced
in SIP at all, and therefore want an entirely new protocol.  I don't
see the need for that, personally, but they can probably argue quite
correctly that SIP wasn't originally intended to have some of the
semantics we're giving it.  It's also asking for an awful lot of work,
vs something that we can build on and get running quickly.

The point is that we all want to support P2P resource registration,
discovery, and session setup for SIP endpoints.   We need to form the
WG around those goals.  Then we need to evaluate what the best path is
to get there.

Bruce


From Markus.Isomaki at nokia.com  Tue Mar 14 20:05:42 2006
From: Markus.Isomaki at nokia.com (Markus.Isomaki@nokia.com)
Date: Wed, 15 Mar 2006 03:05:42 +0200
Subject: [p2p-sip] Problem statement for P2P SIP NAT traversal
In-Reply-To: <4416E17A.1010800@nokia.com>
Message-ID: <C84E0A4ABA6DD74DA5221E0833A35DF305732F5F@esebe101.NOE.Nokia.com>

Hi,

I agree. IETF prefers narrow and focused charters, which is good
engineering, but often it means that some important components of the
overall solution fall through the cracks, and the real interoperable
deployments get delayed for a number of years. In this case I believe
that for P2P-SIP to be usable in practice, also the discovery and
allocation of STUN and TURN servers must belong to the default package.

ICE framework itself is very P2P oriented, so I think it should be
usable as such.

Markus
 

>-----Original Message-----
>From: p2p-sip-bounces at cs.columbia.edu 
>[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of ext Aki Niemi
>Sent: 14 March, 2006 17:30
>To: ext Jonathan Rosenberg
>Cc: P2P-SIP
>Subject: Re: [p2p-sip] Problem statement for P2P SIP NAT traversal
>
>
>
>ext Jonathan Rosenberg wrote:
>> If you want p2p-sip to 'just work', I suggest that time is better 
>> spent figuring out how to eliminate the need to configure 
>the servers 
>> that ICE depends on - turn and stun servers - and figure out how to 
>> dynamically find ones amongst peers in the mesh.
>
>Exactly.
>
>However, last time I looked, this was buried in the charter 
>text as something to work on "later". I think it should be 
>first priority. In fact, I'd hazard a guess that running media 
>relays (and not the SIP
>rendezvous) is the highest costing resource that most 
>desperately needs to be distributed in this system.
>
>Cheers,
>Aki
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From dbarrett at quinthar.com  Tue Mar 14 20:32:55 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Tue, 14 Mar 2006 17:32:55 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <e9132f820603141652k1b0342ak100c173964ceae49@mail.gmail.com>
Message-ID: <200603150133.k2F1X1OZ025330@cs.columbia.edu>

> -----Original Message-----
> From: Bruce Lowekamp
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> I think there are a number of competing ideas on this point.  It's one
> of the major issues we need to evaluate and address in a WG.  My
> personal opinion (as you can probably guess from
> draft-bryan-sipping-p2p) is that the overlay maintenance can, in fact,
> benefit from being based on SIP.   

Hi Bruce.  Actually, I like this plan.  I think SIP would be better off with
distributed resource location and NAT traversal built in.  I totally agree
these could be accomplished with minor, localized improvements to SIP
itself.  And I think it could further be done in a way that doesn't break
interoperation with non-extended clients.

But I've heard a lot of opposition to this point.  ICE has already set the
precedent of creating a separate system to handle the separate problem.
I've made it very clear I disagree with this precedent, but I'm willing to
accept it if that's what it takes to move forward.

So I'm fine with either building distributed resource management and NAT
traversal:

1) Into the SIP base with minor changes
2) Into a separate layer as a totally new protocol

The same reasoning that delegated NAT traversal to ICE is what keeps
resource management out of SIP.  I'd *love* to reconsider that reasoning.
But if not, I'd like to move forward.

-david



From karstbj at yahoo.com  Wed Mar 15 00:51:39 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Tue, 14 Mar 2006 21:51:39 -0800 (PST)
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603150133.k2F1X1OZ025330@cs.columbia.edu>
Message-ID: <20060315055139.71440.qmail@web53011.mail.yahoo.com>

> But I've heard a lot of opposition to this point. 
> ICE has already set the
> precedent of creating a separate system to handle
> the separate problem.
> I've made it very clear I disagree with this
> precedent, but I'm willing to
> accept it if that's what it takes to move forward.

ICE is a proven solution for establishing media
streams, but P2P SIP needs more. In classic SIP, the
end points can exchange signalling through proxies.
The proxies themselves don't have to traverse NAT. In
P2P SIP, the end points have to be able to send their
signalling messages directly, crossing NAT if needed.
I don't get how that can be done without adding some
NAT traversal capability to SIP itself. 

-- Karst





__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From dbarrett at quinthar.com  Wed Mar 15 03:06:01 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Wed, 15 Mar 2006 00:06:01 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
Message-ID: <200603150806.k2F86DOZ010500@cs.columbia.edu>

> -----Original Message-----
> From: Karst Bjorgson [mailto:karstbj at yahoo.com]
> Sent: Tuesday, March 14, 2006 9:52 PM
> To: David Barrett; 'Bruce Lowekamp'
> Cc: 'P2P-SIP'
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> > But I've heard a lot of opposition to this point.
> > ICE has already set the
> > precedent of creating a separate system to handle
> > the separate problem.
> > I've made it very clear I disagree with this
> > precedent, but I'm willing to
> > accept it if that's what it takes to move forward.
> 
> ICE is a proven solution for establishing media
> streams, but P2P SIP needs more. In classic SIP, the
> end points can exchange signalling through proxies.
> The proxies themselves don't have to traverse NAT. In
> P2P SIP, the end points have to be able to send their
> signalling messages directly, crossing NAT if needed.
> I don't get how that can be done without adding some
> NAT traversal capability to SIP itself.

I totally agree -- NAT traversal is a natural fit into the baseline SIP
syntax.  I think it can be done with only very minor changes.  The outline
of my proposal is here: (with the summary plucked out)

https://lists.cs.columbia.edu/pipermail/p2p-sip/2006-January/000289.html

> In summary, I suggest:
> - SIP be expanded with a STUN command that proxies MUST implement
> - User agents REGISTER multiple (ip:port) tuples with the proxy
> - Proxies respond "300 Multiple Choices" to INVITE, to which UA's
>   holepunch
> - Proxies append all of A's (ip:port) tuples when forwarding to B
> - B replies "200 OK" direct to all of A's (ip:port) tuples

In essence, every P2P-SIP proxy becomes a rendezvous service (and
non-P2P-SIP proxies work as before).  And a distributed P2P-SIP proxy
naturally becomes a distributed rendezvous service.  

Is this (or something along these lines) what you had in mind?

(Incidentally, ICE could still play its standard role in negotiating media
streams; this above solution only sets up the SIP signaling stream.)

-david


From jh at tutpro.com  Wed Mar 15 03:49:50 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Wed, 15 Mar 2006 10:49:50 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603150806.k2F86DOZ010500@cs.columbia.edu>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
Message-ID: <17431.54574.497221.987252@rautu.tutpro.com>

David Barrett writes:

 > > - SIP be expanded with a STUN command that proxies MUST implement

last time i checked and tried, STUN didn't work if SIP UA was behind
symmetric NAT.  since most NATs are symmetric, i found STUN to have very
little value and we turned the service off.  has something changed?

-- juha

From aki.niemi at nokia.com  Wed Mar 15 03:57:24 2006
From: aki.niemi at nokia.com (Aki Niemi)
Date: Wed, 15 Mar 2006 10:57:24 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17431.54574.497221.987252@rautu.tutpro.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
Message-ID: <4417D6F4.2080609@nokia.com>


ext Juha Heinanen wrote:
> David Barrett writes:
> 
>  > > - SIP be expanded with a STUN command that proxies MUST implement
> 
> last time i checked and tried, STUN didn't work if SIP UA was behind
> symmetric NAT.  since most NATs are symmetric, i found STUN to have very
> little value and we turned the service off.  has something changed?

I think it still has value in SIP, as in being able to register the
public IP/port instead of the private ones. Of course, with symmetric
NATs, redirects won't work, but forwarding ought to work.

However, aren't we talking about RFC 3581 here?

Cheers,
Aki

From jh at tutpro.com  Wed Mar 15 04:08:33 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Wed, 15 Mar 2006 11:08:33 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <4417D6F4.2080609@nokia.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
	<4417D6F4.2080609@nokia.com>
Message-ID: <17431.55697.306070.831624@rautu.tutpro.com>

Aki Niemi writes:

 > I think it still has value in SIP, as in being able to register the
 > public IP/port instead of the private ones. 

any decent proxy can detect if a UA is behind NAT and register
automatically both public and private addresses/ports.

-- juha

From dbarrett at quinthar.com  Wed Mar 15 04:09:04 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Wed, 15 Mar 2006 01:09:04 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17431.54574.497221.987252@rautu.tutpro.com>
Message-ID: <200603150909.k2F997OZ026285@cs.columbia.edu>

> -----Original Message-----
> From: Juha Heinanen [mailto:jh at tutpro.com]
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> David Barrett writes:
> 
>  > > - SIP be expanded with a STUN command that proxies MUST implement
> 
> last time i checked and tried, STUN didn't work if SIP UA was behind
> symmetric NAT.  since most NATs are symmetric, i found STUN to have very
> little value and we turned the service off.  has something changed?

Well, it partially works.  STUN will still correctly return the external
mapping.  The problem is that mapping will only be valid for that one STUN
server -- nobody else can 'call back' through that hole in the NAT.
However, it does give us a sense of what the 'latest' mapping is, and
thereby give us a starting point for port prediction.  Taken all together,
assuming one or both of us are behind symmetric NATs:

1) You and I connect to the proxy
2) I pick a random STUN server (proxy) and figure out my latest mapping
3) I send INVITE to the proxy, including my latest mapping
4) The proxy replies with a list of your last known mapping
5) The proxy forwards my INVITE to you with my latest mapping
6) You pick a random STUN server (proxy) and figure out your latest mapping
7) You reply to the proxy's INVITE indicating your latest mapping
8) The proxy returns your latest mapping back to me
  --- At this point we both have a good guess of the other's mapping ---
9) I start scanning all your mappings (and predict some more)
10) You start scanning all my mappings (and predict some more)
11) Eventually our scans overlap and we establish a direct connection

It's not a pretty process, and it's not guaranteed to work (that's what TURN
is for), but it's been made to work with extremely high success rates in the
real world.  Basically, the proxy accumulates a list of (ip:port)
combinations that it thinks you can get contacted at.  If you're on a
non-symmetric NAT, this list will never grow.  If you're on a symmetric NAT,
we'll constantly discard the old entries and replace it with the latest.
Regardless, it's the same algorithm for both symmetric and non-symmetric
NATs.  (Though by guessing the NAT type we can optimize even further.)

The key, of course, is having a large pool of STUN servers to draw from,
because you'll need to re-STUN before each new connection.  Otherwise you'll
lose track of your 'latest' mapping and make port prediction (which, in its
most basic form, is just linearly trying a range of N ports ahead of the
remote user's 'latest') work poorly.

-david


From slavitch at objectworld.com  Wed Mar 15 09:51:16 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Wed, 15 Mar 2006 09:51:16 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C780@photon.objectworld.com>

What is the IETF standard for that? 


Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Juha Heinanen
Sent: March 15, 2006 4:09 AM
To: Aki Niemi
Cc: 'P2P-SIP'
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network

Aki Niemi writes:

 > I think it still has value in SIP, as in being able to register the
> public IP/port instead of the private ones. 

any decent proxy can detect if a UA is behind NAT and register
automatically both public and private addresses/ports.

-- juha
_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


--


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From dean.willis at softarmor.com  Wed Mar 15 16:38:35 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Wed, 15 Mar 2006 15:38:35 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603142311.k2ENBax5014890@nylon.softarmor.com>
References: <200603142311.k2ENBax5014890@nylon.softarmor.com>
Message-ID: <3C88084B-A84F-4E58-9EBA-A5F19FC8E0BF@softarmor.com>


On Mar 14, 2006, at 5:11 PM, David Barrett wrote:

>> -----Original Message-----
>> From: Dean Willis [mailto:dean.willis at softarmor.com]
>> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
>>
>> On Mar 14, 2006, at 3:20 PM, David Barrett wrote:
>>
>>> 1) The only problems we're trying to solve are decentralizing the
>>> SIP proxy
>>> and STUN/TURN services.  Anything that doesn't relate to these
>>> problems is
>>> out of scope.  (Specifically, the whole topic of NAT traversal
>>> between UAs
>>> for media streams is entirely out of scope.)
>>
>> maybe I'm confused, but I see "STUN/TURN services" as key components
>> of "NAT traversal between UAs for media streams".
>
> Ah, sorry, I merely meant to mean we'd provide a decentralized  
> directory of
> STUN/TURN servers.  Thus our overlay network doesn't "understand"  
> STUN any
> more than your hard drive "understands" a Word doc.  Both are just  
> storage
> mechanisms that are used by other services.

Ah! I understand. Yes, this sounds good.

--
Dean



From dean.willis at softarmor.com  Wed Mar 15 16:45:58 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Wed, 15 Mar 2006 15:45:58 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17431.54574.497221.987252@rautu.tutpro.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
Message-ID: <76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>


On Mar 15, 2006, at 2:49 AM, Juha Heinanen wrote:

> David Barrett writes:
>
>>> - SIP be expanded with a STUN command that proxies MUST implement
>
> last time i checked and tried, STUN didn't work if SIP UA was behind
> symmetric NAT.  since most NATs are symmetric, i found STUN to have  
> very
> little value and we turned the service off.  has something changed?
>

Yes. You'd also add a TURN component (see latest draft) into your  
proxy, and voila, symmetric NAT traversal.

In my P2P-SIP fantasy world, all the supernodes provide STUN/TURN  
services. They MAY also provide "SIP proxy" interfaces that adapt  
regular SIP client-nodes to the overlay network.

Another model moves the STUN/TURN services (and maybe the proxies)  
into a layer of extraordinary but not super (well-connected but lack  
full route participation) nodes, each of which provide named resource 
(s) in the overlay.

--
Dean




From dean.willis at softarmor.com  Wed Mar 15 16:49:49 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Wed, 15 Mar 2006 15:49:49 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603150909.k2F997OZ026285@cs.columbia.edu>
References: <200603150909.k2F997OZ026285@cs.columbia.edu>
Message-ID: <65669CF3-FEF8-4A8C-8352-2AA919BC7D8A@softarmor.com>


Almost completely off-topic here, but if ICE incorporated your port- 
prediction approach in between the
"stun" and "turn" phases, would it satisfy your requirements?

--
Dean


On Mar 15, 2006, at 3:09 AM, David Barrett wrote:
>
> Well, it partially works.  STUN will still correctly return the  
> external
> mapping.  The problem is that mapping will only be valid for that  
> one STUN
> server -- nobody else can 'call back' through that hole in the NAT.
> However, it does give us a sense of what the 'latest' mapping is, and
> thereby give us a starting point for port prediction.  Taken all  
> together,
> assuming one or both of us are behind symmetric NATs:
>
> 1) You and I connect to the proxy
> 2) I pick a random STUN server (proxy) and figure out my latest  
> mapping
> 3) I send INVITE to the proxy, including my latest mapping
> 4) The proxy replies with a list of your last known mapping
> 5) The proxy forwards my INVITE to you with my latest mapping
> 6) You pick a random STUN server (proxy) and figure out your latest  
> mapping
> 7) You reply to the proxy's INVITE indicating your latest mapping
> 8) The proxy returns your latest mapping back to me
>   --- At this point we both have a good guess of the other's  
> mapping ---
> 9) I start scanning all your mappings (and predict some more)
> 10) You start scanning all my mappings (and predict some more)
> 11) Eventually our scans overlap and we establish a direct connection
>
> It's not a pretty process, and it's not guaranteed to work (that's  
> what TURN
> is for), but it's been made to work with extremely high success  
> rates in the
> real world.  Basically, the proxy accumulates a list of (ip:port)
> combinations that it thinks you can get contacted at.  If you're on a
> non-symmetric NAT, this list will never grow.  If you're on a  
> symmetric NAT,
> we'll constantly discard the old entries and replace it with the  
> latest.
> Regardless, it's the same algorithm for both symmetric and non- 
> symmetric
> NATs.  (Though by guessing the NAT type we can optimize even further.)
>
> The key, of course, is having a large pool of STUN servers to draw  
> from,
> because you'll need to re-STUN before each new connection.   
> Otherwise you'll
> lose track of your 'latest' mapping and make port prediction  
> (which, in its
> most basic form, is just linearly trying a range of N ports ahead  
> of the
> remote user's 'latest') work poorly.
>
> -david
>
>


From jh at tutpro.com  Wed Mar 15 17:18:18 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 00:18:18 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
	<76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
Message-ID: <17432.37546.401236.854304@rautu.tutpro.com>

Dean Willis writes:

 > Yes. You'd also add a TURN component (see latest draft) into your  
 > proxy, and voila, symmetric NAT traversal.

dean,

today there are large number (perhaps hundreds) of open source
ser/openser SIP proxies deployed today that achieve NAT traversal
without any need for STUN in the UAs or in the network.  

they automatically detect from via and contact headers and source IP
addresses/ports if an UA is behind any kind of NAT and if so, start
sending requests and replies to its public IP address.

since this is currently working well (equal to Skype), what added values
(complexity does not count) does STUN offer here?

-- juha

From dean.willis at softarmor.com  Wed Mar 15 17:57:39 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Wed, 15 Mar 2006 16:57:39 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17432.37546.401236.854304@rautu.tutpro.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
	<76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
	<17432.37546.401236.854304@rautu.tutpro.com>
Message-ID: <111BADDF-2135-4623-BEC6-070DB1EA0300@softarmor.com>

On Mar 15, 2006, at 4:18 PM, Juha Heinanen wrote:

> Dean Willis writes:
>
>> Yes. You'd also add a TURN component (see latest draft) into your
>> proxy, and voila, symmetric NAT traversal.
>
> dean,
>
> today there are large number (perhaps hundreds) of open source
> ser/openser SIP proxies deployed today that achieve NAT traversal
> without any need for STUN in the UAs or in the network.
>
> they automatically detect from via and contact headers and source IP
> addresses/ports if an UA is behind any kind of NAT and if so, start
> sending requests and replies to its public IP address.
>
> since this is currently working well (equal to Skype), what added  
> values
> (complexity does not count) does STUN offer here?

Try talking to a college-boy who's behind a double-NAT, which is  
happening a lot now at universities (one at the University level, one  
for the dorm room), and I'll bet you find Skype works a lot better.

Off-topic for P2P, perhaps, but we've found any number of deployments  
(I've been supporting UA design for Vonage, AT&T, and other  
deployments that use some Linksys gear) where this doesn't work all  
that well. The main issue is keepalive, and STUN (in combination with  
the "outbound" draft) gives us a much better answer for keepalive  
than sending a REGISTER request every 20 seconds does.

That said, what you're using in SER is essentially a mini-STUN. STUN  
uses the same fundamental NAT detection technique, gives feedback  
also to the UA about its NATness, and generally just solves a few  
more cases, while maintaining good efficiency levels.

This stuff is finally starting to come together into a generally- 
workable solution. No perfect, but much better than the amalgamated  
glop that has hampered broad-scale deployments for the the pas few  
years.

--
Dean 

From jdrosen at cisco.com  Wed Mar 15 17:59:02 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Wed, 15 Mar 2006 17:59:02 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <200603150909.k2F997OZ026285@cs.columbia.edu>
References: <200603150909.k2F997OZ026285@cs.columbia.edu>
Message-ID: <44189C36.8000706@cisco.com>

There is a lot of misinformation floating around on these threads about 
how traversal for signaling itself works.

Please read:
http://www.ietf.org/internet-drafts/draft-ietf-sip-outbound-02.txt

which deals with this problem.

The net/net is that it allows clients to register to the proxy their 
current TCP connection and/or UDP 5-tuple, and to use keepalives to keep 
the bindings open for incoming calls. The mechanism works through 
symmetric nat as well. Note that it uses STUN messages for the 
keepalives themselves. Please do not confuse this usage of stun with 
'classic' stun in rfc3489 where the stun messages are sent to a separate 
stun server for discovery of nat type and learning of bindings. Here, 
they are used to keep bindings alive and to detect change in bindings. 
The notion of differing stun usages reflects the reality of its 
deployment over the years and is described in rfc3489bis.

Now, in terms of p2p nat traversal for SIP itself, the sip-outbound 
mechanism works just fine for p2p sip as well. The difference is that 
the proxy is just another host in the mesh. Of course, it better be one 
on the public network so everyone can reach it, else you will not be 
able to get incoming calls from everyone.

Now, the only issue as I see it, is that the current sip-outbound 
mechanism requires that proxy to record-route, and to act in proxy mode. 
I can imagine that, for p2p-sip, there would be a desire to not have it 
record-route. I see nothing gained from only using redirection for the 
initial invite; if the client is keeping a binding to the proxy open 
then may as well use that for the initial rendezvous.

To eliminate record-routing and go direct if you can, that is something 
that we can't do with today's sip specs that would be a worthy 
extension. The gruu spec postulated the need for such a thing in the 
future, in fact.

One solution there is that the 200 OK from the UAS include multiple 
contact values, each representing a potentially reachable IP address and 
  a stun username/password. You then, ICE_like, check connectivity to 
each of those, and whichever one works, you use that as the contact for 
direct mid-dialog signaling. If none work, you use the proxy for 
mid-dialog requests too.

-Jonathan R.

David Barrett wrote:

>>-----Original Message-----
>>From: Juha Heinanen [mailto:jh at tutpro.com]
>>Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
>>
>>David Barrett writes:
>>
>> > > - SIP be expanded with a STUN command that proxies MUST implement
>>
>>last time i checked and tried, STUN didn't work if SIP UA was behind
>>symmetric NAT.  since most NATs are symmetric, i found STUN to have very
>>little value and we turned the service off.  has something changed?
> 
> 
> Well, it partially works.  STUN will still correctly return the external
> mapping.  The problem is that mapping will only be valid for that one STUN
> server -- nobody else can 'call back' through that hole in the NAT.
> However, it does give us a sense of what the 'latest' mapping is, and
> thereby give us a starting point for port prediction.  Taken all together,
> assuming one or both of us are behind symmetric NATs:
> 
> 1) You and I connect to the proxy
> 2) I pick a random STUN server (proxy) and figure out my latest mapping
> 3) I send INVITE to the proxy, including my latest mapping
> 4) The proxy replies with a list of your last known mapping
> 5) The proxy forwards my INVITE to you with my latest mapping
> 6) You pick a random STUN server (proxy) and figure out your latest mapping
> 7) You reply to the proxy's INVITE indicating your latest mapping
> 8) The proxy returns your latest mapping back to me
>   --- At this point we both have a good guess of the other's mapping ---
> 9) I start scanning all your mappings (and predict some more)
> 10) You start scanning all my mappings (and predict some more)
> 11) Eventually our scans overlap and we establish a direct connection
> 
> It's not a pretty process, and it's not guaranteed to work (that's what TURN
> is for), but it's been made to work with extremely high success rates in the
> real world.  Basically, the proxy accumulates a list of (ip:port)
> combinations that it thinks you can get contacted at.  If you're on a
> non-symmetric NAT, this list will never grow.  If you're on a symmetric NAT,
> we'll constantly discard the old entries and replace it with the latest.
> Regardless, it's the same algorithm for both symmetric and non-symmetric
> NATs.  (Though by guessing the NAT type we can optimize even further.)
> 
> The key, of course, is having a large pool of STUN servers to draw from,
> because you'll need to re-STUN before each new connection.  Otherwise you'll
> lose track of your 'latest' mapping and make port prediction (which, in its
> most basic form, is just linearly trying a range of N ports ahead of the
> remote user's 'latest') work poorly.
> 
> -david
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From jdrosen at cisco.com  Wed Mar 15 18:00:47 2006
From: jdrosen at cisco.com (Jonathan Rosenberg)
Date: Wed, 15 Mar 2006 18:00:47 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17432.37546.401236.854304@rautu.tutpro.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>	<200603150806.k2F86DOZ010500@cs.columbia.edu>	<17431.54574.497221.987252@rautu.tutpro.com>	<76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
	<17432.37546.401236.854304@rautu.tutpro.com>
Message-ID: <44189C9F.9040400@cisco.com>

The sip-outbound mechanism provides a standards-based way to get sip 
signaling nat traversal to work. Today what people are doing is a hodge 
podge that sometimes doesnt work and is always expensive (frequently 
large sip message retransmits).

-Jonathan R.

Juha Heinanen wrote:

> Dean Willis writes:
> 
>  > Yes. You'd also add a TURN component (see latest draft) into your  
>  > proxy, and voila, symmetric NAT traversal.
> 
> dean,
> 
> today there are large number (perhaps hundreds) of open source
> ser/openser SIP proxies deployed today that achieve NAT traversal
> without any need for STUN in the UAs or in the network.  
> 
> they automatically detect from via and contact headers and source IP
> addresses/ports if an UA is behind any kind of NAT and if so, start
> sending requests and replies to its public IP address.
> 
> since this is currently working well (equal to Skype), what added values
> (complexity does not count) does STUN offer here?
> 
> -- juha
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 

-- 
Jonathan D. Rosenberg, Ph.D.                   600 Lanidex Plaza
Cisco Fellow                                   Parsippany, NJ 07054-2711
Cisco Systems
jdrosen at cisco.com                              FAX:   (973) 952-5050
http://www.jdrosen.net                         PHONE: (973) 952-5000
http://www.cisco.com

From slavitch at objectworld.com  Wed Mar 15 18:11:02 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Wed, 15 Mar 2006 18:11:02 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>

"Try talking to a college-boy who's behind a double-NAT, which is
happening a lot now at universities (one at the University level, one
for the dorm room), and I'll bet you find Skype works a lot better." 

I am at VON right now, in a conference room at the San Jose Convention
Centre, and it's got NATs and things up the yingyang. I'm at the
conference, but in the exhibit hall my employer has a live system
running our software. 
We had to work at it to get around things not under our control. It
required configuration.  Not much, but some. 
And we consider this a selling point, because our competitors have far
more complicated deployment scenarios.  
Other organizations have canned demos at the show.

Now, with Skype, I just turned it on and it worked. That speaks volumes.
This is now an industry requirement, period.

M

Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Dean Willis
Sent: March 15, 2006 2:58 PM
To: Juha Heinanen
Cc: 'P2P-SIP'
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network

On Mar 15, 2006, at 4:18 PM, Juha Heinanen wrote:

> Dean Willis writes:
>
>> Yes. You'd also add a TURN component (see latest draft) into your 
>> proxy, and voila, symmetric NAT traversal.
>
> dean,
>
> today there are large number (perhaps hundreds) of open source 
> ser/openser SIP proxies deployed today that achieve NAT traversal 
> without any need for STUN in the UAs or in the network.
>
> they automatically detect from via and contact headers and source IP 
> addresses/ports if an UA is behind any kind of NAT and if so, start 
> sending requests and replies to its public IP address.
>
> since this is currently working well (equal to Skype), what added 
> values (complexity does not count) does STUN offer here?

Try talking to a college-boy who's behind a double-NAT, which is
happening a lot now at universities (one at the University level, one
for the dorm room), and I'll bet you find Skype works a lot better.

Off-topic for P2P, perhaps, but we've found any number of deployments
(I've been supporting UA design for Vonage, AT&T, and other deployments
that use some Linksys gear) where this doesn't work all that well. The
main issue is keepalive, and STUN (in combination with the "outbound"
draft) gives us a much better answer for keepalive than sending a
REGISTER request every 20 seconds does.

That said, what you're using in SER is essentially a mini-STUN. STUN
uses the same fundamental NAT detection technique, gives feedback also
to the UA about its NATness, and generally just solves a few more cases,
while maintaining good efficiency levels.

This stuff is finally starting to come together into a generally-
workable solution. No perfect, but much better than the amalgamated glop
that has hampered broad-scale deployments for the the pas few years.

--
Dean
_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


-- 


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From jh at tutpro.com  Wed Mar 15 18:11:35 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 01:11:35 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <111BADDF-2135-4623-BEC6-070DB1EA0300@softarmor.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
	<76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
	<17432.37546.401236.854304@rautu.tutpro.com>
	<111BADDF-2135-4623-BEC6-070DB1EA0300@softarmor.com>
Message-ID: <17432.40743.993867.1517@rautu.tutpro.com>

Dean Willis writes:

 > Try talking to a college-boy who's behind a double-NAT, which is  
 > happening a lot now at universities (one at the University level, one  
 > for the dorm room), and I'll bet you find Skype works a lot better.

openser NAT traversal solution works fine for to users behind double
NATs.  can you be more specific here?

 > The main issue is keepalive, and STUN (in combination with  
 > the "outbound" draft) gives us a much better answer for keepalive  
 > than sending a REGISTER request every 20 seconds does.

openser is not sending REGISTER requests as keepalives, but dummy UDP
packets.  it is a bad idea from scalability point of view to make the
proxy to send any kind of keepalives that require establishing a SIP
transaction.  it might be ok for the UA to do so, in which case it would
suffice for the proxy to simply reply statelessly.

 > That said, what you're using in SER is essentially a mini-STUN. STUN  
 > uses the same fundamental NAT detection technique, gives feedback  
 > also to the UA about its NATness, and generally just solves a few  
 > more cases, while maintaining good efficiency levels.

i agree with you here that it might be a good idea for the proxy to
somehow inform the UA that it is behind a NAT so that the UA could
itself handle the keepalives.  but for that purpose, would it simply
suffice for the proxy to a add an indication about NATing in register
reply?  then the only new thing the UA would need to do is to start
sending the keepalives.

-- juha

From jh at tutpro.com  Wed Mar 15 18:20:22 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 01:20:22 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>
Message-ID: <17432.41270.764372.593730@rautu.tutpro.com>

Michael Slavitch writes:

 > Now, with Skype, I just turned it on and it worked. That speaks volumes.
 > This is now an industry requirement, period.

you have to be more specific:  what makes Skype work in your environment
better than openser NAT traversal solution?

-- juha

From dean.willis at softarmor.com  Wed Mar 15 18:59:18 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Wed, 15 Mar 2006 17:59:18 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17432.40743.993867.1517@rautu.tutpro.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
	<76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
	<17432.37546.401236.854304@rautu.tutpro.com>
	<111BADDF-2135-4623-BEC6-070DB1EA0300@softarmor.com>
	<17432.40743.993867.1517@rautu.tutpro.com>
Message-ID: <6EAFB95A-A60A-40B6-A590-D985ED1AADB6@softarmor.com>


On Mar 15, 2006, at 5:11 PM, Juha Heinanen wrote:

> Dean Willis writes:
>
>> Try talking to a college-boy who's behind a double-NAT, which is
>> happening a lot now at universities (one at the University level, one
>> for the dorm room), and I'll bet you find Skype works a lot better.
>
> openser NAT traversal solution works fine for to users behind double
> NATs.  can you be more specific here?

Probably not, but I'm sure somebody can. I haven't gotten SER to work  
in that configuration with Linksys UAs. Maybe Im just doing it wrong,  
but my media always gets black-holed.

>
>> The main issue is keepalive, and STUN (in combination with
>> the "outbound" draft) gives us a much better answer for keepalive
>> than sending a REGISTER request every 20 seconds does.
>
> openser is not sending REGISTER requests as keepalives, but dummy UDP
> packets.  it is a bad idea from scalability point of view to make the
> proxy to send any kind of keepalives that require establishing a SIP
> transaction.  it might be ok for the UA to do so, in which case it  
> would
> suffice for the proxy to simply reply statelessly.
>

Cool. How does this detect and correct for NAT resets?

>> That said, what you're using in SER is essentially a mini-STUN. STUN
>> uses the same fundamental NAT detection technique, gives feedback
>> also to the UA about its NATness, and generally just solves a few
>> more cases, while maintaining good efficiency levels.
>
> i agree with you here that it might be a good idea for the proxy to
> somehow inform the UA that it is behind a NAT so that the UA could
> itself handle the keepalives.  but for that purpose, would it simply
> suffice for the proxy to a add an indication about NATing in register
> reply?  then the only new thing the UA would need to do is to start
> sending the keepalives.

yes, I think that could work. We're essentially moving the outbound- 
stun functionality into the REGISTER transaction. We might still have  
keepalive and reset-detection issues to work out.

--
Dean

From dean.willis at softarmor.com  Wed Mar 15 19:00:15 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Wed, 15 Mar 2006 18:00:15 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17432.41270.764372.593730@rautu.tutpro.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>
	<17432.41270.764372.593730@rautu.tutpro.com>
Message-ID: <3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>


On Mar 15, 2006, at 5:20 PM, Juha Heinanen wrote:

> Michael Slavitch writes:
>
>> Now, with Skype, I just turned it on and it worked. That speaks  
>> volumes.
>> This is now an industry requirement, period.
>
> you have to be more specific:  what makes Skype work in your  
> environment
> better than openser NAT traversal solution?

Nailed-up TCP connections and coaxial media-and-signaling, usually.

--
Dean



From slavitch at gmail.com  Wed Mar 15 19:18:13 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Wed, 15 Mar 2006 16:18:13 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>
	<17432.41270.764372.593730@rautu.tutpro.com>
	<3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>
Message-ID: <1fc32c8f0603151618n505f1f14ub380ddcdb1f4c748@mail.gmail.com>

"you have to be more specific:  what makes Skype work in your environment
better than openser NAT traversal solution?"

The people I need to talk to use Skype.  Nobody that I need to talk to
even knows openser exists, apart from a few alpha geeks, and even
those alpha geeks use Skype because it's easier, the Skype client is
better, and Skype is what their girlfriends use.

Every non-American that I meet at these VON sessions has a Skype ID.
It is ubiquitous, in wide use, and it just works, and consumers love
it.  Mine is running live streaming video right now.

A few moments ago I got to see another conference in a different room.
 Another Skype user turned on Skype Video, turned around his camera,
and I got to see full motion video of the session in a room on the
other side of the conference center:   High- resolution high frame
rate video with excellent sound (muted).  Skype figured out that both
nodes are on the same LAN:  My IP is 10.75.103.247, the other
something else, and there was no delay or jitter.  I turned my camera
around and we got to see two conferences at the same time.
 It was brilliant.

Meanwhile, the session I was in, on venture capital,  had an animated
discussion about how commoditization isn't a good investment vehicle. 
The irony was quite interesting.

In between these two sessions is a  is an exhibit hall where companies
are selling expensive hardware and software  that with difficulty will
provision IMS and carrier VOIP such that with a lot of work they may
be able to offer this service, for a fee of course.

But Skype does it now for free.

Whatever P2PSIP is, it had better be better than that.

M

On 3/15/06, Dean Willis <dean.willis at softarmor.com> wrote:
>
> On Mar 15, 2006, at 5:20 PM, Juha Heinanen wrote:
>
> > Michael Slavitch writes:
> >
> >> Now, with Skype, I just turned it on and it worked. That speaks
> >> volumes.
> >> This is now an industry requirement, period.
> >
> > you have to be more specific:  what makes Skype work in your
> > environment
> > better than openser NAT traversal solution?
>
> Nailed-up TCP connections and coaxial media-and-signaling, usually.
>
> --
> Dean
>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


--
Michael Slavitch
Ottawa, Ontario Canada


From dbarrett at quinthar.com  Wed Mar 15 20:02:22 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Wed, 15 Mar 2006 17:02:22 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <44189C36.8000706@cisco.com>
Message-ID: <200603160102.k2G12QOZ029011@cs.columbia.edu>

> -----Original Message-----
> From: Jonathan Rosenberg [mailto:jdrosen at cisco.com]
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> One solution there is that the 200 OK from the UAS include multiple
> contact values, each representing a potentially reachable IP address and
>   a stun username/password. You then, ICE_like, check connectivity to
> each of those, and whichever one works, you use that as the contact for
> direct mid-dialog signaling. If none work, you use the proxy for
> mid-dialog requests too.

I think I like this solution, but I'm not sure I understand it.  Can you
give a start-to-finish message list (REGISTER, INVITE, 200OK, etc) between
two peers and a proxy?

-david


From dbarrett at quinthar.com  Wed Mar 15 20:11:52 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Wed, 15 Mar 2006 17:11:52 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <65669CF3-FEF8-4A8C-8352-2AA919BC7D8A@softarmor.com>
Message-ID: <200603160111.k2G1BvOZ001673@cs.columbia.edu>

> -----Original Message-----
> From: Dean Willis [mailto:dean.willis at softarmor.com]
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> Almost completely off-topic here, but if ICE incorporated your port-
> prediction approach in between the
> "stun" and "turn" phases, would it satisfy your requirements?

First, it's not really "my" port prediction approach.  It's been well
documented for years, so I don't want to take credit for it.

Second, the issue with ICE isn't so much one of features, but whether those
features should be built into SIP or a supporting protocol.  ICE takes the
supporting protocol approach; I'd prefer it were in SIP itself.

At the very least, it seems straightforward to add NAT traversal for SIP
signaling into the basic SIP protocol itself (like I think is in Jonathan's
latest proposal). 

-david



From dbarrett at quinthar.com  Wed Mar 15 20:11:54 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Wed, 15 Mar 2006 17:11:54 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
Message-ID: <200603160111.k2G1BvOZ001674@cs.columbia.edu>

> -----Original Message-----
> From: Dean Willis [mailto:dean.willis at softarmor.com]
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> Yes. You'd also add a TURN component (see latest draft) into your
> proxy, and voila, symmetric NAT traversal.

Ah, I was unclear.  By 'NAT traversal' I meant the direct establishment of a
connection between two peers.  TURN is clearly the most reliable fallback
position, but it routes traffic through a central relay node -- something
that is costly for the relay, and adds latency to the connection.

-david


From jh at tutpro.com  Wed Mar 15 23:55:33 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 06:55:33 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <44189C9F.9040400@cisco.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
	<76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
	<17432.37546.401236.854304@rautu.tutpro.com>
	<44189C9F.9040400@cisco.com>
Message-ID: <17432.61381.597329.184392@rautu.tutpro.com>

Jonathan Rosenberg writes:

 > The sip-outbound mechanism provides a standards-based way to get sip 
 > signaling nat traversal to work. Today what people are doing is a hodge 
 > podge that sometimes doesnt work and is always expensive (frequently 
 > large sip message retransmits).

jonathan,

i'll read the outbound draft and if it sounds ok, i have no problem of
implementing it in our proxies.  what comes to the current NAT traversal
in openser, i don't consider it hodge-podge and i have never experience
any kind of large sip message retransmits,

-- juha

From jh at tutpro.com  Thu Mar 16 00:15:48 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 07:15:48 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <6EAFB95A-A60A-40B6-A590-D985ED1AADB6@softarmor.com>
References: <20060315055139.71440.qmail@web53011.mail.yahoo.com>
	<200603150806.k2F86DOZ010500@cs.columbia.edu>
	<17431.54574.497221.987252@rautu.tutpro.com>
	<76F4C2C6-C6CF-4A3E-90F0-602994640EC0@softarmor.com>
	<17432.37546.401236.854304@rautu.tutpro.com>
	<111BADDF-2135-4623-BEC6-070DB1EA0300@softarmor.com>
	<17432.40743.993867.1517@rautu.tutpro.com>
	<6EAFB95A-A60A-40B6-A590-D985ED1AADB6@softarmor.com>
Message-ID: <17432.62596.25964.623314@rautu.tutpro.com>

Dean Willis writes:

 > Probably not, but I'm sure somebody can. I haven't gotten SER to work  
 > in that configuration with Linksys UAs. Maybe Im just doing it wrong,  
 > but my media always gets black-holed.

you mean the ones that used to be called sipuras?  they work very well
here.  you just turn off all NAT claverness if they have any and let the
proxy handle rest.  of course you need to figure out, how to configure
your proxy correctly.

 > Cool. How does this detect and correct for NAT resets?

how often does your NAT reset itself?  if very often, buy a better NAT
or get an UPS.  it also helps to set REGISTER interval so something
reasonable, like 15 minutes or so.

 > > but for that purpose, would it simply
 > > suffice for the proxy to a add an indication about NATing in register
 > > reply?  then the only new thing the UA would need to do is to start
 > > sending the keepalives.
 > 
 > yes, I think that could work. We're essentially moving the outbound- 
 > stun functionality into the REGISTER transaction. We might still have  
 > keepalive and reset-detection issues to work out.

if that is what outbound STUN does, then yes.  if keepalives require
response from proxy, which i'm not sure about, then such a SIP keepalive
request would need to be defined.  but why would thr UA need a response?
in openser NAT traversal, keepalives are one way.

reset detection i would not worry about.  just use reasonable short
REGISTER intervals.

-- juha




From karstbj at yahoo.com  Thu Mar 16 01:31:42 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Wed, 15 Mar 2006 22:31:42 -0800 (PST)
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <44189C36.8000706@cisco.com>
Message-ID: <20060316063142.16647.qmail@web53014.mail.yahoo.com>



--- Jonathan Rosenberg <jdrosen at cisco.com> wrote:

> Now, in terms of p2p nat traversal for SIP itself,
> the sip-outbound 
> mechanism works just fine for p2p sip as well. The
> difference is that 
> the proxy is just another host in the mesh. Of
> course, it better be one 
> on the public network so everyone can reach it, else
> you will not be 
> able to get incoming calls from everyone.

I am sorry, but I don't get it. If the "P2P" in P2PSIP
has any meaning, we are speaking of a network where
there are no proxies. To establish a session, one
would:

- Learn the SIP URL of the peer;
- Use a discovery protocol of some kind, maybe a DHT,
to learn the address:proto:port tuple(s) at which the
corresponding SIP UA is reachable;
- Send SIP packets directly to one of the specified
address tuples;
- After the initial signalling exchange, establish the
media streams.

There is an obvious role for STUN, TURN, or for that
matter Teredo: provide a SIP UA with a list of
address:proto:port tuples that might be reachable by
the peers. The list will have to be obtained before
the tuples are published in the discovery service, be
it a DHT or something else. Obviously, to do that, one
will have to discover the adequate servers, or maybe
super-nodes willing to act as servers.

By the way, running a DHT system across multiple NAT
is an interesting challenge...

-- Karst

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From marc.bailly at gmail.com  Thu Mar 16 04:57:11 2006
From: marc.bailly at gmail.com (Marc Bailly)
Date: Thu, 16 Mar 2006 10:57:11 +0100
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <20060316063142.16647.qmail@web53014.mail.yahoo.com>
References: <44189C36.8000706@cisco.com>
	<20060316063142.16647.qmail@web53014.mail.yahoo.com>
Message-ID: <3bbc31e90603160157w4bb87ef0sc037ba9756ab6734@mail.gmail.com>

On 3/16/06, Karst Bjorgson <karstbj at yahoo.com> wrote:
>
>
>
> --- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>
> > Now, in terms of p2p nat traversal for SIP itself,
> > the sip-outbound
> > mechanism works just fine for p2p sip as well. The
> > difference is that
> > the proxy is just another host in the mesh. Of
> > course, it better be one
> > on the public network so everyone can reach it, else
> > you will not be
> > able to get incoming calls from everyone.
>
> I am sorry, but I don't get it. If the "P2P" in P2PSIP
> has any meaning, we are speaking of a network where
> there are no proxies.


In case where a UA is behind a NAT (and can't take part in the P2P overlay),
an ordinary node / super node architecture could be set up. A super node
would act as a proxy (with outbound support) for a set of ordinary nodes in
order to traverse NAT for SIP signaling.

Marc
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060316/a83a9f29/attachment.html

From effis at dsp.co.il  Thu Mar 16 05:59:26 2006
From: effis at dsp.co.il (Effi Shiri)
Date: Thu, 16 Mar 2006 12:59:26 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <216A12DDD3490049BA3E56BAC324F285012FAFB9@ILEXCH2003.il-prod.dspcorp.com>

 

Hi all,

Does anyone have statistic on how many home users are not behind NAT?

 

 

-----------------------------

On 3/16/06, Karst Bjorgson <karstbj at yahoo.com> wrote:



--- Jonathan Rosenberg <jdrosen at cisco.com> wrote:

> Now, in terms of p2p nat traversal for SIP itself,
> the sip-outbound
> mechanism works just fine for p2p sip as well. The 
> difference is that
> the proxy is just another host in the mesh. Of
> course, it better be one
> on the public network so everyone can reach it, else
> you will not be
> able to get incoming calls from everyone. 

I am sorry, but I don't get it. If the "P2P" in P2PSIP
has any meaning, we are speaking of a network where
there are no proxies. 


In case where a UA is behind a NAT (and can't take part in the P2P
overlay), an ordinary node / super node architecture could be set up. A
super node would act as a proxy (with outbound support) for a set of
ordinary nodes in order to traverse NAT for SIP signaling. 


Marc


______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060316/3923a397/attachment.html

From jh at tutpro.com  Thu Mar 16 06:19:15 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 13:19:15 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <1fc32c8f0603151618n505f1f14ub380ddcdb1f4c748@mail.gmail.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>
	<17432.41270.764372.593730@rautu.tutpro.com>
	<3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>
	<1fc32c8f0603151618n505f1f14ub380ddcdb1f4c748@mail.gmail.com>
Message-ID: <17433.18867.650499.329759@rautu.tutpro.com>

Michael Slavitch writes:

 > The people I need to talk to use Skype.  Nobody that I need to talk to
 > even knows openser exists, apart from a few alpha geeks, and even
 > those alpha geeks use Skype because it's easier, the Skype client is
 > better, and Skype is what their girlfriends use.

i tried to ask, why Skype's NAT traversal is technically better than
openser's.  what i have heard about Skype is that it uses very similar
technology, except that mediaproxies are distributed to customers'
computers.

 > Every non-American that I meet at these VON sessions has a Skype ID.
 > It is ubiquitous, in wide use, and it just works, and consumers love
 > it.  Mine is running live streaming video right now.

the only problem with Skype is that it is proprietary and perhaps also
patented technology.  i, and may be also IETF, believe in open systems.
for this same reason i don't use Windows either.  but if you have
already sold your soul to MicroSoft, you may as well sell it to Skype
too.

-- juha

From marc.bailly at gmail.com  Thu Mar 16 06:22:38 2006
From: marc.bailly at gmail.com (Marc Bailly)
Date: Thu, 16 Mar 2006 12:22:38 +0100
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <216A12DDD3490049BA3E56BAC324F285012FAFB9@ILEXCH2003.il-prod.dspcorp.com>
References: <216A12DDD3490049BA3E56BAC324F285012FAFB9@ILEXCH2003.il-prod.dspcorp.com>
Message-ID: <3bbc31e90603160322u1324218dja54cebf01da45a25@mail.gmail.com>

Effi,

As it was mentionned by Salman a few days ago :
- 50% of broadband US/UK MSN users behind NAT,
http://research.microsoft.com/~pablo/papers/CCR.pdf<http://research.microsoft.com/%7Epablo/papers/CCR.pdf>(CCR
Jan 2006)
- According to Nielsen ratings 2 out of 5 Americans have broadband access at
home. http://www.nielsen-netratings.com/pr/pr_050928.pdf

In France for example, I don't know the number but all ISPs now offer ADSL
boxes with a NAT function so the proportion of natted users increases every
day.

Marc

On 3/16/06, Effi Shiri <effis at dsp.co.il> wrote:
>
>
>
> Hi all,
>
> Does anyone have statistic on how many home users are not behind NAT?
>
>
>
>
>
> -----------------------------
>
> On 3/16/06, *Karst Bjorgson* < karstbj at yahoo.com> wrote:
>
>
>
> --- Jonathan Rosenberg <jdrosen at cisco.com> wrote:
>
> > Now, in terms of p2p nat traversal for SIP itself,
> > the sip-outbound
> > mechanism works just fine for p2p sip as well. The
> > difference is that
> > the proxy is just another host in the mesh. Of
> > course, it better be one
> > on the public network so everyone can reach it, else
> > you will not be
> > able to get incoming calls from everyone.
>
> I am sorry, but I don't get it. If the "P2P" in P2PSIP
> has any meaning, we are speaking of a network where
> there are no proxies.
>
>
> In case where a UA is behind a NAT (and can't take part in the P2P
> overlay), an ordinary node / super node architecture could be set up. A
> super node would act as a proxy (with outbound support) for a set of
> ordinary nodes in order to traverse NAT for SIP signaling.
>
>
> Marc
>
>
> ______________________________________________________________________
> DSP Group LTD. automatically scans all emails and attachments using
> MessageLabs Email Security System.
> ______________________________________________________________________
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>
>


--
-------------------
Marc Bailly
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060316/c990c691/attachment-0001.html

From jh at tutpro.com  Thu Mar 16 06:51:21 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 13:51:21 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <44189C36.8000706@cisco.com>
References: <200603150909.k2F997OZ026285@cs.columbia.edu>
	<44189C36.8000706@cisco.com>
Message-ID: <17433.20793.191367.804402@rautu.tutpro.com>

jonathan,

i read the outbound draft (not all glory details yet) and i was pleased
with it.  the main reason being that in addition to NAT traversal (which
openser can handle already now satisfactorily without any new
standardization), it also tackles the redundancy problem.

there is one thing though that i can't accept:

   UAs are configured with one or more SIP URIs representing the default
   outbound-proxy-set.  The specification assumes the set is determined
   via configuration but future specifications may define other
   mechanisms such as using DNS to discover this set.

we MUST have a DNS based way for the UAs to figure out the default
outbound proxy set BEFORE this spec is published as an RFC.  if not
done, there is yet another thing to be configured in the UA, which is
exactly what Skype lovers hate in SIP.

-- juha

From aki.niemi at nokia.com  Thu Mar 16 07:15:42 2006
From: aki.niemi at nokia.com (Aki Niemi)
Date: Thu, 16 Mar 2006 14:15:42 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17433.18867.650499.329759@rautu.tutpro.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>	<17432.41270.764372.593730@rautu.tutpro.com>	<3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>	<1fc32c8f0603151618n505f1f14ub380ddcdb1f4c748@mail.gmail.com>
	<17433.18867.650499.329759@rautu.tutpro.com>
Message-ID: <441956EE.3090206@nokia.com>

Hi Juha,

ext Juha Heinanen wrote:
> Michael Slavitch writes:
> 
>  > The people I need to talk to use Skype.  Nobody that I need to talk to
>  > even knows openser exists, apart from a few alpha geeks, and even
>  > those alpha geeks use Skype because it's easier, the Skype client is
>  > better, and Skype is what their girlfriends use.
> 
> i tried to ask, why Skype's NAT traversal is technically better than
> openser's.  what i have heard about Skype is that it uses very similar
> technology, except that mediaproxies are distributed to customers'
> computers.

I think the main additional thing is that Skype also does a bunch of,
say, "politically incorrect" things, like tunnel via HTTP proxies, to
port 443, etc.

Cheers,
Aki

From jh at tutpro.com  Thu Mar 16 07:42:28 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 14:42:28 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <441956EE.3090206@nokia.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>
	<17432.41270.764372.593730@rautu.tutpro.com>
	<3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>
	<1fc32c8f0603151618n505f1f14ub380ddcdb1f4c748@mail.gmail.com>
	<17433.18867.650499.329759@rautu.tutpro.com>
	<441956EE.3090206@nokia.com>
Message-ID: <17433.23860.181223.294522@rautu.tutpro.com>

Aki Niemi writes:

 > I think the main additional thing is that Skype also does a bunch of,
 > say, "politically incorrect" things, like tunnel via HTTP proxies, to
 > port 443, etc.

how about mapping SIP requests and replies into XML-RPC?

-- juha

From aki.niemi at nokia.com  Thu Mar 16 07:50:42 2006
From: aki.niemi at nokia.com (Aki Niemi)
Date: Thu, 16 Mar 2006 14:50:42 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17433.23860.181223.294522@rautu.tutpro.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>	<17432.41270.764372.593730@rautu.tutpro.com>	<3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>	<1fc32c8f0603151618n505f1f14ub380ddcdb1f4c748@mail.gmail.com>	<17433.18867.650499.329759@rautu.tutpro.com>	<441956EE.3090206@nokia.com>
	<17433.23860.181223.294522@rautu.tutpro.com>
Message-ID: <44195F22.2050401@nokia.com>


ext Juha Heinanen wrote:
> Aki Niemi writes:
> 
>  > I think the main additional thing is that Skype also does a bunch of,
>  > say, "politically incorrect" things, like tunnel via HTTP proxies, to
>  > port 443, etc.
> 
> how about mapping SIP requests and replies into XML-RPC?

Heh, yes. Or vice versa, finally defining that DO method to SIP for
carrying XMLRPC. :)

Cheers,
Aki

From jh at tutpro.com  Thu Mar 16 08:20:24 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Thu, 16 Mar 2006 15:20:24 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <44195F22.2050401@nokia.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7B2@photon.objectworld.com>
	<17432.41270.764372.593730@rautu.tutpro.com>
	<3D088E5B-3131-4744-92E0-94A965BB2693@softarmor.com>
	<1fc32c8f0603151618n505f1f14ub380ddcdb1f4c748@mail.gmail.com>
	<17433.18867.650499.329759@rautu.tutpro.com>
	<441956EE.3090206@nokia.com>
	<17433.23860.181223.294522@rautu.tutpro.com>
	<44195F22.2050401@nokia.com>
Message-ID: <17433.26136.6919.608236@rautu.tutpro.com>

Aki Niemi writes:

 > Heh, yes. Or vice versa, finally defining that DO method to SIP for
 > carrying XMLRPC. :)

i don't know about do.  what i meant is that you map INVITE to XML-RPC
function called sip.INVITE, etc.  then SIP Proxy would become an XML-RPC
server, which listens at port 80 and/or 443.

-- juha


From BeckW at t-systems.com  Thu Mar 16 08:58:58 2006
From: BeckW at t-systems.com (Beck01, Wolfgang)
Date: Thu, 16 Mar 2006 14:58:58 +0100
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <1E4CCB2441C5C0409AD8A929482A09F31BB68F@S4DE9JSAAIG.ost.t-com.de>



>  > Heh, yes. Or vice versa, finally defining that DO method 
> to SIP for  > carrying XMLRPC. :)
> 
> i don't know about do.  what i meant is that you map INVITE 
> to XML-RPC function called sip.INVITE, etc.  then SIP Proxy 
> would become an XML-RPC server, which listens at port 80 and/or 443.
> 
> -- juha
Why not REST?

GET http://example.com/INVITE/bob/


Wolfgang

From ggb at tid.es  Thu Mar 16 09:35:43 2006
From: ggb at tid.es (=?iso-8859-1?Q?Gustavo_Garc=EDa_Bernardo?=)
Date: Thu, 16 Mar 2006 15:35:43 +0100
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <1E4CCB2441C5C0409AD8A929482A09F31BB68F@S4DE9JSAAIG.ost.t-com.de>
Message-ID: <021001c64906$e7fce5a0$4345fea9@matrix2>



In my opinion, we first should define if P2P SIP should work through HTTP
proxies, only HTTPS proxies or none of them.   Not only for SIP messages,
but also for DHT and RTP messages....

As someone has said before, these are "politically incorrect" things...

In any case, I think this feature is a minor detail and doesn't change the
P2P SIP architecture which should be defined before discussing this kind of
details.

G.

> -----Mensaje original-----
> De: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-
> bounces at cs.columbia.edu] En nombre de Beck01, Wolfgang
> Enviado el: jueves, 16 de marzo de 2006 14:59
> Para: jh at tutpro.com; aki.niemi at nokia.com
> CC: p2p-sip at cs.columbia.edu
> Asunto: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> 
> 
> >  > Heh, yes. Or vice versa, finally defining that DO method
> > to SIP for  > carrying XMLRPC. :)
> >
> > i don't know about do.  what i meant is that you map INVITE
> > to XML-RPC function called sip.INVITE, etc.  then SIP Proxy
> > would become an XML-RPC server, which listens at port 80 and/or 443.
> >
> > -- juha
> Why not REST?
> 
> GET http://example.com/INVITE/bob/
> 
> 
> Wolfgang
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From slavitch at objectworld.com  Thu Mar 16 10:18:03 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Thu, 16 Mar 2006 10:18:03 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>

In such a case a nice P2P protocol would set the HTTP request header
user agent designator to be something truly 'P2PSIP', keeping the
firewall guys happy. Make it politically correct, make it nice, and make
it work.

This means that existing firewalls could support P2PSIP monitoring
without any state changes, only a new policy. 
This would make them and their customers very happy and remove a pain
point as discussed yesterday at VON.

It is an interesting way of solving the problem, and making
bootstrap/configuration friendly and polite.

M

Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Aki Niemi
Sent: March 16, 2006 4:16 AM
To: ext Juha Heinanen
Cc: P2P-SIP
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network

Hi Juha,

ext Juha Heinanen wrote:
> Michael Slavitch writes:
> 
>  > The people I need to talk to use Skype.  Nobody that I need to talk

> to  > even knows openser exists, apart from a few alpha geeks, and 
> even  > those alpha geeks use Skype because it's easier, the Skype 
> client is  > better, and Skype is what their girlfriends use.
> 
> i tried to ask, why Skype's NAT traversal is technically better than 
> openser's.  what i have heard about Skype is that it uses very similar

> technology, except that mediaproxies are distributed to customers'
> computers.

I think the main additional thing is that Skype also does a bunch of,
say, "politically incorrect" things, like tunnel via HTTP proxies, to
port 443, etc.

Cheers,
Aki
_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


--


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From RADHIKA.R.ROY at saic.com  Thu Mar 16 12:00:01 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Thu, 16 Mar 2006 12:00:01 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31E9634@0591-ITS-EXMP02.us.saic.com>

G:
 
I think that it is an interesting discussion for working of P2P-SIP with
HTTP proxies.
 
SIP has hired all the things that HTTP does, only in different context: SIP
is almost like P2P keeping primarily human users in mind, while HTTP is a
client-server model.
 
Semantically, many things of SIP will find appropriate place in HTTP
although the syntaxes might be different, and in this way, at least, P2P-SIP
will also be able to work with HTTP proxies, I believe.
 
Best regards,
Radhika

  _____  

From: p2p-sip-bounces at cs.columbia.edu on behalf of Gustavo Garc?a Bernardo
Sent: Thu 3/16/2006 9:35 AM
To: Beck01, Wolfgang
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network





In my opinion, we first should define if P2P SIP should work through HTTP 
proxies, only HTTPS proxies or none of them.   Not only for SIP messages, 
but also for DHT and RTP messages.... 

As someone has said before, these are "politically incorrect" things... 

In any case, I think this feature is a minor detail and doesn't change the 
P2P SIP architecture which should be defined before discussing this kind of 
details. 

G. 



From RADHIKA.R.ROY at saic.com  Thu Mar 16 12:18:49 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Thu, 16 Mar 2006 12:18:49 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31E9637@0591-ITS-EXMP02.us.saic.com>

Juha:
 
Is there any documentation is available on "OPENSER" for FW/NAT crossing as
it is being discussed in this extensively?
 
Best regards,
Radhika

  _____  

From: p2p-sip-bounces at cs.columbia.edu on behalf of Juha Heinanen
Sent: Thu 3/16/2006 6:51 AM
To: Jonathan Rosenberg
Cc: 'P2P-SIP'
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network



jonathan,

i read the outbound draft (not all glory details yet) and i was pleased
with it.  the main reason being that in addition to NAT traversal (which
openser can handle already now satisfactorily without any new
standardization), it also tackles the redundancy problem.

there is one thing though that i can't accept:

   UAs are configured with one or more SIP URIs representing the default
   outbound-proxy-set.  The specification assumes the set is determined
   via configuration but future specifications may define other
   mechanisms such as using DNS to discover this set.

we MUST have a DNS based way for the UAs to figure out the default
outbound proxy set BEFORE this spec is published as an RFC.  if not
done, there is yet another thing to be configured in the UA, which is
exactly what Skype lovers hate in SIP.

-- juha
_______________________________________________



From ag at ag-projects.com  Thu Mar 16 13:43:47 2006
From: ag at ag-projects.com (Adrian Georgescu)
Date: Thu, 16 Mar 2006 10:43:47 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31E9637@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9637@0591-ITS-EXMP02.us.saic.com>
Message-ID: <B010825B-7E5F-40FD-9196-1A11305AE20E@ag-projects.com>

Radhika, see http://www.sipcenter.com/goto/AG-SIP-NAT

Regards
Adrian


On Mar 16, 2006, at 9:18 AM, Roy, Radhika R. wrote:

> Juha:
>
> Is there any documentation is available on "OPENSER" for FW/NAT  
> crossing as
> it is being discussed in this extensively?
>
> Best regards,
> Radhika
>
>   _____
>
> From: p2p-sip-bounces at cs.columbia.edu on behalf of Juha Heinanen
> Sent: Thu 3/16/2006 6:51 AM
> To: Jonathan Rosenberg
> Cc: 'P2P-SIP'
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
>
>
>
> jonathan,
>
> i read the outbound draft (not all glory details yet) and i was  
> pleased
> with it.  the main reason being that in addition to NAT traversal  
> (which
> openser can handle already now satisfactorily without any new
> standardization), it also tackles the redundancy problem.
>
> there is one thing though that i can't accept:
>
>    UAs are configured with one or more SIP URIs representing the  
> default
>    outbound-proxy-set.  The specification assumes the set is  
> determined
>    via configuration but future specifications may define other
>    mechanisms such as using DNS to discover this set.
>
> we MUST have a DNS based way for the UAs to figure out the default
> outbound proxy set BEFORE this spec is published as an RFC.  if not
> done, there is yet another thing to be configured in the UA, which is
> exactly what Skype lovers hate in SIP.
>
> -- juha
> _______________________________________________
>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From slavitch at gmail.com  Thu Mar 16 14:06:16 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Thu, 16 Mar 2006 11:06:16 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31E9634@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9634@0591-ITS-EXMP02.us.saic.com>
Message-ID: <1fc32c8f0603161106qd56c6fcnc5e0c68c34bb562@mail.gmail.com>

Excellent observation.  If we have hired it, hire it well.

Michael

On 3/16/06, Roy, Radhika R. <RADHIKA.R.ROY at saic.com> wrote:
> G:
>
> I think that it is an interesting discussion for working of P2P-SIP with
> HTTP proxies.
>
> SIP has hired all the things that HTTP does, only in different context: SIP
> is almost like P2P keeping primarily human users in mind, while HTTP is a
> client-server model.
>
> Semantically, many things of SIP will find appropriate place in HTTP
> although the syntaxes might be different, and in this way, at least, P2P-SIP
> will also be able to work with HTTP proxies, I believe.
>
> Best regards,
> Radhika
>
>   _____
>
> From: p2p-sip-bounces at cs.columbia.edu on behalf of Gustavo Garc?a Bernardo
> Sent: Thu 3/16/2006 9:35 AM
> To: Beck01, Wolfgang
> Cc: p2p-sip at cs.columbia.edu
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
>
>
>
>
>
> In my opinion, we first should define if P2P SIP should work through HTTP
> proxies, only HTTPS proxies or none of them.   Not only for SIP messages,
> but also for DHT and RTP messages....
>
> As someone has said before, these are "politically incorrect" things...
>
> In any case, I think this feature is a minor detail and doesn't change the
> P2P SIP architecture which should be defined before discussing this kind of
> details.
>
> G.
>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


--
Michael Slavitch
Ottawa, Ontario Canada


From gregd at research.panasonic.com  Thu Mar 16 15:31:02 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Thu, 16 Mar 2006 15:31:02 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>
Message-ID: <4419CB06.5000300@research.panasonic.com>

Hi Michael,

Michael Slavitch wrote:
> In such a case a nice P2P protocol would set the HTTP request header
> user agent designator to be something truly 'P2PSIP', keeping the
> firewall guys happy. Make it politically correct, make it nice, and make
> it work.
> 
> This means that existing firewalls could support P2PSIP monitoring
> without any state changes, only a new policy. 
> This would make them and their customers very happy and remove a pain
> point as discussed yesterday at VON.
> 
> It is an interesting way of solving the problem, and making
> bootstrap/configuration friendly and polite.
> 
> M

We don't want to do this, as it requires modification of proxy
and firewall infrastructure.

There is a reason why we have assigned port numbers.
It's so people can tell what's travelling on a certain port.

If an admin has to install a new firewall or proxy (or upgrade) in
order to stop access of VoIP across my network boundary, they're
either going to be upset at IETF, or not going to let the software
be deployed at all on the network.

So we should have a protocol which is easily identifiable, and
easily blockable (conversely easily openable if need be).  The
security ADs would eat us for dinner if we proposed http traversal
just to get around an organization's legitimately configured
firewall.

Let proprietary vendors do it, and we'll see which protocol is
left on the corporate LAN.

Greg

From slavitch at gmail.com  Thu Mar 16 16:27:33 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Thu, 16 Mar 2006 13:27:33 -0800
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <1fc32c8f0603161325i2788918dmdb83c1810f428ffe@mail.gmail.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>
	<4419CB06.5000300@research.panasonic.com>
	<1fc32c8f0603161325i2788918dmdb83c1810f428ffe@mail.gmail.com>
Message-ID: <1fc32c8f0603161327k96e1b48mca27728ee03bce08@mail.gmail.com>

Forwarded to the list.

---------- Forwarded message ----------
From: Michael Slavitch <slavitch at gmail.com>
Date: Mar 16, 2006 1:25 PM
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
To: Greg Daley <gregd at research.panasonic.com>


Depending on port numbers for security becomes silly when NAT
traversal comes into play.
If your customers depend on static numbering they are truly kidding
themselves. How is it not easily identifiable?  It is a content lookup
on http.  It is polite, W3 stuff. So what is a well-known number?  I
get access to my file system over WEBDAV. WEBDAV runs over http1.1,
giving me access to an entire network structure over http. Outlook and
Exchange use 443 for reading and writing email if you so choose, and
if your settings are correct you have access to a file system through
outlook.  Streaming video, voice, you name it  runs over 80 or 443.
Google mail pop access is over 995 and something else.

I have 2G of disk space on my gmail drive, which I can get access to over 443.

A clever hack to get around a firewall guessing about access is to
have the equivalent of a secure-id card on  a web server,  an http get
request gives you a server port number that only responds to your
request (an interesting abstraction of STUN).

If the firewall cannot read content the firewall is useless.  NAT
traversal by definition breaks the well-known number rule so that must
be off the table.

A few years ago I got tired of a rather nasty firewall and I put an
ssh server on port 443, and with port tunneling I made an instant
homebrew VPN in a few minutes.   The security officer at this firm
insisted that this was impossible,  and didn't understand it even when
I showed him (I was doing a network analysis, it was part of my job).

Depending on 'well-known numbers' when the world can run anything over
80 or 443 means that your customers do not have a firewall, they have
a port restrictor.

Security through stupidity isn't security.

M


On 3/16/06, Greg Daley <gregd at research.panasonic.com> wrote:
> Hi Michael,
>
> Michael Slavitch wrote:
> > In such a case a nice P2P protocol would set the HTTP request header
> > user agent designator to be something truly 'P2PSIP', keeping the
> > firewall guys happy. Make it politically correct, make it nice, and make
> > it work.
> >
> > This means that existing firewalls could support P2PSIP monitoring
> > without any state changes, only a new policy.
> > This would make them and their customers very happy and remove a pain
> > point as discussed yesterday at VON.
> >
> > It is an interesting way of solving the problem, and making
> > bootstrap/configuration friendly and polite.
> >
> > M
>
> We don't want to do this, as it requires modification of proxy
> and firewall infrastructure.
>
> There is a reason why we have assigned port numbers.
> It's so people can tell what's travelling on a certain port.
>
> If an admin has to install a new firewall or proxy (or upgrade) in
> order to stop access of VoIP across my network boundary, they're
> either going to be upset at IETF, or not going to let the software
> be deployed at all on the network.
>
> So we should have a protocol which is easily identifiable, and
> easily blockable (conversely easily openable if need be).  The
> security ADs would eat us for dinner if we proposed http traversal
> just to get around an organization's legitimately configured
> firewall.
>
> Let proprietary vendors do it, and we'll see which protocol is
> left on the corporate LAN.
>
> Greg
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


--
Michael Slavitch
Ottawa, Ontario Canada


--
Michael Slavitch
Ottawa, Ontario Canada


From slavitch at gmail.com  Thu Mar 16 16:33:53 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Thu, 16 Mar 2006 13:33:53 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <1fc32c8f0603161327k96e1b48mca27728ee03bce08@mail.gmail.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>
	<4419CB06.5000300@research.panasonic.com>
	<1fc32c8f0603161325i2788918dmdb83c1810f428ffe@mail.gmail.com>
	<1fc32c8f0603161327k96e1b48mca27728ee03bce08@mail.gmail.com>
Message-ID: <1fc32c8f0603161333p3758c84el5914c141208fcc39@mail.gmail.com>

If WebDav passed security AD's, so must P2PSIP/HTTPS.

E-commerce occurs using WebDav so security is I assume under control
in the application layer.

On 3/16/06, Michael Slavitch <slavitch at gmail.com> wrote:
> Forwarded to the list.
>
> ---------- Forwarded message ----------
> From: Michael Slavitch <slavitch at gmail.com>
> Date: Mar 16, 2006 1:25 PM
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> To: Greg Daley <gregd at research.panasonic.com>
>
>
> Depending on port numbers for security becomes silly when NAT
> traversal comes into play.
> If your customers depend on static numbering they are truly kidding
> themselves. How is it not easily identifiable?  It is a content lookup
> on http.  It is polite, W3 stuff. So what is a well-known number?  I
> get access to my file system over WEBDAV. WEBDAV runs over http1.1,
> giving me access to an entire network structure over http. Outlook and
> Exchange use 443 for reading and writing email if you so choose, and
> if your settings are correct you have access to a file system through
> outlook.  Streaming video, voice, you name it  runs over 80 or 443.
> Google mail pop access is over 995 and something else.
>
> I have 2G of disk space on my gmail drive, which I can get access to over 443.
>
> A clever hack to get around a firewall guessing about access is to
> have the equivalent of a secure-id card on  a web server,  an http get
> request gives you a server port number that only responds to your
> request (an interesting abstraction of STUN).
>
> If the firewall cannot read content the firewall is useless.  NAT
> traversal by definition breaks the well-known number rule so that must
> be off the table.
>
> A few years ago I got tired of a rather nasty firewall and I put an
> ssh server on port 443, and with port tunneling I made an instant
> homebrew VPN in a few minutes.   The security officer at this firm
> insisted that this was impossible,  and didn't understand it even when
> I showed him (I was doing a network analysis, it was part of my job).
>
> Depending on 'well-known numbers' when the world can run anything over
> 80 or 443 means that your customers do not have a firewall, they have
> a port restrictor.
>
> Security through stupidity isn't security.
>
> M
>
>
> On 3/16/06, Greg Daley <gregd at research.panasonic.com> wrote:
> > Hi Michael,
> >
> > Michael Slavitch wrote:
> > > In such a case a nice P2P protocol would set the HTTP request header
> > > user agent designator to be something truly 'P2PSIP', keeping the
> > > firewall guys happy. Make it politically correct, make it nice, and make
> > > it work.
> > >
> > > This means that existing firewalls could support P2PSIP monitoring
> > > without any state changes, only a new policy.
> > > This would make them and their customers very happy and remove a pain
> > > point as discussed yesterday at VON.
> > >
> > > It is an interesting way of solving the problem, and making
> > > bootstrap/configuration friendly and polite.
> > >
> > > M
> >
> > We don't want to do this, as it requires modification of proxy
> > and firewall infrastructure.
> >
> > There is a reason why we have assigned port numbers.
> > It's so people can tell what's travelling on a certain port.
> >
> > If an admin has to install a new firewall or proxy (or upgrade) in
> > order to stop access of VoIP across my network boundary, they're
> > either going to be upset at IETF, or not going to let the software
> > be deployed at all on the network.
> >
> > So we should have a protocol which is easily identifiable, and
> > easily blockable (conversely easily openable if need be).  The
> > security ADs would eat us for dinner if we proposed http traversal
> > just to get around an organization's legitimately configured
> > firewall.
> >
> > Let proprietary vendors do it, and we'll see which protocol is
> > left on the corporate LAN.
> >
> > Greg
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
>
>
> --
> Michael Slavitch
> Ottawa, Ontario Canada
>
>
> --
> Michael Slavitch
> Ottawa, Ontario Canada
>


--
Michael Slavitch
Ottawa, Ontario Canada


From jh at tutpro.com  Thu Mar 16 17:05:23 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Fri, 17 Mar 2006 00:05:23 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31E9637@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31E9637@0591-ITS-EXMP02.us.saic.com>
Message-ID: <17433.57635.788864.877714@rautu.tutpro.com>

Roy, Radhika R. writes:

 > Is there any documentation is available on "OPENSER" for FW/NAT crossing as
 > it is being discussed in this extensively?

all functions that you need to implement NAT traversal in the proxy are
documentation at www.openser.org (see nathelper, mediaproxy, registrar,
and usrloc modules).  in addition you need to install an RTP proxy for
which there are two to choose from.

-- juha

From gregd at research.panasonic.com  Thu Mar 16 17:06:37 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Thu, 16 Mar 2006 17:06:37 -0500
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <1fc32c8f0603161327k96e1b48mca27728ee03bce08@mail.gmail.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>	<4419CB06.5000300@research.panasonic.com>	<1fc32c8f0603161325i2788918dmdb83c1810f428ffe@mail.gmail.com>
	<1fc32c8f0603161327k96e1b48mca27728ee03bce08@mail.gmail.com>
Message-ID: <4419E16D.8070503@research.panasonic.com>

Hi Michael,

You are missing the point.
I'm not for relying on port numbers for security.

I'm for not overloading the mechanisms _yet again_
and allowing systems administrators to:

1) Identify the traffic on the network

2) Provide basic policy control.

People will certainly be able to tunnel stuff out
of the network using customized tools.  I'm just
saying we shouldn't build our protocol on top of the
idea.  Put P2P over SIP over HTTP over SSH over DNS if
you like, just don't propose it when other people are
having a serious attempt at chartering a working group.

People will think the group hasn't made a serious
consideration of operational security requirements.

It's also worth considering that while _outbound_ HTTP
HTTPS and DNS are common on the Internet.  How many
companies allow them inbound?

We're going to have to allow inbound traffic.  Why
not have a well known port?

Greg


Michael Slavitch wrote:
> Forwarded to the list.
> 
> ---------- Forwarded message ----------
> From: Michael Slavitch <slavitch at gmail.com>
> Date: Mar 16, 2006 1:25 PM
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> To: Greg Daley <gregd at research.panasonic.com>
> 
> 
> Depending on port numbers for security becomes silly when NAT
> traversal comes into play.
> If your customers depend on static numbering they are truly kidding
> themselves. How is it not easily identifiable?  It is a content lookup
> on http.  It is polite, W3 stuff. So what is a well-known number?  I
> get access to my file system over WEBDAV. WEBDAV runs over http1.1,
> giving me access to an entire network structure over http. Outlook and
> Exchange use 443 for reading and writing email if you so choose, and
> if your settings are correct you have access to a file system through
> outlook.  Streaming video, voice, you name it  runs over 80 or 443.
> Google mail pop access is over 995 and something else.
> 
> I have 2G of disk space on my gmail drive, which I can get access to over 443.
> 
> A clever hack to get around a firewall guessing about access is to
> have the equivalent of a secure-id card on  a web server,  an http get
> request gives you a server port number that only responds to your
> request (an interesting abstraction of STUN).
> 
> If the firewall cannot read content the firewall is useless.  NAT
> traversal by definition breaks the well-known number rule so that must
> be off the table.
> 
> A few years ago I got tired of a rather nasty firewall and I put an
> ssh server on port 443, and with port tunneling I made an instant
> homebrew VPN in a few minutes.   The security officer at this firm
> insisted that this was impossible,  and didn't understand it even when
> I showed him (I was doing a network analysis, it was part of my job).
> 
> Depending on 'well-known numbers' when the world can run anything over
> 80 or 443 means that your customers do not have a firewall, they have
> a port restrictor.
> 
> Security through stupidity isn't security.
> 
> M
> 
> 
> On 3/16/06, Greg Daley <gregd at research.panasonic.com> wrote:
>> Hi Michael,
>>
>> Michael Slavitch wrote:
>>> In such a case a nice P2P protocol would set the HTTP request header
>>> user agent designator to be something truly 'P2PSIP', keeping the
>>> firewall guys happy. Make it politically correct, make it nice, and make
>>> it work.
>>>
>>> This means that existing firewalls could support P2PSIP monitoring
>>> without any state changes, only a new policy.
>>> This would make them and their customers very happy and remove a pain
>>> point as discussed yesterday at VON.
>>>
>>> It is an interesting way of solving the problem, and making
>>> bootstrap/configuration friendly and polite.
>>>
>>> M
>> We don't want to do this, as it requires modification of proxy
>> and firewall infrastructure.
>>
>> There is a reason why we have assigned port numbers.
>> It's so people can tell what's travelling on a certain port.
>>
>> If an admin has to install a new firewall or proxy (or upgrade) in
>> order to stop access of VoIP across my network boundary, they're
>> either going to be upset at IETF, or not going to let the software
>> be deployed at all on the network.
>>
>> So we should have a protocol which is easily identifiable, and
>> easily blockable (conversely easily openable if need be).  The
>> security ADs would eat us for dinner if we proposed http traversal
>> just to get around an organization's legitimately configured
>> firewall.
>>
>> Let proprietary vendors do it, and we'll see which protocol is
>> left on the corporate LAN.
>>
>> Greg
>> _______________________________________________
>> p2p-sip mailing list
>> p2p-sip at cs.columbia.edu
>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
> 
> 
> --
> Michael Slavitch
> Ottawa, Ontario Canada
> 
> 
> --
> Michael Slavitch
> Ottawa, Ontario Canada
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From gregd at research.panasonic.com  Thu Mar 16 17:27:24 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Thu, 16 Mar 2006 17:27:24 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <1fc32c8f0603161333p3758c84el5914c141208fcc39@mail.gmail.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>	<4419CB06.5000300@research.panasonic.com>	<1fc32c8f0603161325i2788918dmdb83c1810f428ffe@mail.gmail.com>	<1fc32c8f0603161327k96e1b48mca27728ee03bce08@mail.gmail.com>
	<1fc32c8f0603161333p3758c84el5914c141208fcc39@mail.gmail.com>
Message-ID: <4419E64C.8040507@research.panasonic.com>

Hi Michael,

Michael Slavitch wrote:
> If WebDav passed security AD's, so must P2PSIP/HTTPS.
> 
> E-commerce occurs using WebDav so security is I assume under control
> in the application layer.

You've essentially picked a limited, assymmetric application
which requires a server to illustrate how security will work
in a P2P network.  Sure it's fine for file storage.  There are
only a finite number of operations you can do.

Saying something is possible doesn't mean it's a good
match, though.

Providing explicit protocol/port identification allows
straightforward troubleshooting and control.  Sure it doesn't
always work, but it's the principle that most firewalls
are based upon today, and matches the paradigm.

I don't know why you're so against it. Perhaps you have an
esoteric reason you could reveal to the group?

Greg


From dbarrett at quinthar.com  Thu Mar 16 17:28:46 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Thu, 16 Mar 2006 14:28:46 -0800
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <4419E16D.8070503@research.panasonic.com>
Message-ID: <200603162228.k2GMSpOZ028582@cs.columbia.edu>

It seems to me the key sticking point is deciding which is more important:

1) Adhering to the spirit of the IANA
2) Creating a system that "just works" as well as Skype

Clearly, if we can do both, then there's no conflict.  But in the event that
we can't do both, which wins?

I think Michael is arguing that the user trumps all (even if it means
ignoring the IANA).  And I think Greg is arguing that the IANA numbering
can't be violated (even if it means limiting functionality for users).

Personally, I'm in Michael's camp -- if what we make isn't at least as good
for users as Skype is today (and remember, it'll only get better), then
there will be little gain from our effort.

Thus while I like the idea of using well-known IANA port numbers, and I'm
eager to find a way to stick within its guidelines, I'm more concerned with
making the best experience for users.

-david

> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-
> bounces at cs.columbia.edu] On Behalf Of Greg Daley
> Sent: Thursday, March 16, 2006 2:07 PM
> To: Michael Slavitch
> Cc: P2P-SIP
> Subject: Re: [p2p-sip] Fwd: SIP, ICE, and an Overlay Network
> 
> Hi Michael,
> 
> You are missing the point.
> I'm not for relying on port numbers for security.
> 
> I'm for not overloading the mechanisms _yet again_
> and allowing systems administrators to:
> 
> 1) Identify the traffic on the network
> 
> 2) Provide basic policy control.
> 
> People will certainly be able to tunnel stuff out
> of the network using customized tools.  I'm just
> saying we shouldn't build our protocol on top of the
> idea.  Put P2P over SIP over HTTP over SSH over DNS if
> you like, just don't propose it when other people are
> having a serious attempt at chartering a working group.
> 
> People will think the group hasn't made a serious
> consideration of operational security requirements.
> 
> It's also worth considering that while _outbound_ HTTP
> HTTPS and DNS are common on the Internet.  How many
> companies allow them inbound?
> 
> We're going to have to allow inbound traffic.  Why
> not have a well known port?
> 
> Greg
> 
> 
> Michael Slavitch wrote:
> > Forwarded to the list.
> >
> > ---------- Forwarded message ----------
> > From: Michael Slavitch <slavitch at gmail.com>
> > Date: Mar 16, 2006 1:25 PM
> > Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> > To: Greg Daley <gregd at research.panasonic.com>
> >
> >
> > Depending on port numbers for security becomes silly when NAT
> > traversal comes into play.
> > If your customers depend on static numbering they are truly kidding
> > themselves. How is it not easily identifiable?  It is a content lookup
> > on http.  It is polite, W3 stuff. So what is a well-known number?  I
> > get access to my file system over WEBDAV. WEBDAV runs over http1.1,
> > giving me access to an entire network structure over http. Outlook and
> > Exchange use 443 for reading and writing email if you so choose, and
> > if your settings are correct you have access to a file system through
> > outlook.  Streaming video, voice, you name it  runs over 80 or 443.
> > Google mail pop access is over 995 and something else.
> >
> > I have 2G of disk space on my gmail drive, which I can get access to
> over 443.
> >
> > A clever hack to get around a firewall guessing about access is to
> > have the equivalent of a secure-id card on  a web server,  an http get
> > request gives you a server port number that only responds to your
> > request (an interesting abstraction of STUN).
> >
> > If the firewall cannot read content the firewall is useless.  NAT
> > traversal by definition breaks the well-known number rule so that must
> > be off the table.
> >
> > A few years ago I got tired of a rather nasty firewall and I put an
> > ssh server on port 443, and with port tunneling I made an instant
> > homebrew VPN in a few minutes.   The security officer at this firm
> > insisted that this was impossible,  and didn't understand it even when
> > I showed him (I was doing a network analysis, it was part of my job).
> >
> > Depending on 'well-known numbers' when the world can run anything over
> > 80 or 443 means that your customers do not have a firewall, they have
> > a port restrictor.
> >
> > Security through stupidity isn't security.
> >
> > M
> >
> >
> > On 3/16/06, Greg Daley <gregd at research.panasonic.com> wrote:
> >> Hi Michael,
> >>
> >> Michael Slavitch wrote:
> >>> In such a case a nice P2P protocol would set the HTTP request header
> >>> user agent designator to be something truly 'P2PSIP', keeping the
> >>> firewall guys happy. Make it politically correct, make it nice, and
> make
> >>> it work.
> >>>
> >>> This means that existing firewalls could support P2PSIP monitoring
> >>> without any state changes, only a new policy.
> >>> This would make them and their customers very happy and remove a pain
> >>> point as discussed yesterday at VON.
> >>>
> >>> It is an interesting way of solving the problem, and making
> >>> bootstrap/configuration friendly and polite.
> >>>
> >>> M
> >> We don't want to do this, as it requires modification of proxy
> >> and firewall infrastructure.
> >>
> >> There is a reason why we have assigned port numbers.
> >> It's so people can tell what's travelling on a certain port.
> >>
> >> If an admin has to install a new firewall or proxy (or upgrade) in
> >> order to stop access of VoIP across my network boundary, they're
> >> either going to be upset at IETF, or not going to let the software
> >> be deployed at all on the network.
> >>
> >> So we should have a protocol which is easily identifiable, and
> >> easily blockable (conversely easily openable if need be).  The
> >> security ADs would eat us for dinner if we proposed http traversal
> >> just to get around an organization's legitimately configured
> >> firewall.
> >>
> >> Let proprietary vendors do it, and we'll see which protocol is
> >> left on the corporate LAN.
> >>
> >> Greg
> >> _______________________________________________
> >> p2p-sip mailing list
> >> p2p-sip at cs.columbia.edu
> >> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >>
> >
> >
> > --
> > Michael Slavitch
> > Ottawa, Ontario Canada
> >
> >
> > --
> > Michael Slavitch
> > Ottawa, Ontario Canada
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From aki.niemi at nokia.com  Thu Mar 16 17:30:23 2006
From: aki.niemi at nokia.com (Aki Niemi)
Date: Fri, 17 Mar 2006 00:30:23 +0200
Subject: [p2p-sip] Tunneling SIP via HTTP proxies (was: Re:  Fwd:  SIP, ICE,
 and an Overlay Network)
In-Reply-To: <4419E16D.8070503@research.panasonic.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C7C7@photon.objectworld.com>	<4419CB06.5000300@research.panasonic.com>	<1fc32c8f0603161325i2788918dmdb83c1810f428ffe@mail.gmail.com>	<1fc32c8f0603161327k96e1b48mca27728ee03bce08@mail.gmail.com>
	<4419E16D.8070503@research.panasonic.com>
Message-ID: <4419E6FF.60007@nokia.com>

Hi,

Many IM clients make use of the HTTP connect (or alternatively, use
socks) method to get through firewalls. Of course, if you for example
run Oscar, then you'd make HTTP connect to port 5190. Your firewall
admin may or may not allow connects to that port.

I think writing up how to use CONNECT (RFC 2817) to connect to port 5060
is trivial. It also has a better chance of getting through IESG review
than explaining how to tunnel P2PSIP over port 443. Probably some ADs
could chime in here with their opinions on this matter.

Also...

ext Greg Daley wrote:
> We're going to have to allow inbound traffic.  Why
> not have a well known port?

Naturally the TCP connection you open using the HTTP proxy works in both
directions. It does require the supernode you connected to to relay
traffic, instead of simply redirecting, though.

Cheers,
Aki

From jh at tutpro.com  Thu Mar 16 17:48:55 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Fri, 17 Mar 2006 00:48:55 +0200
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <200603162228.k2GMSpOZ028582@cs.columbia.edu>
References: <4419E16D.8070503@research.panasonic.com>
	<200603162228.k2GMSpOZ028582@cs.columbia.edu>
Message-ID: <17433.60247.397968.140795@rautu.tutpro.com>

David Barrett writes:

 > Thus while I like the idea of using well-known IANA port numbers, and I'm
 > eager to find a way to stick within its guidelines, I'm more concerned with
 > making the best experience for users.

since we have DNS SRV records, there are no use for well known port
numbers for any application than perhaps DNS.  a firewall that believes
that port 80 or 443 traffic is web traffic is really stupid.

-- juha

From dbarrett at quinthar.com  Thu Mar 16 17:53:13 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Thu, 16 Mar 2006 14:53:13 -0800
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <17433.60247.397968.140795@rautu.tutpro.com>
Message-ID: <200603162253.k2GMrMOZ014012@cs.columbia.edu>

> -----Original Message-----
> From: Juha Heinanen [mailto:jh at tutpro.com]
> Subject: Re: [p2p-sip] Fwd: SIP, ICE, and an Overlay Network
> 
> David Barrett writes:
> 
>  > Thus while I like the idea of using well-known IANA port numbers, and
> I'm
>  > eager to find a way to stick within its guidelines, I'm more concerned
> with
>  > making the best experience for users.
> 
> since we have DNS SRV records, there are no use for well known port
> numbers for any application than perhaps DNS.  a firewall that believes
> that port 80 or 443 traffic is web traffic is really stupid.

Well, stupid or not, I think the argument is that port 80 is "supposed" to
only be HTTP according to the IANA.  The question is to what degree we are
bound by the IANA's guidelines.

-david


From gregd at research.panasonic.com  Thu Mar 16 17:54:39 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Thu, 16 Mar 2006 17:54:39 -0500
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <200603162227.AEF17639@redwood.research.panasonic.com>
References: <200603162227.AEF17639@redwood.research.panasonic.com>
Message-ID: <4419ECAF.50004@research.panasonic.com>

Dear Dave,

Please let me be clear.

David Barrett wrote:
> It seems to me the key sticking point is deciding which is more important:
> 
> 1) Adhering to the spirit of the IANA
> 2) Creating a system that "just works" as well as Skype
> 
> Clearly, if we can do both, then there's no conflict.  But in the event that
> we can't do both, which wins?
> 
> I think Michael is arguing that the user trumps all (even if it means
> ignoring the IANA).  And I think Greg is arguing that the IANA numbering
> can't be violated (even if it means limiting functionality for users).


I'm not saying it can't be violated, it's just that there
are costs if you do.  Following the leader an putting everything
through the smallest open hole in the firewall isn't a magic bullet.


> Personally, I'm in Michael's camp -- if what we make isn't at least as good
> for users as Skype is today (and remember, it'll only get better), then
> there will be little gain from our effort.
> 
> Thus while I like the idea of using well-known IANA port numbers, and I'm
> eager to find a way to stick within its guidelines, I'm more concerned with
> making the best experience for users.

I personally am not interested in making skype.
Perhaps I am one of the few.

I do not want to do it because it has already been done,
and I think that if we want to make skype we could
save a lot of time and useless bickering by asking
them to come and present the protocol at the BoF.

I want to make something MUCH better than skype, which
sometimes means making different architectural decisions
to their system.  (really!)


So if administrators can switch off P2P SIP at the firewall,
and direct their people to use a CS/SIP proxy as a gateway,
while having completely unrestricted (and otherwise
serverless and unmanaged: actually the proxy could be
configured off the overlay) access within the LAN, I think
this interoperability is a selling point.

I also think that because the administrators know they can
turn it off when they like, they will let it on the network,
whereas corporate IT teams will ban use of applications which
force their way through firewalls.

"users" in this case doesn't mean "individuals".  It really means
"enterprises". Individuals at home don't block outbound ports.

Greg


From hgs at cs.columbia.edu  Thu Mar 16 18:09:34 2006
From: hgs at cs.columbia.edu (Henning Schulzrinne)
Date: Thu, 16 Mar 2006 18:09:34 -0500
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <4419ECAF.50004@research.panasonic.com>
References: <200603162227.AEF17639@redwood.research.panasonic.com>
	<4419ECAF.50004@research.panasonic.com>
Message-ID: <4419F02E.6060002@cs.columbia.edu>

Maybe we need to step back and see what kind of systems make longer-term 
technical sense. I think we need to distinguish, for example, between 
inbound and outbound firewalls. We seem to have:

- Connection-based: Any connection is ok, as long as it is initiated by 
the inside. Some people erroneously believe that NATs provide this 
functionality.

- Port-based: Probably just encourages cloaking if used for outbound 
filtering, but works reasonably well for limiting inbound connections to 
servers that are authorized and designed to provide that service (as 
opposed to some rogue non-patched IIS server running on a personal laptop).

- Content-based: Deep packet inspection seems to be all the vogue, but I 
think in the end, only true protocol participation is going to be 
useful. In other words, the firewall has to be an outbound proxy (or 
work with a trusted outbound proxy) to make sure that only real protocol 
messages make it in or out. I don't see much hope in preventing 
information leakage - it's just too easy to hide information or tunnel 
protocols in what looks like valid HTML or SIP (steganography). Just 
send your protocol requests in base 64 as Via tokens or in Subject 
headers or in a hundred other ways.

I think it also helps to understand the often unstated motivations for 
these filters, which are somewhat different:

- keep the bad guys from connecting to machines that are not meant to 
provide public services;

- keep bad protocol elements out before the IT guys have had a chance to 
fix the buffer overflow;

- keep stupid, untrustworthy and gullible users (as viewed by the IT 
guys) from connecting to bad guys.

Thus, just saying "firewall" tells us very little without being more 
precise what kind of firewall it is and why it is there.

We are now, however, dangerously close to getting into a discussion on 
session border controllers.


FWIW, 
http://www.blackhat.com/presentations/bh-europe-06/bh-eu-06-biondi/bh-eu-06-biondi-up.pdf 
illustrates that these problems are not merely theoretical.


Henning

From slavitch at objectworld.com  Thu Mar 16 20:20:20 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Thu, 16 Mar 2006 20:20:20 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C7E4@photon.objectworld.com>

I'm not sure how limited WebDAV is.  An entire industry is built around
it.

It gives me encrypted r/w access to a whole file system with local
security credentials, replacing the need for a VPN or configuring NAT
firewalls as long as I have accessed to a common server. No more worry
about NFS, SMB, whatever. All I need is WEBDAV and credentials. This is
a good thing.  A Mac that doesn't run SMB can get data from an XP box if
it can log in. No more futzing or fussing or configuring. It is
automatic. 

It allows a company in Cupertino to actually make money selling music by
making it easier to buy than steal.

My local university uses Webdav to offer video archives to its students
so they can catch up on lectures.

Whiteboard collaboration, remote terminals, database read/write, all can
and is being done with WEBDAV, which is RFC status.  WEBDAV is hardly
trivial because it allows you to do pretty well anything over port 443,
except for real-time streaming media.  

That it cannot do, for real time media doesn't work with get/put except
at teletype rates. 

P2PSIP adds media, once done an entire Internet experience can be
encapsulated using two IETF standard protocols.  

All over a single port that is secured and well understood with a
well-described and well-respected protocol, something so well respected
it is used for financial transactions by big conservative banks.   

"I don't know why you're so against it."

Because in practise it no longer exists. It existed in an ancient static
internet that didn't have session establishment protocols and protocol
overloading. Fixed ports were needed pre-SIP and pre-HTTP because there
was no better alternative.  Applications decide on content, not the
protocol.


If the argument is that it cannot be done then the WEBDAV RFC must be
deprecated. There is no alternative.

If WebDav is allowed to remain then P2PSIP should be allowed to follow
its lead. You just can't say yes to one and no to another, especially
when the truth in the field is that well-known numbering no longer
exists in practise except for DNS, SMTP, and HTTP/HTTPS.  

M



Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Greg Daley
Sent: March 16, 2006 2:27 PM
To: Michael Slavitch
Cc: P2P-SIP
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network

Hi Michael,

Michael Slavitch wrote:
> If WebDav passed security AD's, so must P2PSIP/HTTPS.
> 
> E-commerce occurs using WebDav so security is I assume under control 
> in the application layer.

You've essentially picked a limited, assymmetric application which
requires a server to illustrate how security will work in a P2P network.
Sure it's fine for file storage.  There are only a finite number of
operations you can do.

Saying something is possible doesn't mean it's a good match, though.

Providing explicit protocol/port identification allows straightforward
troubleshooting and control.  Sure it doesn't always work, but it's the
principle that most firewalls are based upon today, and matches the
paradigm.

I don't know why you're so against it. Perhaps you have an esoteric
reason you could reveal to the group?

Greg

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


--


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From slavitch at objectworld.com  Thu Mar 16 20:24:17 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Thu, 16 Mar 2006 20:24:17 -0500
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C7E5@photon.objectworld.com>

"We're going to have to allow inbound traffic.  Why not have a well
known port?" 

Because we have to also deal with RTP. We already have to do NAT
traversal. 
By definition if there is NAT traversal well-known ports go out the
window.
Symmetric UDP is about a big a break of firewall security as there can
be.
Those ports are not only opened, they are opened dynamically both
outbound and in.

M

Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Greg Daley
Sent: March 16, 2006 2:07 PM
To: Michael Slavitch
Cc: P2P-SIP
Subject: Re: [p2p-sip] Fwd: SIP, ICE, and an Overlay Network

Hi Michael,

You are missing the point.
I'm not for relying on port numbers for security.

I'm for not overloading the mechanisms _yet again_ and allowing systems
administrators to:

1) Identify the traffic on the network

2) Provide basic policy control.

People will certainly be able to tunnel stuff out of the network using
customized tools.  I'm just saying we shouldn't build our protocol on
top of the idea.  Put P2P over SIP over HTTP over SSH over DNS if you
like, just don't propose it when other people are having a serious
attempt at chartering a working group.

People will think the group hasn't made a serious consideration of
operational security requirements.

It's also worth considering that while _outbound_ HTTP HTTPS and DNS are
common on the Internet.  How many companies allow them inbound?

We're going to have to allow inbound traffic.  Why not have a well known
port?

Greg


Michael Slavitch wrote:
> Forwarded to the list.
> 
> ---------- Forwarded message ----------
> From: Michael Slavitch <slavitch at gmail.com>
> Date: Mar 16, 2006 1:25 PM
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> To: Greg Daley <gregd at research.panasonic.com>
> 
> 
> Depending on port numbers for security becomes silly when NAT 
> traversal comes into play.
> If your customers depend on static numbering they are truly kidding 
> themselves. How is it not easily identifiable?  It is a content lookup

> on http.  It is polite, W3 stuff. So what is a well-known number?  I 
> get access to my file system over WEBDAV. WEBDAV runs over http1.1, 
> giving me access to an entire network structure over http. Outlook and

> Exchange use 443 for reading and writing email if you so choose, and 
> if your settings are correct you have access to a file system through 
> outlook.  Streaming video, voice, you name it  runs over 80 or 443.
> Google mail pop access is over 995 and something else.
> 
> I have 2G of disk space on my gmail drive, which I can get access to
over 443.
> 
> A clever hack to get around a firewall guessing about access is to 
> have the equivalent of a secure-id card on  a web server,  an http get

> request gives you a server port number that only responds to your 
> request (an interesting abstraction of STUN).
> 
> If the firewall cannot read content the firewall is useless.  NAT 
> traversal by definition breaks the well-known number rule so that must

> be off the table.
> 
> A few years ago I got tired of a rather nasty firewall and I put an 
> ssh server on port 443, and with port tunneling I made an instant
> homebrew VPN in a few minutes.   The security officer at this firm
> insisted that this was impossible,  and didn't understand it even when

> I showed him (I was doing a network analysis, it was part of my job).
> 
> Depending on 'well-known numbers' when the world can run anything over

> 80 or 443 means that your customers do not have a firewall, they have 
> a port restrictor.
> 
> Security through stupidity isn't security.
> 
> M
> 
> 
> On 3/16/06, Greg Daley <gregd at research.panasonic.com> wrote:
>> Hi Michael,
>>
>> Michael Slavitch wrote:
>>> In such a case a nice P2P protocol would set the HTTP request header

>>> user agent designator to be something truly 'P2PSIP', keeping the 
>>> firewall guys happy. Make it politically correct, make it nice, and 
>>> make it work.
>>>
>>> This means that existing firewalls could support P2PSIP monitoring 
>>> without any state changes, only a new policy.
>>> This would make them and their customers very happy and remove a 
>>> pain point as discussed yesterday at VON.
>>>
>>> It is an interesting way of solving the problem, and making 
>>> bootstrap/configuration friendly and polite.
>>>
>>> M
>> We don't want to do this, as it requires modification of proxy and 
>> firewall infrastructure.
>>
>> There is a reason why we have assigned port numbers.
>> It's so people can tell what's travelling on a certain port.
>>
>> If an admin has to install a new firewall or proxy (or upgrade) in 
>> order to stop access of VoIP across my network boundary, they're 
>> either going to be upset at IETF, or not going to let the software be

>> deployed at all on the network.
>>
>> So we should have a protocol which is easily identifiable, and easily

>> blockable (conversely easily openable if need be).  The security ADs 
>> would eat us for dinner if we proposed http traversal just to get 
>> around an organization's legitimately configured firewall.
>>
>> Let proprietary vendors do it, and we'll see which protocol is left 
>> on the corporate LAN.
>>
>> Greg
>> _______________________________________________
>> p2p-sip mailing list
>> p2p-sip at cs.columbia.edu
>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
> 
> 
> --
> Michael Slavitch
> Ottawa, Ontario Canada
> 
> 
> --
> Michael Slavitch
> Ottawa, Ontario Canada
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


--


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From dbarrett at quinthar.com  Thu Mar 16 20:28:28 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Thu, 16 Mar 2006 17:28:28 -0800
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
In-Reply-To: <4419ECAF.50004@research.panasonic.com>
Message-ID: <200603170128.k2H1SdOZ009429@cs.columbia.edu>

> -----Original Message-----
> From: Greg Daley [mailto:gregd at research.panasonic.com]
> Subject: Re: [p2p-sip] Fwd: SIP, ICE, and an Overlay Network
> 
> So if administrators can switch off P2P SIP at the firewall,
> and direct their people to use a CS/SIP proxy as a gateway,
> while having completely unrestricted (and otherwise
> serverless and unmanaged: actually the proxy could be
> configured off the overlay) access within the LAN, I think
> this interoperability is a selling point.

Ah, awesome, now I understand what you're saying.  Yes, this sounds like a
good vision.  The trick is distinguishing between environments that
explicitly block SIP due to some active policy, versus environments that
accidentally block SIP due to some combination of misconfigurations.

Skype takes an 'opt out' approach where, unless you explicitly block access
to the central authentication servers, Skype will do everything it can to
figure out a way to work (including tunneling over other ports).

SIP currently takes an 'opt in' approach where, unless everything is tuned
in just right, it assumes it's not supposed to work (and doesn't try to
tunnel). 

I certainly support the notion of keeping admins in control of their
networks.  But I lean towards making PSP-SIP more of an 'opt out' service
like Skype, than maintaining the current 'opt in' bias.  Effective NAT
penetration is one big step in this direction; I think opt-out firewall
tunneling is another good move.

-david


From slavitch at objectworld.com  Thu Mar 16 20:46:06 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Thu, 16 Mar 2006 20:46:06 -0500
Subject: [p2p-sip] Fwd:  SIP, ICE, and an Overlay Network
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C7E6@photon.objectworld.com>

"I certainly support the notion of keeping admins in control of their
networks."

What gives more clear control than an HTTP request header user agent
designator? - "No X allowed here sorry. Go use something else, like
this."
In the extreme case the designator can be overloaded to contain a cert.
In that case the admin can permit/exclude traffic on a strong policy
basis guaranteeing security and identity, and the failure mode can
include an informative redirect to a third party so that it is well
understood to not be a man-in-the-middle attack.  In HTTP-land these are
well-established and well-done things, allowing as much administrative
control as one can muster. 

In SIP this is an open research item. Why?

A well-known port is a port is a port. That is VAX-era stuff.

M



Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David Barrett
Sent: March 16, 2006 5:28 PM
To: 'Greg Daley'
Cc: 'P2P-SIP'
Subject: Re: [p2p-sip] Fwd: SIP, ICE, and an Overlay Network

> -----Original Message-----
> From: Greg Daley [mailto:gregd at research.panasonic.com]
> Subject: Re: [p2p-sip] Fwd: SIP, ICE, and an Overlay Network
> 
> So if administrators can switch off P2P SIP at the firewall, and 
> direct their people to use a CS/SIP proxy as a gateway, while having 
> completely unrestricted (and otherwise serverless and unmanaged: 
> actually the proxy could be configured off the overlay) access within 
> the LAN, I think this interoperability is a selling point.

Ah, awesome, now I understand what you're saying.  Yes, this sounds like
a good vision.  The trick is distinguishing between environments that
explicitly block SIP due to some active policy, versus environments that
accidentally block SIP due to some combination of misconfigurations.

Skype takes an 'opt out' approach where, unless you explicitly block
access to the central authentication servers, Skype will do everything
it can to figure out a way to work (including tunneling over other
ports).

SIP currently takes an 'opt in' approach where, unless everything is
tuned in just right, it assumes it's not supposed to work (and doesn't
try to tunnel). 

I certainly support the notion of keeping admins in control of their
networks.  But I lean towards making PSP-SIP more of an 'opt out'
service like Skype, than maintaining the current 'opt in' bias.
Effective NAT penetration is one big step in this direction; I think
opt-out firewall tunneling is another good move.

-david

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


--


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From RADHIKA.R.ROY at saic.com  Thu Mar 16 21:11:09 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Thu, 16 Mar 2006 21:11:09 -0500
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31E963D@0591-ITS-EXMP02.us.saic.com>

All,
 
The main assumption of HTTP, WevDev, and Web application standardirization
have been to cross the firewall/nat. So, they have done this very easily
without worrying about the fw/nat crossing problems. One advantage, I think,
they have is this: Both signaling and media are in the same plane.
 
In SIP, we could do the same if we would make our number one requirement
that both signaling and media in SIP need to cross the fw/nat, I think.
Since we forgot this most essential requirement during the protocol
development of SIP, we are paying the heavy price now.
 
So, let me turn around the discussion a little differently as follows:
 
What would we do in developing the SIP protocol in the beginning if the
number one requirement would have the crossing of the fw/nat by both
signaling and media?
 
Best regards,
Radhika
 
  _____  

From: p2p-sip-bounces at cs.columbia.edu on behalf of Michael Slavitch
Sent: Thu 3/16/2006 8:20 PM
To: Greg Daley; Michael Slavitch
Cc: P2P-SIP
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network



I'm not sure how limited WebDAV is.  An entire industry is built around 
it. 

It gives me encrypted r/w access to a whole file system with local 
security credentials, replacing the need for a VPN or configuring NAT 
firewalls as long as I have accessed to a common server. No more worry 
about NFS, SMB, whatever. All I need is WEBDAV and credentials. This is 
a good thing.  A Mac that doesn't run SMB can get data from an XP box if 
it can log in. No more futzing or fussing or configuring. It is 
automatic. 

It allows a company in Cupertino to actually make money selling music by 
making it easier to buy than steal. 

My local university uses Webdav to offer video archives to its students 
so they can catch up on lectures. 

Whiteboard collaboration, remote terminals, database read/write, all can 
and is being done with WEBDAV, which is RFC status.  WEBDAV is hardly 
trivial because it allows you to do pretty well anything over port 443, 
except for real-time streaming media.  

That it cannot do, for real time media doesn't work with get/put except 
at teletype rates. 

P2PSIP adds media, once done an entire Internet experience can be 
encapsulated using two IETF standard protocols.  

All over a single port that is secured and well understood with a 
well-described and well-respected protocol, something so well respected 
it is used for financial transactions by big conservative banks.   

"I don't know why you're so against it." 

Because in practise it no longer exists. It existed in an ancient static 
internet that didn't have session establishment protocols and protocol 
overloading. Fixed ports were needed pre-SIP and pre-HTTP because there 
was no better alternative.  Applications decide on content, not the 
protocol. 


If the argument is that it cannot be done then the WEBDAV RFC must be 
deprecated. There is no alternative. 

If WebDav is allowed to remain then P2PSIP should be allowed to follow 
its lead. You just can't say yes to one and no to another, especially 
when the truth in the field is that well-known numbering no longer 
exists in practise except for DNS, SMTP, and HTTP/HTTPS.  

M 



Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada 

 


From dbarrett at quinthar.com  Thu Mar 16 21:34:37 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Thu, 16 Mar 2006 18:34:37 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31E963D@0591-ITS-EXMP02.us.saic.com>
Message-ID: <200603170234.k2H2YkOZ003451@cs.columbia.edu>

> -----Original Message-----
> From: Roy, Radhika R.
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> So, let me turn around the discussion a little differently as follows:
> 
> What would we do in developing the SIP protocol in the beginning if the
> number one requirement would have the crossing of the fw/nat by both
> signaling and media?

Interesting.  Well, this is certainly doable -- in fact, that's what I do
with iGlance: each client only listens on a single port.  Thus all SIP
requests and responses, all RTP/RTCP (across 4 different profiles), from all
peers, all comes in on a single port.  There's some trickiness in
classifying and dispatching incoming packets (need to look at the binary
signature to figure it out), but it works great.

The nice thing about this approach is you do NAT traversal exactly once per
peer, and then re-use that connection for everything.  Given how cumbersome,
non-deterministic, and costly it is to establish and maintain each
connection, it's a huge complexity, time, CPU, and bandwidth savings to
multiplex everything over a single connection.

And it can all be done without any on-the-wire protocol changes: the actual
SIP, RTP, and RTCP packets sent can be fully compliant with their
specifications.  They all just happen to arrive on the same port.

I *love* the idea of standardizing this behavior.  It works great for me and
my users.

-david


From huang-ming.pan at comcast.net  Fri Mar 17 12:49:52 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Fri, 17 Mar 2006 09:49:52 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
References: <200603170234.k2H2YkOZ003451@cs.columbia.edu>
Message-ID: <002901c649eb$32970ca0$6400a8c0@comcast.net>

> Interesting.  Well, this is certainly doable -- in fact, that's what I do
> with iGlance: each client only listens on a single port.  Thus all SIP
> requests and responses, all RTP/RTCP (across 4 different profiles), from
all
> peers, all comes in on a single port.  There's some trickiness in
> classifying and dispatching incoming packets (need to look at the binary
> signature to figure it out), but it works great.

Multiplexing packet streams of different types on one UDP port seems no
different from (and no more helpful than) on separate udp ports when a UA is
behind symmetric nat, regarding nat penetration.

When a node is behind symmetric nat, if its signaling msg can be forwarded
via one supernode while its media msg can be via another, the two msg
streams actually create two external port mappings.

Especially for p2p-sip nodes to be backward compatible with client/server
sip world, where proxy/ua/getway are normally separately located and anyway
there are different external port mappings for all types of packet streams,
standardizing one single port for all msg streams may hinder modular design
of p2p-sip applications.

peter
----- Original Message -----
From: "David Barrett" <dbarrett at quinthar.com>
To: "'Roy, Radhika R.'" <RADHIKA.R.ROY at saic.com>; "'Michael Slavitch'"
<slavitch at objectworld.com>; "'Greg Daley'" <gregd at research.panasonic.com>;
"'Michael Slavitch'" <slavitch at gmail.com>
Cc: "'P2P-SIP'" <p2p-sip at cs.columbia.edu>
Sent: Thursday, March 16, 2006 6:34 PM
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network


> > -----Original Message-----
> > From: Roy, Radhika R.
> > Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> >
> > So, let me turn around the discussion a little differently as follows:
> >
> > What would we do in developing the SIP protocol in the beginning if the
> > number one requirement would have the crossing of the fw/nat by both
> > signaling and media?
>
> Interesting.  Well, this is certainly doable -- in fact, that's what I do
> with iGlance: each client only listens on a single port.  Thus all SIP
> requests and responses, all RTP/RTCP (across 4 different profiles), from
all
> peers, all comes in on a single port.  There's some trickiness in
> classifying and dispatching incoming packets (need to look at the binary
> signature to figure it out), but it works great.
>
> The nice thing about this approach is you do NAT traversal exactly once
per
> peer, and then re-use that connection for everything.  Given how
cumbersome,
> non-deterministic, and costly it is to establish and maintain each
> connection, it's a huge complexity, time, CPU, and bandwidth savings to
> multiplex everything over a single connection.
>
> And it can all be done without any on-the-wire protocol changes: the
actual
> SIP, RTP, and RTCP packets sent can be fully compliant with their
> specifications.  They all just happen to arrive on the same port.
>
> I *love* the idea of standardizing this behavior.  It works great for me
and
> my users.
>
> -david
>
>
>


From dbarrett at quinthar.com  Fri Mar 17 14:32:23 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Fri, 17 Mar 2006 11:32:23 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <002901c649eb$32970ca0$6400a8c0@comcast.net>
Message-ID: <200603171932.k2HJWYOZ008261@cs.columbia.edu>

> -----Original Message-----
> From: Peter Pan [mailto:huang-ming.pan at comcast.net]
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> Multiplexing packet streams of different types on one UDP port seems no
> different from (and no more helpful than) on separate udp ports when a UA
> is
> behind symmetric nat, regarding nat penetration.

Hm, I'm not sure I follow you:

- Assuming it costs X (time, CPU, bandwidth, risk of failure) to establish a
connection between two peers, and 

- Assuming I need to establish 5 different connections between each pair of
peers (SIP, RTP-audio, RTP-video, RTP-filetransfer, RTP-screenshare), then

- It's 5X (longer, more expensive, more bandwidth-intensive, greater risk of
partial failure) to use separate UDP ports than to re-use the same port.

And this is true regardless of what kind of NAT or firewall you're behind
(obviously, the more complex the environment, the more advantageous
multiplexing becomes).

Am I misunderstanding what you were saying?  I'm saying it's better to do a
risky and expensive operation once, rather than five times.


> When a node is behind symmetric nat, if its signaling msg can be forwarded
> via one supernode while its media msg can be via another, the two msg
> streams actually create two external port mappings.

Perhaps this is the hangup: I'm suggesting that once the *direct* connection
is established between peers, *no* supernode does forwarding.

Again, relay is *not* the same as NAT penetration.  NAT penetration is the
art of establishing a direct P2P connection between *two* peers, without
relying upon a third middleman (except for the connection establishment, but
then he gets out of the picture).


> Especially for p2p-sip nodes to be backward compatible with client/server
> sip world, where proxy/ua/getway are normally separately located and
> anyway
> there are different external port mappings for all types of packet
> streams,
> standardizing one single port for all msg streams may hinder modular
> design
> of p2p-sip applications.

I'm not sure we're talking about the same thing, but I'm suggesting that
each UA be able to specify an arbitrary set of (ip:port) for its services.
A UA can optionally use different ports for all services like today, or it
can optionally use the same (ip:port) for all services.

Anyway, I'm just following Roy's suggestion that in a P2P environment when
the connection setup cost is so high, it's useful to reuse a single
(ip:port) for everything.  But I'm certainly not proposing that we mandate a
single (ip:port) for everyone.  I'm merely suggesting we might standardize
'best practices' along these lines to simplify and regularize SIP in a P2P
environment.

-david




From huang-ming.pan at comcast.net  Fri Mar 17 15:25:53 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Fri, 17 Mar 2006 12:25:53 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <005e01c64a00$fd3ce640$6400a8c0@comcast.net>

Maybe my limited knowledge of nat penetration leaves us on different pages.

Just few questions:

1. If peers behind all types of nats can establish direct connection between
each other *easily or anyway", why TURN comes up? (even Skype seems using
some kind of TURN relay)

2. If using two internal ports doubles risk of failure of connection
establishment, does it mean that, for two hosts behind a nat, after one host
gets connected (inward or outward), another will be double harder to get
connected??

3. If the overhead to open a udp socket can be ignored (it can be usually),
does it need double cpu/bandwidth to use the udp socket to communicate with
two peers?

4. From the sentence, "I *love* the idea of standardizing this behavior.  It
works great for me and my users", I thought one-port approach will loved to
be standardized. Sorry, I misunderstood your *love* :-)

I do (+ did) *love* the idea of port convergence, though. Actually when a
secure transport layer is necessarily beneath all packet streams for p2p
security, one-port approach is indeed as sweet as you love to suggest.

peter
----- Original Message -----
From: "David Barrett" <dbarrett at quinthar.com>
To: "'Peter Pan'" <huang-ming.pan at comcast.net>; "'Roy, Radhika R.'"
<RADHIKA.R.ROY at saic.com>; "'Michael Slavitch'" <slavitch at objectworld.com>;
"'Greg Daley'" <gregd at research.panasonic.com>; "'Michael Slavitch'"
<slavitch at gmail.com>
Cc: "'P2P-SIP'" <p2p-sip at cs.columbia.edu>
Sent: Friday, March 17, 2006 11:32 AM
Subject: RE: [p2p-sip] SIP, ICE, and an Overlay Network


> > -----Original Message-----
> > From: Peter Pan [mailto:huang-ming.pan at comcast.net]
> > Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> >
> > Multiplexing packet streams of different types on one UDP port seems no
> > different from (and no more helpful than) on separate udp ports when a
UA
> > is
> > behind symmetric nat, regarding nat penetration.
>
> Hm, I'm not sure I follow you:
>
> - Assuming it costs X (time, CPU, bandwidth, risk of failure) to establish
a
> connection between two peers, and
>
> - Assuming I need to establish 5 different connections between each pair
of
> peers (SIP, RTP-audio, RTP-video, RTP-filetransfer, RTP-screenshare), then
>
> - It's 5X (longer, more expensive, more bandwidth-intensive, greater risk
of
> partial failure) to use separate UDP ports than to re-use the same port.
>
> And this is true regardless of what kind of NAT or firewall you're behind
> (obviously, the more complex the environment, the more advantageous
> multiplexing becomes).
>
> Am I misunderstanding what you were saying?  I'm saying it's better to do
a
> risky and expensive operation once, rather than five times.
>
>
> > When a node is behind symmetric nat, if its signaling msg can be
forwarded
> > via one supernode while its media msg can be via another, the two msg
> > streams actually create two external port mappings.
>
> Perhaps this is the hangup: I'm suggesting that once the *direct*
connection
> is established between peers, *no* supernode does forwarding.
>
> Again, relay is *not* the same as NAT penetration.  NAT penetration is the
> art of establishing a direct P2P connection between *two* peers, without
> relying upon a third middleman (except for the connection establishment,
but
> then he gets out of the picture).
>
>
> > Especially for p2p-sip nodes to be backward compatible with
client/server
> > sip world, where proxy/ua/getway are normally separately located and
> > anyway
> > there are different external port mappings for all types of packet
> > streams,
> > standardizing one single port for all msg streams may hinder modular
> > design
> > of p2p-sip applications.
>
> I'm not sure we're talking about the same thing, but I'm suggesting that
> each UA be able to specify an arbitrary set of (ip:port) for its services.
> A UA can optionally use different ports for all services like today, or it
> can optionally use the same (ip:port) for all services.
>
> Anyway, I'm just following Roy's suggestion that in a P2P environment when
> the connection setup cost is so high, it's useful to reuse a single
> (ip:port) for everything.  But I'm certainly not proposing that we mandate
a
> single (ip:port) for everyone.  I'm merely suggesting we might standardize
> 'best practices' along these lines to simplify and regularize SIP in a P2P
> environment.
>
> -david
>
>
>


From huang-ming.pan at comcast.net  Fri Mar 17 16:44:18 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Fri, 17 Mar 2006 13:44:18 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <006e01c64a0b$f271fc90$6400a8c0@comcast.net>


For p2p-sip to be accepted by enterprises where there are often hundreds of
nodes behind (normally symmetric) nats, do we need to standardize the
"radius" of probabilistic port-scan so that their network administrators
won't treat port scanning by p2p-sip nodes as a security threat?

Does it seem also better to recognize the multiplexing media streams with
SIP signaling stream as a valid and encouraged option in next RFC for
(non-p2p) SIP/3.0 :-)

peter
----- Original Message -----
From: "David Barrett" <dbarrett at quinthar.com>
To: "'Peter Pan'" <huang-ming.pan at comcast.net>; "'Roy, Radhika R.'"
<RADHIKA.R.ROY at saic.com>; "'Michael Slavitch'" <slavitch at objectworld.com>;
"'Greg Daley'" <gregd at research.panasonic.com>; "'Michael Slavitch'"
<slavitch at gmail.com>
Cc: "'P2P-SIP'" <p2p-sip at cs.columbia.edu>
Sent: Friday, March 17, 2006 1:03 PM
Subject: RE: [p2p-sip] SIP, ICE, and an Overlay Network


> > -----Original Message-----
> > From: Peter Pan [mailto:huang-ming.pan at comcast.net]
> > Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> >
> > Maybe my limited knowledge of nat penetration leaves us on different
> > pages.
> >
> > Just few questions:
> >
> > 1. If peers behind all types of nats can establish direct connection
> > between
> > each other *easily or anyway", why TURN comes up? (even Skype seems
using
> > some kind of TURN relay)
>
> Even the best NAT penetration isn't 100% reliable, so the TURN service is
> the fall back plan.  But NAT penetration over 75% is straightforward, and
> high 90% is possible -- even >99% has been claimed.
>
>
> > 2. If using two internal ports doubles risk of failure of connection
> > establishment, does it mean that, for two hosts behind a nat, after one
> > host
> > gets connected (inward or outward), another will be double harder to get
> > connected??
>
> Ah, no I'm referring to the prospect of partial failure.  For example, in
> iGlance you can text-chat, voice-chat, videoconference, screen share, and
> send files.  Were each of these using a different connection, the risk
that
> at least one of these channels fails to penetrate the NAT goes up.  And
then
> you're faced with the problem where your video channel is direct P2P,
while
> your audio channel is relayed through a TURN server.  This leads to all
> sorts of odd behavior for the user.
>
> By having everything either completely work or completely fail, it gives a
> cleaner experience for the user.
>
>
> > 3. If the overhead to open a udp socket can be ignored (it can be
> > usually),
> > does it need double cpu/bandwidth to use the udp socket to communicate
> > with
> > two peers?
>
> The setup cost isn't so much in creating UDP sockets, it's in conducting
the
> connection process.  In the worst case of symmetric NATs, this can involve
> doing a probabilistic port-scan of the remote host.  Thus setting up one
> connection between two peers could conceivably involve hundreds of packets
> going back and forth over many seconds time.
>
>
> > 4. From the sentence, "I *love* the idea of standardizing this behavior.
> > It
> > works great for me and my users", I thought one-port approach will loved
> > to
> > be standardized. Sorry, I misunderstood your *love* :-)
> >
> > I do (+ did) *love* the idea of port convergence, though. Actually when
a
> > secure transport layer is necessarily beneath all packet streams for p2p
> > security, one-port approach is indeed as sweet as you love to suggest.
>
> Heh, I guess what I meant is I'd like my beloved approach to be recognized
> as a valid and encouraged option, but by no means the only option.
>
> -david
>


From dbarrett at quinthar.com  Fri Mar 17 16:03:53 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Fri, 17 Mar 2006 13:03:53 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <005e01c64a00$fd3ce640$6400a8c0@comcast.net>
Message-ID: <200603172216.k2HMG9rj016286@cs.columbia.edu>

> -----Original Message-----
> From: Peter Pan [mailto:huang-ming.pan at comcast.net]
> Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> 
> Maybe my limited knowledge of nat penetration leaves us on different
> pages.
> 
> Just few questions:
> 
> 1. If peers behind all types of nats can establish direct connection
> between
> each other *easily or anyway", why TURN comes up? (even Skype seems using
> some kind of TURN relay)

Even the best NAT penetration isn't 100% reliable, so the TURN service is
the fall back plan.  But NAT penetration over 75% is straightforward, and
high 90% is possible -- even >99% has been claimed.


> 2. If using two internal ports doubles risk of failure of connection
> establishment, does it mean that, for two hosts behind a nat, after one
> host
> gets connected (inward or outward), another will be double harder to get
> connected??

Ah, no I'm referring to the prospect of partial failure.  For example, in
iGlance you can text-chat, voice-chat, videoconference, screen share, and
send files.  Were each of these using a different connection, the risk that
at least one of these channels fails to penetrate the NAT goes up.  And then
you're faced with the problem where your video channel is direct P2P, while
your audio channel is relayed through a TURN server.  This leads to all
sorts of odd behavior for the user.

By having everything either completely work or completely fail, it gives a
cleaner experience for the user.


> 3. If the overhead to open a udp socket can be ignored (it can be
> usually),
> does it need double cpu/bandwidth to use the udp socket to communicate
> with
> two peers?

The setup cost isn't so much in creating UDP sockets, it's in conducting the
connection process.  In the worst case of symmetric NATs, this can involve
doing a probabilistic port-scan of the remote host.  Thus setting up one
connection between two peers could conceivably involve hundreds of packets
going back and forth over many seconds time.


> 4. From the sentence, "I *love* the idea of standardizing this behavior.
> It
> works great for me and my users", I thought one-port approach will loved
> to
> be standardized. Sorry, I misunderstood your *love* :-)
> 
> I do (+ did) *love* the idea of port convergence, though. Actually when a
> secure transport layer is necessarily beneath all packet streams for p2p
> security, one-port approach is indeed as sweet as you love to suggest.

Heh, I guess what I meant is I'd like my beloved approach to be recognized
as a valid and encouraged option, but by no means the only option.

-david


From dean.willis at softarmor.com  Fri Mar 17 18:31:32 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Fri, 17 Mar 2006 17:31:32 -0600
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <17433.20793.191367.804402@rautu.tutpro.com>
References: <200603150909.k2F997OZ026285@cs.columbia.edu>
	<44189C36.8000706@cisco.com>
	<17433.20793.191367.804402@rautu.tutpro.com>
Message-ID: <476B50F8-A426-4A81-AFEA-B6928EF80A9F@softarmor.com>


On Mar 16, 2006, at 5:51 AM, Juha Heinanen wrote:

> jonathan,
>
> i read the outbound draft (not all glory details yet) and i was  
> pleased
> with it.  the main reason being that in addition to NAT traversal  
> (which
> openser can handle already now satisfactorily without any new
> standardization), it also tackles the redundancy problem.
>
> there is one thing though that i can't accept:
>
>    UAs are configured with one or more SIP URIs representing the  
> default
>    outbound-proxy-set.  The specification assumes the set is  
> determined
>    via configuration but future specifications may define other
>    mechanisms such as using DNS to discover this set.
>
> we MUST have a DNS based way for the UAs to figure out the default
> outbound proxy set BEFORE this spec is published as an RFC.  if not
> done, there is yet another thing to be configured in the UA, which is
> exactly what Skype lovers hate in SIP.
>

I think there is some overlap here.

The UA has to have SOME sort of configured information, or it has no  
place to start.

RFC 3263 already defines how a UA can use DNS to find a proxy, based  
on the name of the proxy (if configured) or a registrar (based on the  
domain of registration, if configured). If you don't have SOME of  
this information in the UA, there's no clue where to start looking.

The problem is that DNS is NOT going to tell you about the local  
firewall proxy built into your residential router/firewall/access  
point/media server box.

--
Dean


From jh at tutpro.com  Sat Mar 18 03:05:49 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Sat, 18 Mar 2006 10:05:49 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <476B50F8-A426-4A81-AFEA-B6928EF80A9F@softarmor.com>
References: <200603150909.k2F997OZ026285@cs.columbia.edu>
	<44189C36.8000706@cisco.com>
	<17433.20793.191367.804402@rautu.tutpro.com>
	<476B50F8-A426-4A81-AFEA-B6928EF80A9F@softarmor.com>
Message-ID: <17435.48989.650198.676208@rautu.tutpro.com>

Dean Willis writes:

 > The UA has to have SOME sort of configured information, or it has no  
 > place to start.

yes, AoR, username and password.

 > RFC 3263 already defines how a UA can use DNS to find a proxy, based  
 > on the name of the proxy (if configured) or a registrar (based on the  
 > domain of registration, if configured). If you don't have SOME of  
 > this information in the UA, there's no clue where to start looking.

today, most SIP UAs query DNS based on host part of AoR and this has
worked very well in practise.  the same should work for
outbound-proxy-set.  if you have some special requirements, a good UA
lets you override the default behavior.

-- juha

From huang-ming.pan at comcast.net  Sat Mar 18 11:56:19 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Sat, 18 Mar 2006 08:56:19 -0800
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
Message-ID: <001101c64aac$e1b20b10$6400a8c0@comcast.net>

Though, with cost of time and bandwidth, predictive or probabilistic
port-scan can shoot msg to a node behind a nat eventually, is it possible
that such port-scan can interfere other nodes or existing sessions behind
enterprise firewalls which often have many nodes (sessions) any time so that
have created many "penetratable" holes on the firewalls?

That is, even if an inward probing msg can penetrate a firewall after some
attempts, it might visit many wrong peers (sessions) before it reaches the
right one.

Also, if there are N nodes behind firewall A and N nodes behind firewall B
scan (shoot) each other simultaneously trying to setup N 1-to-1 sessions,
the match can have N*N combinations. Will p2p-sip WG define some protocol to
resolve the possible N*(N-1) "ill-matched" connections?

Or, these are another stupid questions besides the one which questioned how
to interoperate nodes of a p2p-sip Chord overlay with those of a p2p-sip
Kademlia overlay -- peered or bridged???

peter
----- Original Message -----
From: "Peter Pan" <huang-ming.pan at comcast.net>
To: "David Barrett" <dbarrett at quinthar.com>; "'Roy, Radhika R.'"
<RADHIKA.R.ROY at saic.com>; "'Michael Slavitch'" <slavitch at objectworld.com>;
"'Greg Daley'" <gregd at research.panasonic.com>; "'Michael Slavitch'"
<slavitch at gmail.com>
Cc: "'P2P-SIP'" <p2p-sip at cs.columbia.edu>
Sent: Friday, March 17, 2006 1:44 PM
Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network


>
> For p2p-sip to be accepted by enterprises where there are often hundreds
of
> nodes behind (normally symmetric) nats, do we need to standardize the
> "radius" of probabilistic port-scan so that their network administrators
> won't treat port scanning by p2p-sip nodes as a security threat?
>
> Does it seem also better to recognize the multiplexing media streams with
> SIP signaling stream as a valid and encouraged option in next RFC for
> (non-p2p) SIP/3.0 :-)
>
> peter
> ----- Original Message -----
> From: "David Barrett" <dbarrett at quinthar.com>
> To: "'Peter Pan'" <huang-ming.pan at comcast.net>; "'Roy, Radhika R.'"
> <RADHIKA.R.ROY at saic.com>; "'Michael Slavitch'" <slavitch at objectworld.com>;
> "'Greg Daley'" <gregd at research.panasonic.com>; "'Michael Slavitch'"
> <slavitch at gmail.com>
> Cc: "'P2P-SIP'" <p2p-sip at cs.columbia.edu>
> Sent: Friday, March 17, 2006 1:03 PM
> Subject: RE: [p2p-sip] SIP, ICE, and an Overlay Network
>
>
> > > -----Original Message-----
> > > From: Peter Pan [mailto:huang-ming.pan at comcast.net]
> > > Subject: Re: [p2p-sip] SIP, ICE, and an Overlay Network
> > >
> > > Maybe my limited knowledge of nat penetration leaves us on different
> > > pages.
> > >
> > > Just few questions:
> > >
> > > 1. If peers behind all types of nats can establish direct connection
> > > between
> > > each other *easily or anyway", why TURN comes up? (even Skype seems
> using
> > > some kind of TURN relay)
> >
> > Even the best NAT penetration isn't 100% reliable, so the TURN service
is
> > the fall back plan.  But NAT penetration over 75% is straightforward,
and
> > high 90% is possible -- even >99% has been claimed.
> >
> >
> > > 2. If using two internal ports doubles risk of failure of connection
> > > establishment, does it mean that, for two hosts behind a nat, after
one
> > > host
> > > gets connected (inward or outward), another will be double harder to
get
> > > connected??
> >
> > Ah, no I'm referring to the prospect of partial failure.  For example,
in
> > iGlance you can text-chat, voice-chat, videoconference, screen share,
and
> > send files.  Were each of these using a different connection, the risk
> that
> > at least one of these channels fails to penetrate the NAT goes up.  And
> then
> > you're faced with the problem where your video channel is direct P2P,
> while
> > your audio channel is relayed through a TURN server.  This leads to all
> > sorts of odd behavior for the user.
> >
> > By having everything either completely work or completely fail, it gives
a
> > cleaner experience for the user.
> >
> >
> > > 3. If the overhead to open a udp socket can be ignored (it can be
> > > usually),
> > > does it need double cpu/bandwidth to use the udp socket to communicate
> > > with
> > > two peers?
> >
> > The setup cost isn't so much in creating UDP sockets, it's in conducting
> the
> > connection process.  In the worst case of symmetric NATs, this can
involve
> > doing a probabilistic port-scan of the remote host.  Thus setting up one
> > connection between two peers could conceivably involve hundreds of
packets
> > going back and forth over many seconds time.
> >
> >
> > > 4. From the sentence, "I *love* the idea of standardizing this
behavior.
> > > It
> > > works great for me and my users", I thought one-port approach will
loved
> > > to
> > > be standardized. Sorry, I misunderstood your *love* :-)
> > >
> > > I do (+ did) *love* the idea of port convergence, though. Actually
when
> a
> > > secure transport layer is necessarily beneath all packet streams for
p2p
> > > security, one-port approach is indeed as sweet as you love to suggest.
> >
> > Heh, I guess what I meant is I'd like my beloved approach to be
recognized
> > as a valid and encouraged option, but by no means the only option.
> >
> > -david
> >
>



From enrico.marocco at telecomitalia.it  Sun Mar 19 14:59:05 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Sun, 19 Mar 2006 13:59:05 -0600
Subject: [p2p-sip] Running code in Dallas
Message-ID: <441DB809.5060406@telecomitalia.it>

Folks,

I'm looking for someone who's done with slideware and would like to play 
with /real/ software ;-)

In the past few days I've been able to set up a little project for 
implementing David's draft; right now it is at a very early stage, but 
it still implements some (maybe broken) functionalities and - most 
important - could be released under GPL.  Even though it doesn't deal 
with issues which are "hot" on this ml (e.g. NAT traversal...), I think 
that such a kind of shared work is essential for getting the feeling of 
what we really want to do.

The project consists of a module in the cvs server hosted on my laptop; 
unfortunatly, at the moment I can't access any public server for sharing 
it, so, if you are interested, you have to directly contact me (by mail 
or live at sip:me at sipdemo.tilab.com, jid:enrico at tilab.com; 
tel+393357940115, in strict reliability and cost order).

Ah... it works on linux and theoretically on macos using Nokia's sofia-sip.

Enrico
--------------------------------------------------------------------

CONFIDENTIALITY NOTICE

This message and its attachments are addressed solely to the persons above and may contain confidential information. If you have received the message in error, be informed that any use of the content hereof is prohibited. Please return it immediately to the sender and delete the message. Should you have any questions, please contact us by replying to webmaster at telecomitalia.it.

        Thank you

                                        www.telecomitalia.it

--------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060319/1fd929b4/attachment.html

From philip_matthews at magma.ca  Mon Mar 20 19:36:26 2006
From: philip_matthews at magma.ca (Philip Matthews)
Date: Mon, 20 Mar 2006 18:36:26 -0600
Subject: [p2p-sip] NATs and P2P
Message-ID: <9151EA5D-BA55-4A69-8F6A-7118492C4283@magma.ca>

[I guess all this traffic on the mailing list will teach me not to go  
on vacation
just before an IETF meeting!]

As I catch up on all the messages about NATs over the past two weeks,
it seems to me that many people are thinking only of the "Public P2P  
Service Provider"
use-case as described in the use-cases document. In other words, a Skype
competitor.

However, the use-cases document that David Bryan and his co-authors  
wrote
identified a number of other use-cases and it seems to me that these  
have somewhat
different NAT traversal requirements.

In particular, in some of these other use-cases, it seems to me that  
we CANNOT assume
there are peers with public IP addresses.

For example, consider the "Presence using Multimedia Consumer  
Electronics Devices"
use-case (section 3.1.3) -- essentially a P2P network of multimedia  
consumer electronics devices
that need presence information. Who is going to pay the extra money  
to give their digital camera (or those
neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4  
address?? On the contrary,
devices like this are almost certainly going to have private IP  
addresses -- it is very common today for
wireless internet providers to place a big NAT in front of their  
entire network and give private addresses
to all their customers.

Or consider the "IP PBX" use-case -- a IP PBX system for a company  
with a number of small branches
scattered throughout the world. Each branch is going to have a NAT in  
front of its network, and all
the phones in that branch are going to have private IP addresses.  
None of the phones are going to
have public IP addresses.

It is handling the NAT traversal issues for use-cases like these that  
Eric Cooper and I wrote our
internet-draft on NAT Traversal for P2P:
       http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
and-overlays-00.txt

- Philip


From lowekamp at cs.wm.edu  Mon Mar 20 20:58:02 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Mon, 20 Mar 2006 19:58:02 -0600
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <9151EA5D-BA55-4A69-8F6A-7118492C4283@magma.ca>
References: <9151EA5D-BA55-4A69-8F6A-7118492C4283@magma.ca>
Message-ID: <e9132f820603201758n727d6d25p62f4c4d60dac42d9@mail.gmail.com>

Philip,

I absolutely agree that we need to have a solution for this type of
environment.   IMHO, some of the worries about NATs causing problems
have been somewhat overstated.  As you point out, the majority of NATs
in the wild are address independent mapping, and it is relatively
trivial to punch holes to establish communications.  But you're also
right that, especially for corporate networks, this doesn't solve the
entire problem.

I spent some time thinking about how to solve the specific problems
you pose in terms of the protocol we have been playing with in
draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
there (not because we don't think it's a challenge, but just because
we want to work on a basic protocol first).  If I was to address some
of the problems you pose using that framework (modified chord), then
one approach would be to remove the requirement that DHT operations be
iterative and allow them to be recursive.  If DHT operations can be
recursive, then the number of connections that a node must maintain
through the firewall is limited by the size of the finger table.  That
number doesn't have to be very big to have a rather efficient DHT. 
(never more than 128, and I think in practice the number can be much
smaller.  If your guess of 160 is correct, we're more than OK here.)

I'm a bit unhappy with that because I really like the iterative
approach for a variety of reasons, but certainly the protocol could be
modified that way and a hybrid environment could be established where
some nodes are open (or easy to request opens) and support iterative
queries with anyone, but others are recursive queries only.  One
problem is that the burden of having a finger-table entry to a
recursive-only host is placed on the non-recursive node that would
otherwise redirect to its finger-table entry, but now must recurse the
request.

Right now the scheme would also have problems because, as Eric rightly
points out in a separate thread on this mailing list, we don't have a
good way of selecting Identifiers for nodes behind NATs (in
particular).  But if you have node certificates or another way of
authenticating your nodes, this is less of a problem.

I want to make it clear that I'm not ready to propose a specific
solution, but I think there are ways to support structured DHT
algorithms that are not really very painful and work reasonably well
inside these environments.  Clearly support for such an environment
needs to be a listed in eventual requirements docs.

Bruce


On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> [I guess all this traffic on the mailing list will teach me not to go
> on vacation
> just before an IETF meeting!]
>
> As I catch up on all the messages about NATs over the past two weeks,
> it seems to me that many people are thinking only of the "Public P2P
> Service Provider"
> use-case as described in the use-cases document. In other words, a Skype
> competitor.
>
> However, the use-cases document that David Bryan and his co-authors
> wrote
> identified a number of other use-cases and it seems to me that these
> have somewhat
> different NAT traversal requirements.
>
> In particular, in some of these other use-cases, it seems to me that
> we CANNOT assume
> there are peers with public IP addresses.
>
> For example, consider the "Presence using Multimedia Consumer
> Electronics Devices"
> use-case (section 3.1.3) -- essentially a P2P network of multimedia
> consumer electronics devices
> that need presence information. Who is going to pay the extra money
> to give their digital camera (or those
> neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4
> address?? On the contrary,
> devices like this are almost certainly going to have private IP
> addresses -- it is very common today for
> wireless internet providers to place a big NAT in front of their
> entire network and give private addresses
> to all their customers.
>
> Or consider the "IP PBX" use-case -- a IP PBX system for a company
> with a number of small branches
> scattered throughout the world. Each branch is going to have a NAT in
> front of its network, and all
> the phones in that branch are going to have private IP addresses.
> None of the phones are going to
> have public IP addresses.
>
> It is handling the NAT traversal issues for use-cases like these that
> Eric Cooper and I wrote our
> internet-draft on NAT Traversal for P2P:
>        http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats-
> and-overlays-00.txt
>
> - Philip
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>


From philip_matthews at magma.ca  Mon Mar 20 20:25:38 2006
From: philip_matthews at magma.ca (Philip Matthews)
Date: Mon, 20 Mar 2006 19:25:38 -0600
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <500E3F6B-96A7-48F3-B2C5-71AAB9031064@magma.ca>

In advance of Wednesday's BOF session, I thought I should
send out a few comments on the use-cases document
http://www.ietf.org/internet-drafts/draft-bryan-sipping-p2p- 
usecases-00.txt

First of all, I think that having such a document is a great idea.
We can use this document to pick the use-cases that we want to support.

Some comments:

* I was initially confused when reading section 3. After reading a  
few times,
it seems to me that some of the sections describe a network type
(e.g., "global Internet" or "Limited Internet Connectivity") while  
others describe
a particular P2P use-case. I would be helpful if the document made  
the distinction
between these two types of descriptions clearer.

* I think it would be helpful to describe each use-case in more  
detail, and
make it clear how the use-cases differ from each other.

* Section 3.1.1 "Public P2P VoIP Service Providers" doesn't really  
describe
this use-case -- right now, it is mostly a few comments on Skype
(which seem out-of-place). I might suggest that this use-case could
be characterized as follows:
-      Very large numbers of peers (millions)
-      Some may be behind NATs, while others may have public IP  
addresses
-      Protection against malicious peers is very important
-      Connection to PSTN and CS SIP is important

* What is the distinction between 3.1.1 Public P2P VoIP Service  
Providers"
and 3.1.2  "Open Global P2P VoIP Network"??  Is the only difference  
the fact
that in 3.1.1, there is an organization who determines who is allowed  
to join the
overlay, while in 3.1.2, there is no such organization?

* I would like to add another use-case under 3.1 "Global Internet  
Environment".
This use-case is a IP PBX for a company which has a number of  
branches scattered
throughout the world. Each branch consists of a number of phones, and  
each branch
has a NAT or Firewall box connecting its private network to the  
Internet.
The phones in all the branches connect together into a single P2P PBX  
system.
This use-case differs from 3.4.1 because in this use-case some phones  
must
connect to each other via the Internet.

* Finally, I find calling these "use-cases" a bit strange. They seem  
much too high-level
and vague to be anything like what the term "use-case" means in UML.
I might suggest a term like "application scenarios" might be more  
appropriate.


- Philip


From dbarrett at quinthar.com  Mon Mar 20 22:19:13 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Mon, 20 Mar 2006 19:19:13 -0800
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <500E3F6B-96A7-48F3-B2C5-71AAB9031064@magma.ca>
Message-ID: <200603210319.k2L3JIrj001729@cs.columbia.edu>

I'd add to that this key question:

	"Will we extend SIP, or create a new protocol?"

I'm finding it hard to make any headway without understanding the above.
Basically, I see two major directions we could go:

1) Extend SIP with overlay-maintenance and resource-location messages.

2) Create a new overlay protocol and develop bindings for SIP and ICE (eg,
distributed proxy service and STUN/TURN resource-location service).

It seems this high-level decision keeps coming up again and again in
discussions of the smaller issues.

-david


> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-
> bounces at cs.columbia.edu] On Behalf Of Philip Matthews
> Sent: Monday, March 20, 2006 5:26 PM
> To: P2P-SIP
> Subject: [p2p-sip] Some comments on Use-cases document
> 
> In advance of Wednesday's BOF session, I thought I should
> send out a few comments on the use-cases document
> http://www.ietf.org/internet-drafts/draft-bryan-sipping-p2p-
> usecases-00.txt
> 
> First of all, I think that having such a document is a great idea.
> We can use this document to pick the use-cases that we want to support.
> 
> Some comments:
> 
> * I was initially confused when reading section 3. After reading a
> few times,
> it seems to me that some of the sections describe a network type
> (e.g., "global Internet" or "Limited Internet Connectivity") while
> others describe
> a particular P2P use-case. I would be helpful if the document made
> the distinction
> between these two types of descriptions clearer.
> 
> * I think it would be helpful to describe each use-case in more
> detail, and
> make it clear how the use-cases differ from each other.
> 
> * Section 3.1.1 "Public P2P VoIP Service Providers" doesn't really
> describe
> this use-case -- right now, it is mostly a few comments on Skype
> (which seem out-of-place). I might suggest that this use-case could
> be characterized as follows:
> -      Very large numbers of peers (millions)
> -      Some may be behind NATs, while others may have public IP
> addresses
> -      Protection against malicious peers is very important
> -      Connection to PSTN and CS SIP is important
> 
> * What is the distinction between 3.1.1 Public P2P VoIP Service
> Providers"
> and 3.1.2  "Open Global P2P VoIP Network"??  Is the only difference
> the fact
> that in 3.1.1, there is an organization who determines who is allowed
> to join the
> overlay, while in 3.1.2, there is no such organization?
> 
> * I would like to add another use-case under 3.1 "Global Internet
> Environment".
> This use-case is a IP PBX for a company which has a number of
> branches scattered
> throughout the world. Each branch consists of a number of phones, and
> each branch
> has a NAT or Firewall box connecting its private network to the
> Internet.
> The phones in all the branches connect together into a single P2P PBX
> system.
> This use-case differs from 3.4.1 because in this use-case some phones
> must
> connect to each other via the Internet.
> 
> * Finally, I find calling these "use-cases" a bit strange. They seem
> much too high-level
> and vague to be anything like what the term "use-case" means in UML.
> I might suggest a term like "application scenarios" might be more
> appropriate.
> 
> 
> - Philip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From effis at dsp.co.il  Tue Mar 21 03:22:16 2006
From: effis at dsp.co.il (Effi Shiri)
Date: Tue, 21 Mar 2006 10:22:16 +0200
Subject: [p2p-sip] NATs and P2P
Message-ID: <216A12DDD3490049BA3E56BAC324F285012FB007@ILEXCH2003.il-prod.dspcorp.com>


This is very interesting
Do you have statistics on how many Home users actually have Public IP?


-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Philip Matthews
Sent: Tuesday, March 21, 2006 2:36 AM
To: P2P-SIP
Subject: [p2p-sip] NATs and P2P

[I guess all this traffic on the mailing list will teach me not to go  
on vacation
just before an IETF meeting!]

As I catch up on all the messages about NATs over the past two weeks,
it seems to me that many people are thinking only of the "Public P2P  
Service Provider"
use-case as described in the use-cases document. In other words, a Skype
competitor.

However, the use-cases document that David Bryan and his co-authors  
wrote
identified a number of other use-cases and it seems to me that these  
have somewhat
different NAT traversal requirements.

In particular, in some of these other use-cases, it seems to me that  
we CANNOT assume
there are peers with public IP addresses.

For example, consider the "Presence using Multimedia Consumer  
Electronics Devices"
use-case (section 3.1.3) -- essentially a P2P network of multimedia  
consumer electronics devices
that need presence information. Who is going to pay the extra money  
to give their digital camera (or those
neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4  
address?? On the contrary,
devices like this are almost certainly going to have private IP  
addresses -- it is very common today for
wireless internet providers to place a big NAT in front of their  
entire network and give private addresses
to all their customers.

Or consider the "IP PBX" use-case -- a IP PBX system for a company  
with a number of small branches
scattered throughout the world. Each branch is going to have a NAT in  
front of its network, and all
the phones in that branch are going to have private IP addresses.  
None of the phones are going to
have public IP addresses.

It is handling the NAT traversal issues for use-cases like these that  
Eric Cooper and I wrote our
internet-draft on NAT Traversal for P2P:
       http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
and-overlays-00.txt

- Philip

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip

______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________


From l4hurd at 163.com  Tue Mar 21 06:44:12 2006
From: l4hurd at 163.com (Zhou Ya Jin)
Date: Tue, 21 Mar 2006 19:44:12 +0800
Subject: [p2p-sip] NATs and P2P
References: <216A12DDD3490049BA3E56BAC324F285012FB007@ILEXCH2003.il-prod.dspcorp.com>
Message-ID: <000a01c64cdc$c726e950$7b01a8c0@killbill>

Actually in China, many internet user does not have public IP. As far as I am concerned, people in my office share a public IP. We use a router for NAT.
----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 4:22 PM
Subject: Re: [p2p-sip] NATs and P2P


> 
> 
> This is very interesting
> Do you have statistics on how many Home users actually have Public IP?
> 
> 
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Philip Matthews
> Sent: Tuesday, March 21, 2006 2:36 AM
> To: P2P-SIP
> Subject: [p2p-sip] NATs and P2P
> 
> [I guess all this traffic on the mailing list will teach me not to go  
> on vacation
> just before an IETF meeting!]
> 
> As I catch up on all the messages about NATs over the past two weeks,
> it seems to me that many people are thinking only of the "Public P2P  
> Service Provider"
> use-case as described in the use-cases document. In other words, a Skype
> competitor.
> 
> However, the use-cases document that David Bryan and his co-authors  
> wrote
> identified a number of other use-cases and it seems to me that these  
> have somewhat
> different NAT traversal requirements.
> 
> In particular, in some of these other use-cases, it seems to me that  
> we CANNOT assume
> there are peers with public IP addresses.
> 
> For example, consider the "Presence using Multimedia Consumer  
> Electronics Devices"
> use-case (section 3.1.3) -- essentially a P2P network of multimedia  
> consumer electronics devices
> that need presence information. Who is going to pay the extra money  
> to give their digital camera (or those
> neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4  
> address?? On the contrary,
> devices like this are almost certainly going to have private IP  
> addresses -- it is very common today for
> wireless internet providers to place a big NAT in front of their  
> entire network and give private addresses
> to all their customers.
> 
> Or consider the "IP PBX" use-case -- a IP PBX system for a company  
> with a number of small branches
> scattered throughout the world. Each branch is going to have a NAT in  
> front of its network, and all
> the phones in that branch are going to have private IP addresses.  
> None of the phones are going to
> have public IP addresses.
> 
> It is handling the NAT traversal issues for use-cases like these that  
> Eric Cooper and I wrote our
> internet-draft on NAT Traversal for P2P:
>       http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
> and-overlays-00.txt
> 
> - Philip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> ______________________________________________________________________
> DSP Group LTD. automatically scans all emails and attachments using
> MessageLabs Email Security System.
> ______________________________________________________________________
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> 
> __________ NOD32 1.1452 (20060320) Information __________
> 
> This message was checked by NOD32 antivirus system.
> http://www.eset.com
> 
>


From effis at dsp.co.il  Tue Mar 21 06:59:32 2006
From: effis at dsp.co.il (Effi Shiri)
Date: Tue, 21 Mar 2006 13:59:32 +0200
Subject: [p2p-sip] NATs and P2P
Message-ID: <216A12DDD3490049BA3E56BAC324F285012FB018@ILEXCH2003.il-prod.dspcorp.com>

Thanks Zhou!

So if most people do not have Public addresses, I guess the whole p2p is
not p2p anymore? 



-----Original Message-----
From: Zhou Ya Jin [mailto:l4hurd at 163.com] 
Sent: Tuesday, March 21, 2006 1:44 PM
To: Effi Shiri; p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] NATs and P2P

Actually in China, many internet user does not have public IP. As far as
I am concerned, people in my office share a public IP. We use a router
for NAT.
----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 4:22 PM
Subject: Re: [p2p-sip] NATs and P2P


> 
> 
> This is very interesting
> Do you have statistics on how many Home users actually have Public IP?
> 
> 
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Philip Matthews
> Sent: Tuesday, March 21, 2006 2:36 AM
> To: P2P-SIP
> Subject: [p2p-sip] NATs and P2P
> 
> [I guess all this traffic on the mailing list will teach me not to go

> on vacation
> just before an IETF meeting!]
> 
> As I catch up on all the messages about NATs over the past two weeks,
> it seems to me that many people are thinking only of the "Public P2P  
> Service Provider"
> use-case as described in the use-cases document. In other words, a
Skype
> competitor.
> 
> However, the use-cases document that David Bryan and his co-authors  
> wrote
> identified a number of other use-cases and it seems to me that these  
> have somewhat
> different NAT traversal requirements.
> 
> In particular, in some of these other use-cases, it seems to me that  
> we CANNOT assume
> there are peers with public IP addresses.
> 
> For example, consider the "Presence using Multimedia Consumer  
> Electronics Devices"
> use-case (section 3.1.3) -- essentially a P2P network of multimedia  
> consumer electronics devices
> that need presence information. Who is going to pay the extra money  
> to give their digital camera (or those
> neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4

> address?? On the contrary,
> devices like this are almost certainly going to have private IP  
> addresses -- it is very common today for
> wireless internet providers to place a big NAT in front of their  
> entire network and give private addresses
> to all their customers.
> 
> Or consider the "IP PBX" use-case -- a IP PBX system for a company  
> with a number of small branches
> scattered throughout the world. Each branch is going to have a NAT in

> front of its network, and all
> the phones in that branch are going to have private IP addresses.  
> None of the phones are going to
> have public IP addresses.
> 
> It is handling the NAT traversal issues for use-cases like these that

> Eric Cooper and I wrote our
> internet-draft on NAT Traversal for P2P:
>       http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
> and-overlays-00.txt
> 
> - Philip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> ______________________________________________________________________
> DSP Group LTD. automatically scans all emails and attachments using
> MessageLabs Email Security System.
> ______________________________________________________________________
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> 
> __________ NOD32 1.1452 (20060320) Information __________
> 
> This message was checked by NOD32 antivirus system.
> http://www.eset.com
> 
>

______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________


From l4hurd at 163.com  Tue Mar 21 07:16:25 2006
From: l4hurd at 163.com (Zhou Ya Jin)
Date: Tue, 21 Mar 2006 20:16:25 +0800
Subject: [p2p-sip] NATs and P2P
References: <216A12DDD3490049BA3E56BAC324F285012FB018@ILEXCH2003.il-prod.dspcorp.com>
Message-ID: <001701c64ce1$470d89e0$7b01a8c0@killbill>

As far as I know, in China, we use  NAT_traversal to connect two parts of P2P software.  You can find more information at information at http://en.wikipedia.org/wiki/NAT_traversal. 
Technically, you can say this is not P2P anymore because it needs another server who has public IP.  :)


----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: "Zhou Ya Jin" <l4hurd at 163.com>; <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 7:59 PM
Subject: RE: [p2p-sip] NATs and P2P



Thanks Zhou!

So if most people do not have Public addresses, I guess the whole p2p is
not p2p anymore? 



-----Original Message-----
From: Zhou Ya Jin [mailto:l4hurd at 163.com] 
Sent: Tuesday, March 21, 2006 1:44 PM
To: Effi Shiri; p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] NATs and P2P

Actually in China, many internet user does not have public IP. As far as
I am concerned, people in my office share a public IP. We use a router
for NAT.
----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 4:22 PM
Subject: Re: [p2p-sip] NATs and P2P


> 
> 
> This is very interesting
> Do you have statistics on how many Home users actually have Public IP?
> 
> 
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Philip Matthews
> Sent: Tuesday, March 21, 2006 2:36 AM
> To: P2P-SIP
> Subject: [p2p-sip] NATs and P2P
> 
> [I guess all this traffic on the mailing list will teach me not to go

> on vacation
> just before an IETF meeting!]
> 
> As I catch up on all the messages about NATs over the past two weeks,
> it seems to me that many people are thinking only of the "Public P2P  
> Service Provider"
> use-case as described in the use-cases document. In other words, a
Skype
> competitor.
> 
> However, the use-cases document that David Bryan and his co-authors  
> wrote
> identified a number of other use-cases and it seems to me that these  
> have somewhat
> different NAT traversal requirements.
> 
> In particular, in some of these other use-cases, it seems to me that  
> we CANNOT assume
> there are peers with public IP addresses.
> 
> For example, consider the "Presence using Multimedia Consumer  
> Electronics Devices"
> use-case (section 3.1.3) -- essentially a P2P network of multimedia  
> consumer electronics devices
> that need presence information. Who is going to pay the extra money  
> to give their digital camera (or those
> neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4

> address?? On the contrary,
> devices like this are almost certainly going to have private IP  
> addresses -- it is very common today for
> wireless internet providers to place a big NAT in front of their  
> entire network and give private addresses
> to all their customers.
> 
> Or consider the "IP PBX" use-case -- a IP PBX system for a company  
> with a number of small branches
> scattered throughout the world. Each branch is going to have a NAT in

> front of its network, and all
> the phones in that branch are going to have private IP addresses.  
> None of the phones are going to
> have public IP addresses.
> 
> It is handling the NAT traversal issues for use-cases like these that

> Eric Cooper and I wrote our
> internet-draft on NAT Traversal for P2P:
>       http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
> and-overlays-00.txt
> 
> - Philip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> ______________________________________________________________________
> DSP Group LTD. automatically scans all emails and attachments using
> MessageLabs Email Security System.
> ______________________________________________________________________
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> 
> __________ NOD32 1.1452 (20060320) Information __________
> 
> This message was checked by NOD32 antivirus system.
> http://www.eset.com
> 
>

______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________


__________ NOD32 1.1452 (20060320) Information __________

This message was checked by NOD32 antivirus system.
http://www.eset.com



From effis at dsp.co.il  Tue Mar 21 07:22:50 2006
From: effis at dsp.co.il (Effi Shiri)
Date: Tue, 21 Mar 2006 14:22:50 +0200
Subject: [p2p-sip] NATs and P2P
Message-ID: <216A12DDD3490049BA3E56BAC324F285012FB01A@ILEXCH2003.il-prod.dspcorp.com>

One more thing, what is the most common type of NAT? 
(Symmetric, Full Coned, Address Restricted or Port Restricted)

Thanks

-----Original Message-----
From: Zhou Ya Jin [mailto:l4hurd at 163.com] 
Sent: Tuesday, March 21, 2006 2:16 PM
To: Effi Shiri
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] NATs and P2P

As far as I know, in China, we use  NAT_traversal to connect two parts
of P2P software.  You can find more information at information at
http://en.wikipedia.org/wiki/NAT_traversal. 
Technically, you can say this is not P2P anymore because it needs
another server who has public IP.  :)


----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: "Zhou Ya Jin" <l4hurd at 163.com>; <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 7:59 PM
Subject: RE: [p2p-sip] NATs and P2P



Thanks Zhou!

So if most people do not have Public addresses, I guess the whole p2p is
not p2p anymore? 



-----Original Message-----
From: Zhou Ya Jin [mailto:l4hurd at 163.com] 
Sent: Tuesday, March 21, 2006 1:44 PM
To: Effi Shiri; p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] NATs and P2P

Actually in China, many internet user does not have public IP. As far as
I am concerned, people in my office share a public IP. We use a router
for NAT.
----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 4:22 PM
Subject: Re: [p2p-sip] NATs and P2P


> 
> 
> This is very interesting
> Do you have statistics on how many Home users actually have Public IP?
> 
> 
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Philip Matthews
> Sent: Tuesday, March 21, 2006 2:36 AM
> To: P2P-SIP
> Subject: [p2p-sip] NATs and P2P
> 
> [I guess all this traffic on the mailing list will teach me not to go

> on vacation
> just before an IETF meeting!]
> 
> As I catch up on all the messages about NATs over the past two weeks,
> it seems to me that many people are thinking only of the "Public P2P  
> Service Provider"
> use-case as described in the use-cases document. In other words, a
Skype
> competitor.
> 
> However, the use-cases document that David Bryan and his co-authors  
> wrote
> identified a number of other use-cases and it seems to me that these  
> have somewhat
> different NAT traversal requirements.
> 
> In particular, in some of these other use-cases, it seems to me that  
> we CANNOT assume
> there are peers with public IP addresses.
> 
> For example, consider the "Presence using Multimedia Consumer  
> Electronics Devices"
> use-case (section 3.1.3) -- essentially a P2P network of multimedia  
> consumer electronics devices
> that need presence information. Who is going to pay the extra money  
> to give their digital camera (or those
> neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4

> address?? On the contrary,
> devices like this are almost certainly going to have private IP  
> addresses -- it is very common today for
> wireless internet providers to place a big NAT in front of their  
> entire network and give private addresses
> to all their customers.
> 
> Or consider the "IP PBX" use-case -- a IP PBX system for a company  
> with a number of small branches
> scattered throughout the world. Each branch is going to have a NAT in

> front of its network, and all
> the phones in that branch are going to have private IP addresses.  
> None of the phones are going to
> have public IP addresses.
> 
> It is handling the NAT traversal issues for use-cases like these that

> Eric Cooper and I wrote our
> internet-draft on NAT Traversal for P2P:
>       http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
> and-overlays-00.txt
> 
> - Philip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> ______________________________________________________________________
> DSP Group LTD. automatically scans all emails and attachments using
> MessageLabs Email Security System.
> ______________________________________________________________________
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> 
> __________ NOD32 1.1452 (20060320) Information __________
> 
> This message was checked by NOD32 antivirus system.
> http://www.eset.com
> 
>

______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________


__________ NOD32 1.1452 (20060320) Information __________

This message was checked by NOD32 antivirus system.
http://www.eset.com


______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________


From l4hurd at 163.com  Tue Mar 21 07:43:07 2006
From: l4hurd at 163.com (Zhou Ya Jin)
Date: Tue, 21 Mar 2006 20:43:07 +0800
Subject: [p2p-sip] NATs and P2P
References: <216A12DDD3490049BA3E56BAC324F285012FB01A@ILEXCH2003.il-prod.dspcorp.com>
Message-ID: <002101c64ce5$01ea6c80$7b01a8c0@killbill>

Most of NAT in china is Full Coned.  That is based on my knowledeg, maybe not correct.
----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: "Zhou Ya Jin" <l4hurd at 163.com>
Cc: <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 8:22 PM
Subject: RE: [p2p-sip] NATs and P2P



One more thing, what is the most common type of NAT? 
(Symmetric, Full Coned, Address Restricted or Port Restricted)

Thanks

-----Original Message-----
From: Zhou Ya Jin [mailto:l4hurd at 163.com] 
Sent: Tuesday, March 21, 2006 2:16 PM
To: Effi Shiri
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] NATs and P2P

As far as I know, in China, we use  NAT_traversal to connect two parts
of P2P software.  You can find more information at information at
http://en.wikipedia.org/wiki/NAT_traversal. 
Technically, you can say this is not P2P anymore because it needs
another server who has public IP.  :)


----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: "Zhou Ya Jin" <l4hurd at 163.com>; <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 7:59 PM
Subject: RE: [p2p-sip] NATs and P2P



Thanks Zhou!

So if most people do not have Public addresses, I guess the whole p2p is
not p2p anymore? 



-----Original Message-----
From: Zhou Ya Jin [mailto:l4hurd at 163.com] 
Sent: Tuesday, March 21, 2006 1:44 PM
To: Effi Shiri; p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] NATs and P2P

Actually in China, many internet user does not have public IP. As far as
I am concerned, people in my office share a public IP. We use a router
for NAT.
----- Original Message ----- 
From: "Effi Shiri" <effis at dsp.co.il>
To: <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 4:22 PM
Subject: Re: [p2p-sip] NATs and P2P


> 
> 
> This is very interesting
> Do you have statistics on how many Home users actually have Public IP?
> 
> 
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Philip Matthews
> Sent: Tuesday, March 21, 2006 2:36 AM
> To: P2P-SIP
> Subject: [p2p-sip] NATs and P2P
> 
> [I guess all this traffic on the mailing list will teach me not to go

> on vacation
> just before an IETF meeting!]
> 
> As I catch up on all the messages about NATs over the past two weeks,
> it seems to me that many people are thinking only of the "Public P2P  
> Service Provider"
> use-case as described in the use-cases document. In other words, a
Skype
> competitor.
> 
> However, the use-cases document that David Bryan and his co-authors  
> wrote
> identified a number of other use-cases and it seems to me that these  
> have somewhat
> different NAT traversal requirements.
> 
> In particular, in some of these other use-cases, it seems to me that  
> we CANNOT assume
> there are peers with public IP addresses.
> 
> For example, consider the "Presence using Multimedia Consumer  
> Electronics Devices"
> use-case (section 3.1.3) -- essentially a P2P network of multimedia  
> consumer electronics devices
> that need presence information. Who is going to pay the extra money  
> to give their digital camera (or those
> neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4

> address?? On the contrary,
> devices like this are almost certainly going to have private IP  
> addresses -- it is very common today for
> wireless internet providers to place a big NAT in front of their  
> entire network and give private addresses
> to all their customers.
> 
> Or consider the "IP PBX" use-case -- a IP PBX system for a company  
> with a number of small branches
> scattered throughout the world. Each branch is going to have a NAT in

> front of its network, and all
> the phones in that branch are going to have private IP addresses.  
> None of the phones are going to
> have public IP addresses.
> 
> It is handling the NAT traversal issues for use-cases like these that

> Eric Cooper and I wrote our
> internet-draft on NAT Traversal for P2P:
>       http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats- 
> and-overlays-00.txt
> 
> - Philip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> ______________________________________________________________________
> DSP Group LTD. automatically scans all emails and attachments using
> MessageLabs Email Security System.
> ______________________________________________________________________
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> 
> __________ NOD32 1.1452 (20060320) Information __________
> 
> This message was checked by NOD32 antivirus system.
> http://www.eset.com
> 
>

______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________


__________ NOD32 1.1452 (20060320) Information __________

This message was checked by NOD32 antivirus system.
http://www.eset.com


______________________________________________________________________
DSP Group LTD. automatically scans all emails and attachments using
MessageLabs Email Security System.
______________________________________________________________________


__________ NOD32 1.1452 (20060320) Information __________

This message was checked by NOD32 antivirus system.
http://www.eset.com



From kumarb at research.panasonic.com  Tue Mar 21 10:17:42 2006
From: kumarb at research.panasonic.com (Brijesh Kumar)
Date: Tue, 21 Mar 2006 10:17:42 -0500
Subject: [p2p-sip] NATs and P2P
References: <9151EA5D-BA55-4A69-8F6A-7118492C4283@magma.ca>
	<e9132f820603201758n727d6d25p62f4c4d60dac42d9@mail.gmail.com>
Message-ID: <017d01c64cfa$9f978480$01000001@Brijesh2K>

I think Bruce's approach makes sense.

The first task should be to come out with a  base specification
that works. If nothing else, this can be deployed in a hypothetical
pure IPv6 world, or  in a  limited enterprise universe where
traversing NAT may not be an issue.

The second task should be to modify the spec  that works with
NAT.  Trying to resolve NAT issue from day one is good idea, but
one can equally treat is as an extension of the core design.  It is just
a matter of convenience rather than  any hard logic.

Cheers,

--bk

----- Original Message ----- 
From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
To: "Philip Matthews" <philip_matthews at magma.ca>
Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
Sent: Monday, March 20, 2006 8:58 PM
Subject: Re: [p2p-sip] NATs and P2P


> Philip,
>
> I absolutely agree that we need to have a solution for this type of
> environment.   IMHO, some of the worries about NATs causing problems
> have been somewhat overstated.  As you point out, the majority of NATs
> in the wild are address independent mapping, and it is relatively
> trivial to punch holes to establish communications.  But you're also
> right that, especially for corporate networks, this doesn't solve the
> entire problem.
>
> I spent some time thinking about how to solve the specific problems
> you pose in terms of the protocol we have been playing with in
> draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
> there (not because we don't think it's a challenge, but just because
> we want to work on a basic protocol first).  If I was to address some
> of the problems you pose using that framework (modified chord), then
> one approach would be to remove the requirement that DHT operations be
> iterative and allow them to be recursive.  If DHT operations can be
> recursive, then the number of connections that a node must maintain
> through the firewall is limited by the size of the finger table.  That
> number doesn't have to be very big to have a rather efficient DHT.
> (never more than 128, and I think in practice the number can be much
> smaller.  If your guess of 160 is correct, we're more than OK here.)
>
> I'm a bit unhappy with that because I really like the iterative
> approach for a variety of reasons, but certainly the protocol could be
> modified that way and a hybrid environment could be established where
> some nodes are open (or easy to request opens) and support iterative
> queries with anyone, but others are recursive queries only.  One
> problem is that the burden of having a finger-table entry to a
> recursive-only host is placed on the non-recursive node that would
> otherwise redirect to its finger-table entry, but now must recurse the
> request.
>
> Right now the scheme would also have problems because, as Eric rightly
> points out in a separate thread on this mailing list, we don't have a
> good way of selecting Identifiers for nodes behind NATs (in
> particular).  But if you have node certificates or another way of
> authenticating your nodes, this is less of a problem.
>
> I want to make it clear that I'm not ready to propose a specific
> solution, but I think there are ways to support structured DHT
> algorithms that are not really very painful and work reasonably well
> inside these environments.  Clearly support for such an environment
> needs to be a listed in eventual requirements docs.
>
> Bruce
>
>
> On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > [I guess all this traffic on the mailing list will teach me not to go
> > on vacation
> > just before an IETF meeting!]
> >
> > As I catch up on all the messages about NATs over the past two weeks,
> > it seems to me that many people are thinking only of the "Public P2P
> > Service Provider"
> > use-case as described in the use-cases document. In other words, a Skype
> > competitor.
> >
> > However, the use-cases document that David Bryan and his co-authors
> > wrote
> > identified a number of other use-cases and it seems to me that these
> > have somewhat
> > different NAT traversal requirements.
> >
> > In particular, in some of these other use-cases, it seems to me that
> > we CANNOT assume
> > there are peers with public IP addresses.
> >
> > For example, consider the "Presence using Multimedia Consumer
> > Electronics Devices"
> > use-case (section 3.1.3) -- essentially a P2P network of multimedia
> > consumer electronics devices
> > that need presence information. Who is going to pay the extra money
> > to give their digital camera (or those
> > neat 770 tablets that Nokia is demoing here in Dallas?) a public IPv4
> > address?? On the contrary,
> > devices like this are almost certainly going to have private IP
> > addresses -- it is very common today for
> > wireless internet providers to place a big NAT in front of their
> > entire network and give private addresses
> > to all their customers.
> >
> > Or consider the "IP PBX" use-case -- a IP PBX system for a company
> > with a number of small branches
> > scattered throughout the world. Each branch is going to have a NAT in
> > front of its network, and all
> > the phones in that branch are going to have private IP addresses.
> > None of the phones are going to
> > have public IP addresses.
> >
> > It is handling the NAT traversal issues for use-cases like these that
> > Eric Cooper and I wrote our
> > internet-draft on NAT Traversal for P2P:
> >        http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats-
> > and-overlays-00.txt
> >
> > - Philip
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From slavitch at objectworld.com  Tue Mar 21 10:14:58 2006
From: slavitch at objectworld.com (Michael Slavitch)
Date: Tue, 21 Mar 2006 10:14:58 -0500
Subject: [p2p-sip] NATs and P2P
Message-ID: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>

This I believe is a disastrous approach.  If it's not resolved day one
it will not happen.

SIP put this off ten years ago, because NAT was thought bad, something
that would go away. It didn't happen.

I'm sorry, but if it doesn't address NAT we might as well go home.  The
whole point of the overlay is to get around structures such as NAT,
different addressing (IPV4/IPV6), walled gardens and the like.

In a pure IPV6 world we already have a P2P SIP protocol.  It is called
SIP.

M 


Michael Slavitch           Lead, Strategic Initiatives 
slavitch at objectworld.com   Objectworld Communications Corp. 
+1.613.599.9698 .225       Ottawa, Ontario, Canada

-----Original Message-----
From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Brijesh Kumar
Sent: March 21, 2006 10:18 AM
To: Bruce Lowekamp; Philip Matthews
Cc: P2P-SIP
Subject: Re: [p2p-sip] NATs and P2P

I think Bruce's approach makes sense.

The first task should be to come out with a  base specification that
works. If nothing else, this can be deployed in a hypothetical pure IPv6
world, or  in a  limited enterprise universe where traversing NAT may
not be an issue.

The second task should be to modify the spec  that works with NAT.
Trying to resolve NAT issue from day one is good idea, but one can
equally treat is as an extension of the core design.  It is just a
matter of convenience rather than  any hard logic.

Cheers,

--bk

----- Original Message -----
From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
To: "Philip Matthews" <philip_matthews at magma.ca>
Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
Sent: Monday, March 20, 2006 8:58 PM
Subject: Re: [p2p-sip] NATs and P2P


> Philip,
>
> I absolutely agree that we need to have a solution for this type of
> environment.   IMHO, some of the worries about NATs causing problems
> have been somewhat overstated.  As you point out, the majority of NATs
> in the wild are address independent mapping, and it is relatively
> trivial to punch holes to establish communications.  But you're also
> right that, especially for corporate networks, this doesn't solve the
> entire problem.
>
> I spent some time thinking about how to solve the specific problems
> you pose in terms of the protocol we have been playing with in
> draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
> there (not because we don't think it's a challenge, but just because
> we want to work on a basic protocol first).  If I was to address some
> of the problems you pose using that framework (modified chord), then
> one approach would be to remove the requirement that DHT operations be
> iterative and allow them to be recursive.  If DHT operations can be
> recursive, then the number of connections that a node must maintain
> through the firewall is limited by the size of the finger table.  That
> number doesn't have to be very big to have a rather efficient DHT.
> (never more than 128, and I think in practice the number can be much
> smaller.  If your guess of 160 is correct, we're more than OK here.)
>
> I'm a bit unhappy with that because I really like the iterative
> approach for a variety of reasons, but certainly the protocol could be
> modified that way and a hybrid environment could be established where
> some nodes are open (or easy to request opens) and support iterative
> queries with anyone, but others are recursive queries only.  One
> problem is that the burden of having a finger-table entry to a
> recursive-only host is placed on the non-recursive node that would
> otherwise redirect to its finger-table entry, but now must recurse the
> request.
>
> Right now the scheme would also have problems because, as Eric rightly
> points out in a separate thread on this mailing list, we don't have a
> good way of selecting Identifiers for nodes behind NATs (in
> particular).  But if you have node certificates or another way of
> authenticating your nodes, this is less of a problem.
>
> I want to make it clear that I'm not ready to propose a specific
> solution, but I think there are ways to support structured DHT
> algorithms that are not really very painful and work reasonably well
> inside these environments.  Clearly support for such an environment
> needs to be a listed in eventual requirements docs.
>
> Bruce
>
>
> On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > [I guess all this traffic on the mailing list will teach me not to
go
> > on vacation
> > just before an IETF meeting!]
> >
> > As I catch up on all the messages about NATs over the past two
weeks,
> > it seems to me that many people are thinking only of the "Public P2P
> > Service Provider"
> > use-case as described in the use-cases document. In other words, a
Skype
> > competitor.
> >
> > However, the use-cases document that David Bryan and his co-authors
> > wrote
> > identified a number of other use-cases and it seems to me that these
> > have somewhat
> > different NAT traversal requirements.
> >
> > In particular, in some of these other use-cases, it seems to me that
> > we CANNOT assume
> > there are peers with public IP addresses.
> >
> > For example, consider the "Presence using Multimedia Consumer
> > Electronics Devices"
> > use-case (section 3.1.3) -- essentially a P2P network of multimedia
> > consumer electronics devices
> > that need presence information. Who is going to pay the extra money
> > to give their digital camera (or those
> > neat 770 tablets that Nokia is demoing here in Dallas?) a public
IPv4
> > address?? On the contrary,
> > devices like this are almost certainly going to have private IP
> > addresses -- it is very common today for
> > wireless internet providers to place a big NAT in front of their
> > entire network and give private addresses
> > to all their customers.
> >
> > Or consider the "IP PBX" use-case -- a IP PBX system for a company
> > with a number of small branches
> > scattered throughout the world. Each branch is going to have a NAT
in
> > front of its network, and all
> > the phones in that branch are going to have private IP addresses.
> > None of the phones are going to
> > have public IP addresses.
> >
> > It is handling the NAT traversal issues for use-cases like these
that
> > Eric Cooper and I wrote our
> > internet-draft on NAT Traversal for P2P:
> >
http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats-
> > and-overlays-00.txt
> >
> > - Philip
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


-- 


-----------------------------------------------

This transmission (including any attachments) may contain confidential information, privileged material (including material protected by the solicitor-client or other applicable privileges), or constitute non-public information. Any use of this information by anyone other than the intended recipient is prohibited. If you have received this transmission in error, please immediately reply to the sender and delete this information from your system. Use, dissemination, distribution, or reproduction of this transmission by unintended recipients is not authorized and may be unlawful. 

Any views, opinions and content presented in this e-mail are solely those of the author and may not represent those of Objectworld Communications Corp. 


From eunsoo at research.panasonic.com  Tue Mar 21 10:28:32 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Tue, 21 Mar 2006 10:28:32 -0500
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
Message-ID: <44201BA0.30407@research.panasonic.com>

As the proposed charter says, I believe that we have the consensus that
NAT traversal should be addressed from day one.
That's why we have talked so much about ICE and so on.
Thanks.

Eunsoo

Michael Slavitch wrote:

>This I believe is a disastrous approach.  If it's not resolved day one
>it will not happen.
>
>SIP put this off ten years ago, because NAT was thought bad, something
>that would go away. It didn't happen.
>
>I'm sorry, but if it doesn't address NAT we might as well go home.  The
>whole point of the overlay is to get around structures such as NAT,
>different addressing (IPV4/IPV6), walled gardens and the like.
>
>In a pure IPV6 world we already have a P2P SIP protocol.  It is called
>SIP.
>
>M 
>
>
>Michael Slavitch           Lead, Strategic Initiatives 
>slavitch at objectworld.com   Objectworld Communications Corp. 
>+1.613.599.9698 .225       Ottawa, Ontario, Canada
>
>-----Original Message-----
>From: p2p-sip-bounces at cs.columbia.edu
>[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Brijesh Kumar
>Sent: March 21, 2006 10:18 AM
>To: Bruce Lowekamp; Philip Matthews
>Cc: P2P-SIP
>Subject: Re: [p2p-sip] NATs and P2P
>
>I think Bruce's approach makes sense.
>
>The first task should be to come out with a  base specification that
>works. If nothing else, this can be deployed in a hypothetical pure IPv6
>world, or  in a  limited enterprise universe where traversing NAT may
>not be an issue.
>
>The second task should be to modify the spec  that works with NAT.
>Trying to resolve NAT issue from day one is good idea, but one can
>equally treat is as an extension of the core design.  It is just a
>matter of convenience rather than  any hard logic.
>
>Cheers,
>
>--bk
>
>----- Original Message -----
>From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
>To: "Philip Matthews" <philip_matthews at magma.ca>
>Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
>Sent: Monday, March 20, 2006 8:58 PM
>Subject: Re: [p2p-sip] NATs and P2P
>
>
>  
>
>>Philip,
>>
>>I absolutely agree that we need to have a solution for this type of
>>environment.   IMHO, some of the worries about NATs causing problems
>>have been somewhat overstated.  As you point out, the majority of NATs
>>in the wild are address independent mapping, and it is relatively
>>trivial to punch holes to establish communications.  But you're also
>>right that, especially for corporate networks, this doesn't solve the
>>entire problem.
>>
>>I spent some time thinking about how to solve the specific problems
>>you pose in terms of the protocol we have been playing with in
>>draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
>>there (not because we don't think it's a challenge, but just because
>>we want to work on a basic protocol first).  If I was to address some
>>of the problems you pose using that framework (modified chord), then
>>one approach would be to remove the requirement that DHT operations be
>>iterative and allow them to be recursive.  If DHT operations can be
>>recursive, then the number of connections that a node must maintain
>>through the firewall is limited by the size of the finger table.  That
>>number doesn't have to be very big to have a rather efficient DHT.
>>(never more than 128, and I think in practice the number can be much
>>smaller.  If your guess of 160 is correct, we're more than OK here.)
>>
>>I'm a bit unhappy with that because I really like the iterative
>>approach for a variety of reasons, but certainly the protocol could be
>>modified that way and a hybrid environment could be established where
>>some nodes are open (or easy to request opens) and support iterative
>>queries with anyone, but others are recursive queries only.  One
>>problem is that the burden of having a finger-table entry to a
>>recursive-only host is placed on the non-recursive node that would
>>otherwise redirect to its finger-table entry, but now must recurse the
>>request.
>>
>>Right now the scheme would also have problems because, as Eric rightly
>>points out in a separate thread on this mailing list, we don't have a
>>good way of selecting Identifiers for nodes behind NATs (in
>>particular).  But if you have node certificates or another way of
>>authenticating your nodes, this is less of a problem.
>>
>>I want to make it clear that I'm not ready to propose a specific
>>solution, but I think there are ways to support structured DHT
>>algorithms that are not really very painful and work reasonably well
>>inside these environments.  Clearly support for such an environment
>>needs to be a listed in eventual requirements docs.
>>
>>Bruce
>>
>>
>>On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
>>    
>>
>>>[I guess all this traffic on the mailing list will teach me not to
>>>      
>>>
>go
>  
>
>>>on vacation
>>>just before an IETF meeting!]
>>>
>>>As I catch up on all the messages about NATs over the past two
>>>      
>>>
>weeks,
>  
>
>>>it seems to me that many people are thinking only of the "Public P2P
>>>Service Provider"
>>>use-case as described in the use-cases document. In other words, a
>>>      
>>>
>Skype
>  
>
>>>competitor.
>>>
>>>However, the use-cases document that David Bryan and his co-authors
>>>wrote
>>>identified a number of other use-cases and it seems to me that these
>>>have somewhat
>>>different NAT traversal requirements.
>>>
>>>In particular, in some of these other use-cases, it seems to me that
>>>we CANNOT assume
>>>there are peers with public IP addresses.
>>>
>>>For example, consider the "Presence using Multimedia Consumer
>>>Electronics Devices"
>>>use-case (section 3.1.3) -- essentially a P2P network of multimedia
>>>consumer electronics devices
>>>that need presence information. Who is going to pay the extra money
>>>to give their digital camera (or those
>>>neat 770 tablets that Nokia is demoing here in Dallas?) a public
>>>      
>>>
>IPv4
>  
>
>>>address?? On the contrary,
>>>devices like this are almost certainly going to have private IP
>>>addresses -- it is very common today for
>>>wireless internet providers to place a big NAT in front of their
>>>entire network and give private addresses
>>>to all their customers.
>>>
>>>Or consider the "IP PBX" use-case -- a IP PBX system for a company
>>>with a number of small branches
>>>scattered throughout the world. Each branch is going to have a NAT
>>>      
>>>
>in
>  
>
>>>front of its network, and all
>>>the phones in that branch are going to have private IP addresses.
>>>None of the phones are going to
>>>have public IP addresses.
>>>
>>>It is handling the NAT traversal issues for use-cases like these
>>>      
>>>
>that
>  
>
>>>Eric Cooper and I wrote our
>>>internet-draft on NAT Traversal for P2P:
>>>
>>>      
>>>
>http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats-
>  
>
>>>and-overlays-00.txt
>>>
>>>- Philip
>>>
>>>_______________________________________________
>>>p2p-sip mailing list
>>>p2p-sip at cs.columbia.edu
>>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>
>>>
>>>      
>>>
>>_______________________________________________
>>p2p-sip mailing list
>>p2p-sip at cs.columbia.edu
>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
>>    
>>
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>
>  
>



From philip_matthews at magma.ca  Tue Mar 21 10:31:34 2006
From: philip_matthews at magma.ca (Philip Matthews)
Date: Tue, 21 Mar 2006 09:31:34 -0600
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
Message-ID: <48F47E6D-B65A-43B6-83AF-5557533CB38D@magma.ca>

I must agree with Michael on this one.

At Avaya, we have been actively investigating various P2P-SIP  
approaches.
We started with the approach suggested by Brijesh (ignore NATs at  
first).
However, when we started considering NATs, we were amazed at how much
this influenced the overall architecture. We ended up re-thinking  
quite a few
things.

- Philip


On 21-Mar-06, at 9:14 , Michael Slavitch wrote:

> This I believe is a disastrous approach.  If it's not resolved day one
> it will not happen.
>
> SIP put this off ten years ago, because NAT was thought bad, something
> that would go away. It didn't happen.
>
> I'm sorry, but if it doesn't address NAT we might as well go home.   
> The
> whole point of the overlay is to get around structures such as NAT,
> different addressing (IPV4/IPV6), walled gardens and the like.
>
> In a pure IPV6 world we already have a P2P SIP protocol.  It is called
> SIP.
>
> M
>
>
> Michael Slavitch           Lead, Strategic Initiatives
> slavitch at objectworld.com   Objectworld Communications Corp.
> +1.613.599.9698 .225       Ottawa, Ontario, Canada
>
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Brijesh Kumar
> Sent: March 21, 2006 10:18 AM
> To: Bruce Lowekamp; Philip Matthews
> Cc: P2P-SIP
> Subject: Re: [p2p-sip] NATs and P2P
>
> I think Bruce's approach makes sense.
>
> The first task should be to come out with a  base specification that
> works. If nothing else, this can be deployed in a hypothetical pure  
> IPv6
> world, or  in a  limited enterprise universe where traversing NAT may
> not be an issue.
>
> The second task should be to modify the spec  that works with NAT.
> Trying to resolve NAT issue from day one is good idea, but one can
> equally treat is as an extension of the core design.  It is just a
> matter of convenience rather than  any hard logic.
>
> Cheers,
>
> --bk
>
> ----- Original Message -----
> From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
> To: "Philip Matthews" <philip_matthews at magma.ca>
> Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
> Sent: Monday, March 20, 2006 8:58 PM
> Subject: Re: [p2p-sip] NATs and P2P
>
>
>> Philip,
>>
>> I absolutely agree that we need to have a solution for this type of
>> environment.   IMHO, some of the worries about NATs causing problems
>> have been somewhat overstated.  As you point out, the majority of  
>> NATs
>> in the wild are address independent mapping, and it is relatively
>> trivial to punch holes to establish communications.  But you're also
>> right that, especially for corporate networks, this doesn't solve the
>> entire problem.
>>
>> I spent some time thinking about how to solve the specific problems
>> you pose in terms of the protocol we have been playing with in
>> draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
>> there (not because we don't think it's a challenge, but just because
>> we want to work on a basic protocol first).  If I was to address some
>> of the problems you pose using that framework (modified chord), then
>> one approach would be to remove the requirement that DHT  
>> operations be
>> iterative and allow them to be recursive.  If DHT operations can be
>> recursive, then the number of connections that a node must maintain
>> through the firewall is limited by the size of the finger table.   
>> That
>> number doesn't have to be very big to have a rather efficient DHT.
>> (never more than 128, and I think in practice the number can be much
>> smaller.  If your guess of 160 is correct, we're more than OK here.)
>>
>> I'm a bit unhappy with that because I really like the iterative
>> approach for a variety of reasons, but certainly the protocol  
>> could be
>> modified that way and a hybrid environment could be established where
>> some nodes are open (or easy to request opens) and support iterative
>> queries with anyone, but others are recursive queries only.  One
>> problem is that the burden of having a finger-table entry to a
>> recursive-only host is placed on the non-recursive node that would
>> otherwise redirect to its finger-table entry, but now must recurse  
>> the
>> request.
>>
>> Right now the scheme would also have problems because, as Eric  
>> rightly
>> points out in a separate thread on this mailing list, we don't have a
>> good way of selecting Identifiers for nodes behind NATs (in
>> particular).  But if you have node certificates or another way of
>> authenticating your nodes, this is less of a problem.
>>
>> I want to make it clear that I'm not ready to propose a specific
>> solution, but I think there are ways to support structured DHT
>> algorithms that are not really very painful and work reasonably well
>> inside these environments.  Clearly support for such an environment
>> needs to be a listed in eventual requirements docs.
>>
>> Bruce
>>
>>
>> On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
>>> [I guess all this traffic on the mailing list will teach me not to
> go
>>> on vacation
>>> just before an IETF meeting!]
>>>
>>> As I catch up on all the messages about NATs over the past two
> weeks,
>>> it seems to me that many people are thinking only of the "Public P2P
>>> Service Provider"
>>> use-case as described in the use-cases document. In other words, a
> Skype
>>> competitor.
>>>
>>> However, the use-cases document that David Bryan and his co-authors
>>> wrote
>>> identified a number of other use-cases and it seems to me that these
>>> have somewhat
>>> different NAT traversal requirements.
>>>
>>> In particular, in some of these other use-cases, it seems to me that
>>> we CANNOT assume
>>> there are peers with public IP addresses.
>>>
>>> For example, consider the "Presence using Multimedia Consumer
>>> Electronics Devices"
>>> use-case (section 3.1.3) -- essentially a P2P network of multimedia
>>> consumer electronics devices
>>> that need presence information. Who is going to pay the extra money
>>> to give their digital camera (or those
>>> neat 770 tablets that Nokia is demoing here in Dallas?) a public
> IPv4
>>> address?? On the contrary,
>>> devices like this are almost certainly going to have private IP
>>> addresses -- it is very common today for
>>> wireless internet providers to place a big NAT in front of their
>>> entire network and give private addresses
>>> to all their customers.
>>>
>>> Or consider the "IP PBX" use-case -- a IP PBX system for a company
>>> with a number of small branches
>>> scattered throughout the world. Each branch is going to have a NAT
> in
>>> front of its network, and all
>>> the phones in that branch are going to have private IP addresses.
>>> None of the phones are going to
>>> have public IP addresses.
>>>
>>> It is handling the NAT traversal issues for use-cases like these
> that
>>> Eric Cooper and I wrote our
>>> internet-draft on NAT Traversal for P2P:
>>>
> http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats-
>>> and-overlays-00.txt
>>>
>>> - Philip
>>>
>>> _______________________________________________
>>> p2p-sip mailing list
>>> p2p-sip at cs.columbia.edu
>>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>
>>>
>>
>> _______________________________________________
>> p2p-sip mailing list
>> p2p-sip at cs.columbia.edu
>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>
> -- 
>
>
> -----------------------------------------------
>
> This transmission (including any attachments) may contain  
> confidential information, privileged material (including material  
> protected by the solicitor-client or other applicable privileges),  
> or constitute non-public information. Any use of this information  
> by anyone other than the intended recipient is prohibited. If you  
> have received this transmission in error, please immediately reply  
> to the sender and delete this information from your system. Use,  
> dissemination, distribution, or reproduction of this transmission  
> by unintended recipients is not authorized and may be unlawful.
>
> Any views, opinions and content presented in this e-mail are solely  
> those of the author and may not represent those of Objectworld  
> Communications Corp.
>


From sathya at research.panasonic.com  Tue Mar 21 10:40:24 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Tue, 21 Mar 2006 10:40:24 -0500
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <2e794a83.4a832e79@pintlmail.MITL.Research.Panasonic.COM>




> I'd add to that this key question:
> 
> 	"Will we extend SIP, or create a new protocol?"

I agree that this is a "key" question. IMHO, this is the first thing that needs to be addressed by the WG  to create some (rough) consensus.

This was discussed in the ad-hoc in Vancouver and if my memory serves me, there was some sense in the room - but I guess we need to re-visit this issue first thing. Preferably during the friday ad-hoc so that we don't have this hanging over our head for another three months.

- Sathya
 
> I'm finding it hard to make any headway without understanding the 
> above.Basically, I see two major directions we could go:
> 
> 1) Extend SIP with overlay-maintenance and resource-location messages.
> 
> 2) Create a new overlay protocol and develop bindings for SIP and 
> ICE (eg,
> distributed proxy service and STUN/TURN resource-location service).
> 
> It seems this high-level decision keeps coming up again and again in
> discussions of the smaller issues.
> 
> -david
> 
> 
> > -----Original Message-----
> > From: p2p-sip-bounces at cs.columbia.edu [p2p-sip-
> > bounces at cs.columbia.edu] On Behalf Of Philip Matthews
> > Sent: Monday, March 20, 2006 5:26 PM
> > To: P2P-SIP
> > Subject: [p2p-sip] Some comments on Use-cases document
> > 
> > In advance of Wednesday's BOF session, I thought I should
> > send out a few comments on the use-cases document
> > http://www.ietf.org/internet-drafts/draft-bryan-sipping-p2p-
> > usecases-00.txt
> > 
> > First of all, I think that having such a document is a great idea.
> > We can use this document to pick the use-cases that we want to 
> support.> 
> > Some comments:
> > 
> > * I was initially confused when reading section 3. After reading a
> > few times,
> > it seems to me that some of the sections describe a network type
> > (e.g., "global Internet" or "Limited Internet Connectivity") while
> > others describe
> > a particular P2P use-case. I would be helpful if the document made
> > the distinction
> > between these two types of descriptions clearer.
> > 
> > * I think it would be helpful to describe each use-case in more
> > detail, and
> > make it clear how the use-cases differ from each other.
> > 
> > * Section 3.1.1 "Public P2P VoIP Service Providers" doesn't really
> > describe
> > this use-case -- right now, it is mostly a few comments on Skype
> > (which seem out-of-place). I might suggest that this use-case could
> > be characterized as follows:
> > -      Very large numbers of peers (millions)
> > -      Some may be behind NATs, while others may have public IP
> > addresses
> > -      Protection against malicious peers is very important
> > -      Connection to PSTN and CS SIP is important
> > 
> > * What is the distinction between 3.1.1 Public P2P VoIP Service
> > Providers"
> > and 3.1.2  "Open Global P2P VoIP Network"??  Is the only difference
> > the fact
> > that in 3.1.1, there is an organization who determines who is 
> allowed> to join the
> > overlay, while in 3.1.2, there is no such organization?
> > 
> > * I would like to add another use-case under 3.1 "Global Internet
> > Environment".
> > This use-case is a IP PBX for a company which has a number of
> > branches scattered
> > throughout the world. Each branch consists of a number of 
> phones, and
> > each branch
> > has a NAT or Firewall box connecting its private network to the
> > Internet.
> > The phones in all the branches connect together into a single 
> P2P PBX
> > system.
> > This use-case differs from 3.4.1 because in this use-case some 
> phones> must
> > connect to each other via the Internet.
> > 
> > * Finally, I find calling these "use-cases" a bit strange. They seem
> > much too high-level
> > and vague to be anything like what the term "use-case" means in UML.
> > I might suggest a term like "application scenarios" might be more
> > appropriate.
> > 
> > 
> > - Philip
> > 
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From kumarb at research.panasonic.com  Tue Mar 21 11:33:00 2006
From: kumarb at research.panasonic.com (Brijesh Kumar)
Date: Tue, 21 Mar 2006 11:33:00 -0500
Subject: [p2p-sip] NATs and P2P
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<44201BA0.30407@research.panasonic.com>
Message-ID: <022701c64d05$a5e8caf0$01000001@Brijesh2K>

Eunsoo:

Thanks for pointing to the past discussion, and use cases.

After re-reading the  use cases and some of the threads for
requirements of NAT to enable these use cases.   I  realize that
support for NAT is needed from  early  on, and hence  fully
support your comment that the integration of the NAT is needed
in the architecture from the beginning.  Count me in for this, please.

Cheers,

--brijesh

----- Original Message ----- 
From: "Eunsoo Shim" <eunsoo at research.panasonic.com>
To: "Michael Slavitch" <slavitch at objectworld.com>
Cc: "Brijesh Kumar" <kumarb at research.panasonic.com>; "Bruce Lowekamp"
<lowekamp at cs.wm.edu>; "Philip Matthews" <philip_matthews at magma.ca>;
"P2P-SIP" <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 10:28 AM
Subject: Re: [p2p-sip] NATs and P2P


> As the proposed charter says, I believe that we have the consensus that
> NAT traversal should be addressed from day one.
> That's why we have talked so much about ICE and so on.
> Thanks.
>
> Eunsoo
>
> Michael Slavitch wrote:
>
> >This I believe is a disastrous approach.  If it's not resolved day one
> >it will not happen.
> >
> >SIP put this off ten years ago, because NAT was thought bad, something
> >that would go away. It didn't happen.
> >
> >I'm sorry, but if it doesn't address NAT we might as well go home.  The
> >whole point of the overlay is to get around structures such as NAT,
> >different addressing (IPV4/IPV6), walled gardens and the like.
> >
> >In a pure IPV6 world we already have a P2P SIP protocol.  It is called
> >SIP.
> >
> >M
> >
> >
> >Michael Slavitch           Lead, Strategic Initiatives
> >slavitch at objectworld.com   Objectworld Communications Corp.
> >+1.613.599.9698 .225       Ottawa, Ontario, Canada
> >
> >-----Original Message-----
> >From: p2p-sip-bounces at cs.columbia.edu
> >[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Brijesh Kumar
> >Sent: March 21, 2006 10:18 AM
> >To: Bruce Lowekamp; Philip Matthews
> >Cc: P2P-SIP
> >Subject: Re: [p2p-sip] NATs and P2P
> >
> >I think Bruce's approach makes sense.
> >
> >The first task should be to come out with a  base specification that
> >works. If nothing else, this can be deployed in a hypothetical pure IPv6
> >world, or  in a  limited enterprise universe where traversing NAT may
> >not be an issue.
> >
> >The second task should be to modify the spec  that works with NAT.
> >Trying to resolve NAT issue from day one is good idea, but one can
> >equally treat is as an extension of the core design.  It is just a
> >matter of convenience rather than  any hard logic.
> >
> >Cheers,
> >
> >--bk
> >
> >----- Original Message -----
> >From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
> >To: "Philip Matthews" <philip_matthews at magma.ca>
> >Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
> >Sent: Monday, March 20, 2006 8:58 PM
> >Subject: Re: [p2p-sip] NATs and P2P
> >
> >
> >
> >
> >>Philip,
> >>
> >>I absolutely agree that we need to have a solution for this type of
> >>environment.   IMHO, some of the worries about NATs causing problems
> >>have been somewhat overstated.  As you point out, the majority of NATs
> >>in the wild are address independent mapping, and it is relatively
> >>trivial to punch holes to establish communications.  But you're also
> >>right that, especially for corporate networks, this doesn't solve the
> >>entire problem.
> >>
> >>I spent some time thinking about how to solve the specific problems
> >>you pose in terms of the protocol we have been playing with in
> >>draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
> >>there (not because we don't think it's a challenge, but just because
> >>we want to work on a basic protocol first).  If I was to address some
> >>of the problems you pose using that framework (modified chord), then
> >>one approach would be to remove the requirement that DHT operations be
> >>iterative and allow them to be recursive.  If DHT operations can be
> >>recursive, then the number of connections that a node must maintain
> >>through the firewall is limited by the size of the finger table.  That
> >>number doesn't have to be very big to have a rather efficient DHT.
> >>(never more than 128, and I think in practice the number can be much
> >>smaller.  If your guess of 160 is correct, we're more than OK here.)
> >>
> >>I'm a bit unhappy with that because I really like the iterative
> >>approach for a variety of reasons, but certainly the protocol could be
> >>modified that way and a hybrid environment could be established where
> >>some nodes are open (or easy to request opens) and support iterative
> >>queries with anyone, but others are recursive queries only.  One
> >>problem is that the burden of having a finger-table entry to a
> >>recursive-only host is placed on the non-recursive node that would
> >>otherwise redirect to its finger-table entry, but now must recurse the
> >>request.
> >>
> >>Right now the scheme would also have problems because, as Eric rightly
> >>points out in a separate thread on this mailing list, we don't have a
> >>good way of selecting Identifiers for nodes behind NATs (in
> >>particular).  But if you have node certificates or another way of
> >>authenticating your nodes, this is less of a problem.
> >>
> >>I want to make it clear that I'm not ready to propose a specific
> >>solution, but I think there are ways to support structured DHT
> >>algorithms that are not really very painful and work reasonably well
> >>inside these environments.  Clearly support for such an environment
> >>needs to be a listed in eventual requirements docs.
> >>
> >>Bruce
> >>
> >>
> >>On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> >>
> >>
> >>>[I guess all this traffic on the mailing list will teach me not to
> >>>
> >>>
> >go
> >
> >
> >>>on vacation
> >>>just before an IETF meeting!]
> >>>
> >>>As I catch up on all the messages about NATs over the past two
> >>>
> >>>
> >weeks,
> >
> >
> >>>it seems to me that many people are thinking only of the "Public P2P
> >>>Service Provider"
> >>>use-case as described in the use-cases document. In other words, a
> >>>
> >>>
> >Skype
> >
> >
> >>>competitor.
> >>>
> >>>However, the use-cases document that David Bryan and his co-authors
> >>>wrote
> >>>identified a number of other use-cases and it seems to me that these
> >>>have somewhat
> >>>different NAT traversal requirements.
> >>>
> >>>In particular, in some of these other use-cases, it seems to me that
> >>>we CANNOT assume
> >>>there are peers with public IP addresses.
> >>>
> >>>For example, consider the "Presence using Multimedia Consumer
> >>>Electronics Devices"
> >>>use-case (section 3.1.3) -- essentially a P2P network of multimedia
> >>>consumer electronics devices
> >>>that need presence information. Who is going to pay the extra money
> >>>to give their digital camera (or those
> >>>neat 770 tablets that Nokia is demoing here in Dallas?) a public
> >>>
> >>>
> >IPv4
> >
> >
> >>>address?? On the contrary,
> >>>devices like this are almost certainly going to have private IP
> >>>addresses -- it is very common today for
> >>>wireless internet providers to place a big NAT in front of their
> >>>entire network and give private addresses
> >>>to all their customers.
> >>>
> >>>Or consider the "IP PBX" use-case -- a IP PBX system for a company
> >>>with a number of small branches
> >>>scattered throughout the world. Each branch is going to have a NAT
> >>>
> >>>
> >in
> >
> >
> >>>front of its network, and all
> >>>the phones in that branch are going to have private IP addresses.
> >>>None of the phones are going to
> >>>have public IP addresses.
> >>>
> >>>It is handling the NAT traversal issues for use-cases like these
> >>>
> >>>
> >that
> >
> >
> >>>Eric Cooper and I wrote our
> >>>internet-draft on NAT Traversal for P2P:
> >>>
> >>>
> >>>
> >http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats-
> >
> >
> >>>and-overlays-00.txt
> >>>
> >>>- Philip
> >>>
> >>>_______________________________________________
> >>>p2p-sip mailing list
> >>>p2p-sip at cs.columbia.edu
> >>>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >>>
> >>>
> >>>
> >>>
> >>_______________________________________________
> >>p2p-sip mailing list
> >>p2p-sip at cs.columbia.edu
> >>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >>
> >>
> >>
> >
> >_______________________________________________
> >p2p-sip mailing list
> >p2p-sip at cs.columbia.edu
> >https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
> >
> >
>
>
>


From dean.willis at softarmor.com  Tue Mar 21 12:02:06 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 21 Mar 2006 11:02:06 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <200603210319.k2L3JIrj001729@cs.columbia.edu>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
Message-ID: <961760B0-982A-42EC-AA1F-97689977A095@softarmor.com>


On Mar 20, 2006, at 9:19 PM, David Barrett wrote:

> I'd add to that this key question:
>
> 	"Will we extend SIP, or create a new protocol?"
>
> I'm finding it hard to make any headway without understanding the  
> above.
> Basically, I see two major directions we could go:
>
> 1) Extend SIP with overlay-maintenance and resource-location messages.
>
> 2) Create a new overlay protocol and develop bindings for SIP and  
> ICE (eg,
> distributed proxy service and STUN/TURN resource-location service).
>
> It seems this high-level decision keeps coming up again and again in
> discussions of the smaller issues.
>

I propose that we first define an abstract model of an overlay  
network that includes the requisite primitives INSERT and QUERY  
(anything else needed?) and use this abstraction as our middle layer.

For the lower-layer, we do a binding from the abstraction to one or  
more distributed database models, probably starting with CHORD. This  
leaves our P2PSIP architecture independent of the database, but gives  
a proof-of-concept adequate for implementation.

For the upper-layer, we have one model defined using SIP REGISTER  
messages. We've also heard of work that uses REGISTER for the  
"INSERT" primitive and SUBSCRIBE/NOTIFY for the "query" primitive.

I'm also very interested in seeing if there's an obvious binding for  
TURN here, as that might kill two birds with one stone.

I propose we start with the REGISTER/REGISTER model from the bryan  
draft and talk that through a little further before we think about  
changing it.

--
Dean


From huang-ming.pan at comcast.net  Tue Mar 21 13:30:11 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Tue, 21 Mar 2006 10:30:11 -0800
Subject: [p2p-sip] NATs and P2P
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
Message-ID: <009901c64d15$7dd2b250$6400a8c0@comcast.net>

> I'm sorry, but if it doesn't address NAT we might as well go home.

I believe any incremental p2p-sip approach a disaster, too. Since all
technologies involved in p2p-sip have been there integrally proven for years
(by Skype) and most of them seem ought to be tightly coupled for feasibility
and interoperability, I believe a "do it once for all and forever" is the
right approach for the late comers (-: us :-) to attract world attention and
deroot all weeds in the future.

I wish I may add one more sentence: "I'm sorry, but if it doesn't address
SECURITY, TOP-LEVEL FEATURES, SPECIFIC OVERLAY(S)... among others we might
as well go home. The whole point of the overlay is to get around structures
such as NAT ... have STRONG SECURITY beneath it, have TOP-LEVEL FEATURES
(voicemail, centrex, ...) above it, and have INTEROPERABILITY surrounding
it."

peter

*** no bash pls. we are of the same country ***

----- Original Message -----
From: "Michael Slavitch" <slavitch at objectworld.com>
To: "Brijesh Kumar" <kumarb at research.panasonic.com>; "Bruce Lowekamp"
<lowekamp at cs.wm.edu>; "Philip Matthews" <philip_matthews at magma.ca>
Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 7:14 AM
Subject: Re: [p2p-sip] NATs and P2P


> This I believe is a disastrous approach.  If it's not resolved day one
> it will not happen.
>
> SIP put this off ten years ago, because NAT was thought bad, something
> that would go away. It didn't happen.
>
> I'm sorry, but if it doesn't address NAT we might as well go home.  The
> whole point of the overlay is to get around structures such as NAT,
> different addressing (IPV4/IPV6), walled gardens and the like.
>
> In a pure IPV6 world we already have a P2P SIP protocol.  It is called
> SIP.
>
> M
>
>
> Michael Slavitch           Lead, Strategic Initiatives
> slavitch at objectworld.com   Objectworld Communications Corp.
> +1.613.599.9698 .225       Ottawa, Ontario, Canada
>
> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Brijesh Kumar
> Sent: March 21, 2006 10:18 AM
> To: Bruce Lowekamp; Philip Matthews
> Cc: P2P-SIP
> Subject: Re: [p2p-sip] NATs and P2P
>
> I think Bruce's approach makes sense.
>
> The first task should be to come out with a  base specification that
> works. If nothing else, this can be deployed in a hypothetical pure IPv6
> world, or  in a  limited enterprise universe where traversing NAT may
> not be an issue.
>
> The second task should be to modify the spec  that works with NAT.
> Trying to resolve NAT issue from day one is good idea, but one can
> equally treat is as an extension of the core design.  It is just a
> matter of convenience rather than  any hard logic.
>
> Cheers,
>
> --bk
>
> ----- Original Message -----
> From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
> To: "Philip Matthews" <philip_matthews at magma.ca>
> Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
> Sent: Monday, March 20, 2006 8:58 PM
> Subject: Re: [p2p-sip] NATs and P2P
>
>
> > Philip,
> >
> > I absolutely agree that we need to have a solution for this type of
> > environment.   IMHO, some of the worries about NATs causing problems
> > have been somewhat overstated.  As you point out, the majority of NATs
> > in the wild are address independent mapping, and it is relatively
> > trivial to punch holes to establish communications.  But you're also
> > right that, especially for corporate networks, this doesn't solve the
> > entire problem.
> >
> > I spent some time thinking about how to solve the specific problems
> > you pose in terms of the protocol we have been playing with in
> > draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
> > there (not because we don't think it's a challenge, but just because
> > we want to work on a basic protocol first).  If I was to address some
> > of the problems you pose using that framework (modified chord), then
> > one approach would be to remove the requirement that DHT operations be
> > iterative and allow them to be recursive.  If DHT operations can be
> > recursive, then the number of connections that a node must maintain
> > through the firewall is limited by the size of the finger table.  That
> > number doesn't have to be very big to have a rather efficient DHT.
> > (never more than 128, and I think in practice the number can be much
> > smaller.  If your guess of 160 is correct, we're more than OK here.)
> >
> > I'm a bit unhappy with that because I really like the iterative
> > approach for a variety of reasons, but certainly the protocol could be
> > modified that way and a hybrid environment could be established where
> > some nodes are open (or easy to request opens) and support iterative
> > queries with anyone, but others are recursive queries only.  One
> > problem is that the burden of having a finger-table entry to a
> > recursive-only host is placed on the non-recursive node that would
> > otherwise redirect to its finger-table entry, but now must recurse the
> > request.
> >
> > Right now the scheme would also have problems because, as Eric rightly
> > points out in a separate thread on this mailing list, we don't have a
> > good way of selecting Identifiers for nodes behind NATs (in
> > particular).  But if you have node certificates or another way of
> > authenticating your nodes, this is less of a problem.
> >
> > I want to make it clear that I'm not ready to propose a specific
> > solution, but I think there are ways to support structured DHT
> > algorithms that are not really very painful and work reasonably well
> > inside these environments.  Clearly support for such an environment
> > needs to be a listed in eventual requirements docs.
> >
> > Bruce
> >
> >
> > On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > > [I guess all this traffic on the mailing list will teach me not to
> go
> > > on vacation
> > > just before an IETF meeting!]
> > >
> > > As I catch up on all the messages about NATs over the past two
> weeks,
> > > it seems to me that many people are thinking only of the "Public P2P
> > > Service Provider"
> > > use-case as described in the use-cases document. In other words, a
> Skype
> > > competitor.
> > >
> > > However, the use-cases document that David Bryan and his co-authors
> > > wrote
> > > identified a number of other use-cases and it seems to me that these
> > > have somewhat
> > > different NAT traversal requirements.
> > >
> > > In particular, in some of these other use-cases, it seems to me that
> > > we CANNOT assume
> > > there are peers with public IP addresses.
> > >
> > > For example, consider the "Presence using Multimedia Consumer
> > > Electronics Devices"
> > > use-case (section 3.1.3) -- essentially a P2P network of multimedia
> > > consumer electronics devices
> > > that need presence information. Who is going to pay the extra money
> > > to give their digital camera (or those
> > > neat 770 tablets that Nokia is demoing here in Dallas?) a public
> IPv4
> > > address?? On the contrary,
> > > devices like this are almost certainly going to have private IP
> > > addresses -- it is very common today for
> > > wireless internet providers to place a big NAT in front of their
> > > entire network and give private addresses
> > > to all their customers.
> > >
> > > Or consider the "IP PBX" use-case -- a IP PBX system for a company
> > > with a number of small branches
> > > scattered throughout the world. Each branch is going to have a NAT
> in
> > > front of its network, and all
> > > the phones in that branch are going to have private IP addresses.
> > > None of the phones are going to
> > > have public IP addresses.
> > >
> > > It is handling the NAT traversal issues for use-cases like these
> that
> > > Eric Cooper and I wrote our
> > > internet-draft on NAT Traversal for P2P:
> > >
> http://www.ietf.org/internet-drafts/draft-matthews-p2psip-nats-
> > > and-overlays-00.txt
> > >
> > > - Philip
> > >
> > > _______________________________________________
> > > p2p-sip mailing list
> > > p2p-sip at cs.columbia.edu
> > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > >
> > >
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>
> --
>
>
> -----------------------------------------------
>
> This transmission (including any attachments) may contain confidential
information, privileged material (including material protected by the
solicitor-client or other applicable privileges), or constitute non-public
information. Any use of this information by anyone other than the intended
recipient is prohibited. If you have received this transmission in error,
please immediately reply to the sender and delete this information from your
system. Use, dissemination, distribution, or reproduction of this
transmission by unintended recipients is not authorized and may be unlawful.
>
> Any views, opinions and content presented in this e-mail are solely those
of the author and may not represent those of Objectworld Communications
Corp.
>
>
>


From sathya at research.panasonic.com  Tue Mar 21 14:10:03 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Tue, 21 Mar 2006 14:10:03 -0500
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <842f8a3e.8a3e842f@pintlmail.MITL.Research.Panasonic.COM>

Dean -

> 
> I propose that we first define an abstract model of an overlay  
> network that includes the requisite primitives INSERT and QUERY  
> (anything else needed?) and use this abstraction as our middle layer.

There is a section in draft-shim-sipping-p2p-arch-00.txt on P2P Overlay API which proposes a set of such primitives.
 
> For the lower-layer, we do a binding from the abstraction to one or  
> more distributed database models, probably starting with CHORD. This  
> leaves our P2PSIP architecture independent of the database, but gives a proof-of-concept adequate for implementation.

I am not sure if CHORD is the best option to start with. But, I will wait to hear other opinions/reasons before taking a position on this one.

> 
> For the upper-layer, we have one model defined using SIP REGISTER  
> messages. We've also heard of work that uses REGISTER for the  
> "INSERT" primitive and SUBSCRIBE/NOTIFY for the "query" primitive.
> 
> I'm also very interested in seeing if there's an obvious binding for  
> TURN here, as that might kill two birds with one stone.
> 
> I propose we start with the REGISTER/REGISTER model from the bryan 
> draft and talk that through a little further before we think about changing it.

My understanding of Bryan's draft is that it tries to extend SIP for overlay maintenance and resource location, i.e. a single layer approach (Please correct me if I am wrong). So, proposing to start with that model implicitely chooses that approach and ignores the other possibility of having two separate layers as proposed in draft-shim. I think we should have an explicit discussion on that "key" question before proceeding any further. 

thanks,
Sathya

> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From philip_matthews at magma.ca  Tue Mar 21 14:19:20 2006
From: philip_matthews at magma.ca (Philip Matthews)
Date: Tue, 21 Mar 2006 13:19:20 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <200603210319.k2L3JIrj001729@cs.columbia.edu>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
Message-ID: <B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>

On 20-Mar-06, at 21:19 , David Barrett wrote:

> I'd add to that this key question:
>
> 	"Will we extend SIP, or create a new protocol?"
>
> I'm finding it hard to make any headway without understanding the  
> above.
> Basically, I see two major directions we could go:
>
> 1) Extend SIP with overlay-maintenance and resource-location messages.
>
> 2) Create a new overlay protocol and develop bindings for SIP and  
> ICE (eg,
> distributed proxy service and STUN/TURN resource-location service).
>
> It seems this high-level decision keeps coming up again and again in
> discussions of the smaller issues.
>

For what it is worth, I can say that my own views on this subject  
have changed.

For all of last year, I strongly believed that there should be two  
distinct layers:
a SIP layer and a P2P layer (i.e., option 2). See the arguments I  
wrote in
   http://www.p2psip.org/drafts/draft-matthews-sipping-p2p-industrial- 
strength-00.txt
as well as those on Alan Johnston in
   http://www.p2psip.org/drafts/draft-johnston-sipping-p2p-ipcom-01.txt

However, in the last few months, I have come to see that there are  
some good reasons
to put the two layers together into one (i.e., option 1):
	a) NAT Traversal becomes easier because there is just one port,  
rather than two
	   (this assumes that the P2P layer would run on a different port  
than SIP)
	b) Possible performance improvements. With two layers, you have to  
first ask
	  "where is user U?" and then send the Invite message. With one  
layer, there is
	  the possibility of sending the Invite and having it routed to the  
user.
	  (David et al removed this from their latest draft, but this was an  
option in the
	  earlier version, and also in the work done by Henning's group).

As others have pointed out, there are also drawbacks, so I haven't  
concluded anything
yet, but I am a lot more open to option 1 than I was a few months ago.


- Philip

PS. What this has to do with the use-cases document, however, I am  
not clear on  ;-)

From sathya at research.panasonic.com  Tue Mar 21 14:42:47 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Tue, 21 Mar 2006 14:42:47 -0500
Subject: [p2p-sip] Single layer vs two layers
Message-ID: <69b45c18.5c1869b4@pintlmail.MITL.Research.Panasonic.COM>

After I wrote this email, I went back and read Bryan's draft and now I am not sure if it is proposing a single layer approach. It looks to me like it is proposing a two-layer approach, with the SIP layer exactly same as SIP and the overlay layer using a new SIP-like protocol. Is that a correct description?

thanks,
Sathya

----- Original Message -----
From: Sathya Narayanan <sathya at research.panasonic.com>
Date: Tuesday, March 21, 2006 2:10 pm
Subject: Re: [p2p-sip] Some comments on Use-cases document

> Dean -
> 
> > 
> > I propose that we first define an abstract model of an overlay  
> > network that includes the requisite primitives INSERT and QUERY  
> > (anything else needed?) and use this abstraction as our middle 
> layer.
> There is a section in draft-shim-sipping-p2p-arch-00.txt on P2P 
> Overlay API which proposes a set of such primitives.
> 
> > For the lower-layer, we do a binding from the abstraction to one 
> or  
> > more distributed database models, probably starting with CHORD. 
> This  
> > leaves our P2PSIP architecture independent of the database, but 
> gives a proof-of-concept adequate for implementation.
> 
> I am not sure if CHORD is the best option to start with. But, I 
> will wait to hear other opinions/reasons before taking a position 
> on this one.
> 
> > 
> > For the upper-layer, we have one model defined using SIP 
> REGISTER  
> > messages. We've also heard of work that uses REGISTER for the  
> > "INSERT" primitive and SUBSCRIBE/NOTIFY for the "query" primitive.
> > 
> > I'm also very interested in seeing if there's an obvious binding 
> for  
> > TURN here, as that might kill two birds with one stone.
> > 
> > I propose we start with the REGISTER/REGISTER model from the 
> bryan 
> > draft and talk that through a little further before we think 
> about changing it.
> 
> My understanding of Bryan's draft is that it tries to extend SIP 
> for overlay maintenance and resource location, i.e. a single layer 
> approach (Please correct me if I am wrong). So, proposing to start 
> with that model implicitely chooses that approach and ignores the 
> other possibility of having two separate layers as proposed in 
> draft-shim. I think we should have an explicit discussion on that 
> "key" question before proceeding any further. 
> 
> thanks,
> Sathya
> 
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > 
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From sathya at research.panasonic.com  Tue Mar 21 14:59:47 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Tue, 21 Mar 2006 14:59:47 -0500
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <81ed652d.652d81ed@pintlmail.MITL.Research.Panasonic.COM>

Philips -

<snip>

> 	b) Possible performance improvements. With two layers, you have to  
> first ask "where is user U?" and then send the Invite message. 
> With one layer, there is the possibility of sending the Invite and having it 
> routed to the user.

FWIW, I am against exactly the idea of merging QUERY and INVITE operations (Hope I can change your mind back on this one :-) ). How to verify the validity of  the resources records at the storage node, how to integrate with replications of stored resource records  (I am assuming we will (at some point) need replications to address reliability of the overlay), etc. becomes a lot more difficult if we do this merging.

In my mind, these two layers and their operations separate logically very well and keeping them independent will keep things simpler and allow for extending/improving each of these layers functionality independently without affecting the other.

thanks,
Sathya


>          (David et al removed this from their latest draft, but 
> this was an  
> option in the
>          earlier version, and also in the work done by Henning's 
> group).
> As others have pointed out, there are also drawbacks, so I haven't 
> 
> concluded anything
> yet, but I am a lot more open to option 1 than I was a few months ago.
> 
> 
> - Philip
> 
> PS. What this has to do with the use-cases document, however, I am 
> 
> not clear on  ;-)
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From lowekamp at cs.wm.edu  Tue Mar 21 15:10:50 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 21 Mar 2006 14:10:50 -0600
Subject: [p2p-sip] Single layer vs two layers
In-Reply-To: <69b45c18.5c1869b4@pintlmail.MITL.Research.Panasonic.COM>
References: <69b45c18.5c1869b4@pintlmail.MITL.Research.Panasonic.COM>
Message-ID: <e9132f820603211210v51a29c9ai157003571270ae50@mail.gmail.com>

I would say it's SIP, not a SIP-like protocol, but that's a minor
terminology issue.

-01 used INVITE to search the DHT for the contact and to place the
call.  There was some concern that this might introduce issues when
connecting to non-P2P UAs with p2p headers (even if they are just
options or extension headers).

To address that issue, -02 uses REGISTER to query for the resource
registration (as specified by 3261, a REGISTER without a contact is a
query).   The UAC then places the call using purely conventional SIP
to the target UAS.

So it's two-layer in the sense that operations never traverse from the
DHT to non-DHT portions of the protocol.  But it's all implemented on
top of a standard SIP stack.

Bruce


On 3/21/06, Sathya Narayanan <sathya at research.panasonic.com> wrote:
> After I wrote this email, I went back and read Bryan's draft and now I am not sure if it is proposing a single layer approach. It looks to me like it is proposing a two-layer approach, with the SIP layer exactly same as SIP and the overlay layer using a new SIP-like protocol. Is that a correct description?
>
> thanks,
> Sathya
>
> ----- Original Message -----
> From: Sathya Narayanan <sathya at research.panasonic.com>
> Date: Tuesday, March 21, 2006 2:10 pm
> Subject: Re: [p2p-sip] Some comments on Use-cases document
>
> > Dean -
> >
> > >
> > > I propose that we first define an abstract model of an overlay
> > > network that includes the requisite primitives INSERT and QUERY
> > > (anything else needed?) and use this abstraction as our middle
> > layer.
> > There is a section in draft-shim-sipping-p2p-arch-00.txt on P2P
> > Overlay API which proposes a set of such primitives.
> >
> > > For the lower-layer, we do a binding from the abstraction to one
> > or
> > > more distributed database models, probably starting with CHORD.
> > This
> > > leaves our P2PSIP architecture independent of the database, but
> > gives a proof-of-concept adequate for implementation.
> >
> > I am not sure if CHORD is the best option to start with. But, I
> > will wait to hear other opinions/reasons before taking a position
> > on this one.
> >
> > >
> > > For the upper-layer, we have one model defined using SIP
> > REGISTER
> > > messages. We've also heard of work that uses REGISTER for the
> > > "INSERT" primitive and SUBSCRIBE/NOTIFY for the "query" primitive.
> > >
> > > I'm also very interested in seeing if there's an obvious binding
> > for
> > > TURN here, as that might kill two birds with one stone.
> > >
> > > I propose we start with the REGISTER/REGISTER model from the
> > bryan
> > > draft and talk that through a little further before we think
> > about changing it.
> >
> > My understanding of Bryan's draft is that it tries to extend SIP
> > for overlay maintenance and resource location, i.e. a single layer
> > approach (Please correct me if I am wrong). So, proposing to start
> > with that model implicitely chooses that approach and ignores the
> > other possibility of having two separate layers as proposed in
> > draft-shim. I think we should have an explicit discussion on that
> > "key" question before proceeding any further.
> >
> > thanks,
> > Sathya
> >
> > > _______________________________________________
> > > p2p-sip mailing list
> > > p2p-sip at cs.columbia.edu
> > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > >
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>


From enrico.marocco at telecomitalia.it  Tue Mar 21 15:14:08 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Tue, 21 Mar 2006 14:14:08 -0600
Subject: [p2p-sip] Single layer vs two layers
In-Reply-To: <69b45c18.5c1869b4@pintlmail.MITL.Research.Panasonic.COM>
References: <69b45c18.5c1869b4@pintlmail.MITL.Research.Panasonic.COM>
Message-ID: <44205E90.7070205@telecomitalia.it>

Sathya Narayanan wrote:
> After I wrote this email, I went back and read Bryan's draft and now I
> am not sure if it is proposing a single layer approach. It looks to me
> like it is proposing a two-layer approach, with the SIP layer exactly
> same as SIP and the overlay layer using a new SIP-like protocol. Is that
> a correct description?

It is.  Note that, pragmatically, the only detail that makes the
difference is that in version 1 user queries are accomplished with a
INVITE message and the result is conveyed in the last 302 response,
while in version 2 it is accomplished with a REGISTER and the result is
conveyed in the 200 response.

Enrico
--------------------------------------------------------------------

CONFIDENTIALITY NOTICE

This message and its attachments are addressed solely to the persons above and may contain confidential information. If you have received the message in error, be informed that any use of the content hereof is prohibited. Please return it immediately to the sender and delete the message. Should you have any questions, please contact us by replying to webmaster at telecomitalia.it.

        Thank you

                                        www.telecomitalia.it

--------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060321/c781cbd5/attachment.html

From sathya at research.panasonic.com  Tue Mar 21 15:30:26 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Tue, 21 Mar 2006 15:30:26 -0500
Subject: [p2p-sip] Single layer vs two layers
Message-ID: <4ca2879a.879a4ca2@pintlmail.MITL.Research.Panasonic.COM>


> I would say it's SIP, not a SIP-like protocol, but that's a minor terminology issue.

It is sorta minor. But, I would like to explore this point further, please indulge me. I called it SIP-like because there are proposed changes to the packet formats (new options) and behavior of nodes that receive these messages (for example the behavior in response to the receipt of a REGISTER message), to achieve something SIP was not designed for in the first place. 

Even assuming that the semantics of REGISTER message matches the required semantics for node join and resource add of a DHT, how are we going to address replication of resources on DHT. We have to extend the way REGISTER message is used to support whatever form of replication is developed, while AFAIK replication is not part of the REGISTER semantics today. And I am sure there will be other requirements to have a stable overlay network that will need further changes.

So, the question is how far we have to go away from the current purpose of SIP protocol before we start calling it by some other name?

my two cents,
Sathya

PS: I don't have a position regarding the use of something very much like SIP for the overlay protocol yet, I am just trying to understand the pros and cons.

> 
> -01 used INVITE to search the DHT for the contact and to place the
> call.  There was some concern that this might introduce issues when
> connecting to non-P2P UAs with p2p headers (even if they are just
> options or extension headers).
> 
> To address that issue, -02 uses REGISTER to query for the resource
> registration (as specified by 3261, a REGISTER without a contact 
> is a
> query).   The UAC then places the call using purely conventional SIP
> to the target UAS.
> 
> So it's two-layer in the sense that operations never traverse from the
> DHT to non-DHT portions of the protocol.  But it's all implemented on
> top of a standard SIP stack.
> 
> Bruce
> 
> 
> On 3/21/06, Sathya Narayanan <sathya at research.panasonic.com> wrote:
> > After I wrote this email, I went back and read Bryan's draft and 
> now I am not sure if it is proposing a single layer approach. It 
> looks to me like it is proposing a two-layer approach, with the 
> SIP layer exactly same as SIP and the overlay layer using a new 
> SIP-like protocol. Is that a correct description?
> >
> > thanks,
> > Sathya
> >
> > ----- Original Message -----
> > From: Sathya Narayanan <sathya at research.panasonic.com>
> > Date: Tuesday, March 21, 2006 2:10 pm
> > Subject: Re: [p2p-sip] Some comments on Use-cases document
> >
> > > Dean -
> > >
> > > >
> > > > I propose that we first define an abstract model of an overlay
> > > > network that includes the requisite primitives INSERT and QUERY
> > > > (anything else needed?) and use this abstraction as our middle
> > > layer.
> > > There is a section in draft-shim-sipping-p2p-arch-00.txt on P2P
> > > Overlay API which proposes a set of such primitives.
> > >
> > > > For the lower-layer, we do a binding from the abstraction to one
> > > or
> > > > more distributed database models, probably starting with CHORD.
> > > This
> > > > leaves our P2PSIP architecture independent of the database, but
> > > gives a proof-of-concept adequate for implementation.
> > >
> > > I am not sure if CHORD is the best option to start with. But, I
> > > will wait to hear other opinions/reasons before taking a position
> > > on this one.
> > >
> > > >
> > > > For the upper-layer, we have one model defined using SIP
> > > REGISTER
> > > > messages. We've also heard of work that uses REGISTER for the
> > > > "INSERT" primitive and SUBSCRIBE/NOTIFY for the "query" 
> primitive.> > >
> > > > I'm also very interested in seeing if there's an obvious binding
> > > for
> > > > TURN here, as that might kill two birds with one stone.
> > > >
> > > > I propose we start with the REGISTER/REGISTER model from the
> > > bryan
> > > > draft and talk that through a little further before we think
> > > about changing it.
> > >
> > > My understanding of Bryan's draft is that it tries to extend SIP
> > > for overlay maintenance and resource location, i.e. a single layer
> > > approach (Please correct me if I am wrong). So, proposing to start
> > > with that model implicitely chooses that approach and ignores the
> > > other possibility of having two separate layers as proposed in
> > > draft-shim. I think we should have an explicit discussion on that
> > > "key" question before proceeding any further.
> > >
> > > thanks,
> > > Sathya
> > >
> > > > _______________________________________________
> > > > p2p-sip mailing list
> > > > p2p-sip at cs.columbia.edu
> > > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > > >
> > >
> > > _______________________________________________
> > > p2p-sip mailing list
> > > p2p-sip at cs.columbia.edu
> > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > >
> >
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
> 


From lowekamp at cs.wm.edu  Tue Mar 21 15:37:34 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 21 Mar 2006 14:37:34 -0600
Subject: [p2p-sip] Single layer vs two layers
In-Reply-To: <4ca2879a.879a4ca2@pintlmail.MITL.Research.Panasonic.COM>
References: <4ca2879a.879a4ca2@pintlmail.MITL.Research.Panasonic.COM>
Message-ID: <e9132f820603211237k443f4a09x9212b447995b974@mail.gmail.com>

Our -02 draft suggests the idea of making a UA responsible for storing
replicas by modifying the resource URI and registering each replica
separately.  There are pros and cons to this, but it does avoid having
the responsible node also be responsible for maintaining replicas just
because it received a REGISTER.  I haven't contemplated yet whether we
can simply use sip-outbound techniques to accomplish replicas, but
that might be an interesting question.

Bruce

On 3/21/06, Sathya Narayanan <sathya at research.panasonic.com> wrote:
> Even assuming that the semantics of REGISTER message matches the required semantics for node join and resource add of a DHT, how are we going to address replication of resources on DHT. We have to extend the way REGISTER message is used to support whatever form of replication is developed, while AFAIK replication is not part of the REGISTER semantics today. And I am sure there will be other requirements to have a stable overlay network that will need further changes.


From lowekamp at cs.wm.edu  Tue Mar 21 15:41:06 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 21 Mar 2006 14:41:06 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
	<B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
	<e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
Message-ID: <e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>

resending since this hasn't been acked by the mailing list or appeared
after over 30 minutes.  (I'm sure it will now immediately appear)

On 3/21/06, Bruce Lowekamp <lowekamp at cs.wm.edu> wrote:
> I don't thnk it has anything to do with use-cases, either
>
> but
>
> Building DHT maintenance on top of SIP gives us all of the advantages
> of the routing, addressing, naming, and security issues already built
> into SIP.  Plus the NAT traversal capabilities of STUN, TURN, and ICE.
>  While not perfect for our use, I think with very minimal
> modifications (whatever the final protocol is) they will provide a
> very good solution.
>
> A proposal to use something else needs to:
> - explain what the shortcomings of SIP are for this purpose
> - explain how a new or different solution will provide equivalent functionality
> - explain how/why the new protocol will be better after resolving the
> complexities that SIP has become so complex to address
> - make a convincing enough case to justify deploying devices (and
> we're frequently talking about very small devices) to implement two
> separate protocol stacks.
>
>
> The exact DHT algorithm we implement, the exact way the we use SIP,
> the methods, etc I think are all going to need some careful thought.
> But for P2P SIP, I think SIP is the obvious choice for DHT operations,
> and I would consider it the default unless a convincing argument is
> made against it.
>
> A number of comments have been made stating that NATs must be taken
> into account from the beginning.  Again, one of the issues that using
> SIP helps address already is NAT traversal.  It's not perfect for
> signalling, but the framework is there.
>
> Bruce
>
> On 3/21/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > On 20-Mar-06, at 21:19 , David Barrett wrote:
> >
> > > I'd add to that this key question:
> > >
> > >       "Will we extend SIP, or create a new protocol?"
> > >
> > > I'm finding it hard to make any headway without understanding the
> > > above.
> > > Basically, I see two major directions we could go:
> > >
> > > 1) Extend SIP with overlay-maintenance and resource-location messages.
> > >
> > > 2) Create a new overlay protocol and develop bindings for SIP and
> > > ICE (eg,
> > > distributed proxy service and STUN/TURN resource-location service).
> > >
> > > It seems this high-level decision keeps coming up again and again in
> > > discussions of the smaller issues.
> > >
> >
> > For what it is worth, I can say that my own views on this subject
> > have changed.
> >
> > For all of last year, I strongly believed that there should be two
> > distinct layers:
> > a SIP layer and a P2P layer (i.e., option 2). See the arguments I
> > wrote in
> >    http://www.p2psip.org/drafts/draft-matthews-sipping-p2p-industrial-
> > strength-00.txt
> > as well as those on Alan Johnston in
> >    http://www.p2psip.org/drafts/draft-johnston-sipping-p2p-ipcom-01.txt
> >
> > However, in the last few months, I have come to see that there are
> > some good reasons
> > to put the two layers together into one (i.e., option 1):
> >         a) NAT Traversal becomes easier because there is just one port,
> > rather than two
> >            (this assumes that the P2P layer would run on a different port
> > than SIP)
> >         b) Possible performance improvements. With two layers, you have to
> > first ask
> >           "where is user U?" and then send the Invite message. With one
> > layer, there is
> >           the possibility of sending the Invite and having it routed to the
> > user.
> >           (David et al removed this from their latest draft, but this was an
> > option in the
> >           earlier version, and also in the work done by Henning's group).
> >
> > As others have pointed out, there are also drawbacks, so I haven't
> > concluded anything
> > yet, but I am a lot more open to option 1 than I was a few months ago.
> >
> >
> > - Philip
> >
> > PS. What this has to do with the use-cases document, however, I am
> > not clear on  ;-)
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
>


From lowekamp at cs.wm.edu  Tue Mar 21 15:02:05 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 21 Mar 2006 14:02:05 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
	<B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
Message-ID: <e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>

I don't thnk it has anything to do with use-cases, either

but

Building DHT maintenance on top of SIP gives us all of the advantages
of the routing, addressing, naming, and security issues already built
into SIP.  Plus the NAT traversal capabilities of STUN, TURN, and ICE.
 While not perfect for our use, I think with very minimal
modifications (whatever the final protocol is) they will provide a
very good solution.

A proposal to use something else needs to:
- explain what the shortcomings of SIP are for this purpose
- explain how a new or different solution will provide equivalent functionality
- explain how/why the new protocol will be better after resolving the
complexities that SIP has become so complex to address
- make a convincing enough case to justify deploying devices (and
we're frequently talking about very small devices) to implement two
separate protocol stacks.


The exact DHT algorithm we implement, the exact way the we use SIP,
the methods, etc I think are all going to need some careful thought. 
But for P2P SIP, I think SIP is the obvious choice for DHT operations,
and I would consider it the default unless a convincing argument is
made against it.

A number of comments have been made stating that NATs must be taken
into account from the beginning.  Again, one of the issues that using
SIP helps address already is NAT traversal.  It's not perfect for
signalling, but the framework is there.

Bruce

On 3/21/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> On 20-Mar-06, at 21:19 , David Barrett wrote:
>
> > I'd add to that this key question:
> >
> >       "Will we extend SIP, or create a new protocol?"
> >
> > I'm finding it hard to make any headway without understanding the
> > above.
> > Basically, I see two major directions we could go:
> >
> > 1) Extend SIP with overlay-maintenance and resource-location messages.
> >
> > 2) Create a new overlay protocol and develop bindings for SIP and
> > ICE (eg,
> > distributed proxy service and STUN/TURN resource-location service).
> >
> > It seems this high-level decision keeps coming up again and again in
> > discussions of the smaller issues.
> >
>
> For what it is worth, I can say that my own views on this subject
> have changed.
>
> For all of last year, I strongly believed that there should be two
> distinct layers:
> a SIP layer and a P2P layer (i.e., option 2). See the arguments I
> wrote in
>    http://www.p2psip.org/drafts/draft-matthews-sipping-p2p-industrial-
> strength-00.txt
> as well as those on Alan Johnston in
>    http://www.p2psip.org/drafts/draft-johnston-sipping-p2p-ipcom-01.txt
>
> However, in the last few months, I have come to see that there are
> some good reasons
> to put the two layers together into one (i.e., option 1):
>         a) NAT Traversal becomes easier because there is just one port,
> rather than two
>            (this assumes that the P2P layer would run on a different port
> than SIP)
>         b) Possible performance improvements. With two layers, you have to
> first ask
>           "where is user U?" and then send the Invite message. With one
> layer, there is
>           the possibility of sending the Invite and having it routed to the
> user.
>           (David et al removed this from their latest draft, but this was an
> option in the
>           earlier version, and also in the work done by Henning's group).
>
> As others have pointed out, there are also drawbacks, so I haven't
> concluded anything
> yet, but I am a lot more open to option 1 than I was a few months ago.
>
>
> - Philip
>
> PS. What this has to do with the use-cases document, however, I am
> not clear on  ;-)
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>


From huang-ming.pan at comcast.net  Tue Mar 21 16:20:18 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Tue, 21 Mar 2006 13:20:18 -0800
Subject: [p2p-sip] Some comments on Use-cases document
References: <200603210319.k2L3JIrj001729@cs.columbia.edu><B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca><e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
	<e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
Message-ID: <010b01c64d2d$41c340a0$6400a8c0@comcast.net>

> The exact DHT algorithm we implement, the exact way the we use SIP,
> the methods, etc I think are all going to need some careful thought.
> But for P2P SIP, I think SIP is the obvious choice for DHT operations,
> and I would consider it the default unless a convincing argument is
> made against it.

Not an argument but a question: how to efficiently implement Kademlia
STORE_VALUE primitive in SIP INVITE/REGISTER when the stored values can be
of any length and type?

peter

*** dont bash me, we are of the same country ***

----- Original Message -----
From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
To: "Philip Matthews" <philip_matthews at magma.ca>
Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 12:41 PM
Subject: Re: [p2p-sip] Some comments on Use-cases document


> resending since this hasn't been acked by the mailing list or appeared
> after over 30 minutes.  (I'm sure it will now immediately appear)
>
> On 3/21/06, Bruce Lowekamp <lowekamp at cs.wm.edu> wrote:
> > I don't thnk it has anything to do with use-cases, either
> >
> > but
> >
> > Building DHT maintenance on top of SIP gives us all of the advantages
> > of the routing, addressing, naming, and security issues already built
> > into SIP.  Plus the NAT traversal capabilities of STUN, TURN, and ICE.
> >  While not perfect for our use, I think with very minimal
> > modifications (whatever the final protocol is) they will provide a
> > very good solution.
> >
> > A proposal to use something else needs to:
> > - explain what the shortcomings of SIP are for this purpose
> > - explain how a new or different solution will provide equivalent
functionality
> > - explain how/why the new protocol will be better after resolving the
> > complexities that SIP has become so complex to address
> > - make a convincing enough case to justify deploying devices (and
> > we're frequently talking about very small devices) to implement two
> > separate protocol stacks.
> >
> >
> > The exact DHT algorithm we implement, the exact way the we use SIP,
> > the methods, etc I think are all going to need some careful thought.
> > But for P2P SIP, I think SIP is the obvious choice for DHT operations,
> > and I would consider it the default unless a convincing argument is
> > made against it.
> >
> > A number of comments have been made stating that NATs must be taken
> > into account from the beginning.  Again, one of the issues that using
> > SIP helps address already is NAT traversal.  It's not perfect for
> > signalling, but the framework is there.
> >
> > Bruce
> >
> > On 3/21/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > > On 20-Mar-06, at 21:19 , David Barrett wrote:
> > >
> > > > I'd add to that this key question:
> > > >
> > > >       "Will we extend SIP, or create a new protocol?"
> > > >
> > > > I'm finding it hard to make any headway without understanding the
> > > > above.
> > > > Basically, I see two major directions we could go:
> > > >
> > > > 1) Extend SIP with overlay-maintenance and resource-location
messages.
> > > >
> > > > 2) Create a new overlay protocol and develop bindings for SIP and
> > > > ICE (eg,
> > > > distributed proxy service and STUN/TURN resource-location service).
> > > >
> > > > It seems this high-level decision keeps coming up again and again in
> > > > discussions of the smaller issues.
> > > >
> > >
> > > For what it is worth, I can say that my own views on this subject
> > > have changed.
> > >
> > > For all of last year, I strongly believed that there should be two
> > > distinct layers:
> > > a SIP layer and a P2P layer (i.e., option 2). See the arguments I
> > > wrote in
> > >    http://www.p2psip.org/drafts/draft-matthews-sipping-p2p-industrial-
> > > strength-00.txt
> > > as well as those on Alan Johnston in
> > >
http://www.p2psip.org/drafts/draft-johnston-sipping-p2p-ipcom-01.txt
> > >
> > > However, in the last few months, I have come to see that there are
> > > some good reasons
> > > to put the two layers together into one (i.e., option 1):
> > >         a) NAT Traversal becomes easier because there is just one
port,
> > > rather than two
> > >            (this assumes that the P2P layer would run on a different
port
> > > than SIP)
> > >         b) Possible performance improvements. With two layers, you
have to
> > > first ask
> > >           "where is user U?" and then send the Invite message. With
one
> > > layer, there is
> > >           the possibility of sending the Invite and having it routed
to the
> > > user.
> > >           (David et al removed this from their latest draft, but this
was an
> > > option in the
> > >           earlier version, and also in the work done by Henning's
group).
> > >
> > > As others have pointed out, there are also drawbacks, so I haven't
> > > concluded anything
> > > yet, but I am a lot more open to option 1 than I was a few months ago.
> > >
> > >
> > > - Philip
> > >
> > > PS. What this has to do with the use-cases document, however, I am
> > > not clear on  ;-)
> > > _______________________________________________
> > > p2p-sip mailing list
> > > p2p-sip at cs.columbia.edu
> > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > >
> > >
> >
>
>
>


From slavitch at gmail.com  Tue Mar 21 16:23:07 2006
From: slavitch at gmail.com (Michael Slavitch)
Date: Tue, 21 Mar 2006 15:23:07 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
	<B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
	<e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
	<e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
Message-ID: <1fc32c8f0603211323l2e741cebnef4c0fa6bd7d8978@mail.gmail.com>

I believe that SIP is now such a monster that any attempt to P2P the SIP
protocol itself is going to be a total implementation mess.   A separate
overlay that links nicely with SIP but requires no changes to SIP/RTP/SDP
apart from understanding different naming syntaxes (ie, something not DNS or
numbers) is a lot more clean with a lot less screaming from other working
groups that are obsessed with IMS and the CSS7.

There are problems with overlays, but the problem space is being slowly
examined by people in this mailing list.

The key issues with overlays are as follows:

1. Establishment/bootstrap
2. Identity and naming
3. Persistence
4. Security

All without servers,  all working perhaps without acesss to a public
internet.  I think these can only be done via a separate overlay.   I've
done some investigation, it appears that the biggest issues
(Identity/Traversal/Persistance) are less important issues.  I'll discuss
more at the BOF and the Ad Hoc, I've not had time to get things down on
paper in internet-draft form.   Will do tho.

To change SIP to support all those things is going to be a bloody
nightmare.  A smart lean overlay that sits under SIP?
Done well, the only noticed change for other working groups would be to add
a new naming scheme outside DNS and perhaps some minor changes to ICE to
support  P2PSIP namespaces and movement between P2PSIP domains and SIP
domains.  That's it.  Statewise I don't think we have any need to change ICE
requirements.

So if we want to meet the goals of the charter and to follow IETF rules we
should strongly consider SIP and the P2P overlay as different things.  If
the charter is to follow the rules of reuse, minimal changes,  and respect
the very clear statements coming from MMUSIC (which I fully respect, btw),
SIP and SIPING working groups I don't think there is a choice.







--
Michael Slavitch
Ottawa, Ontario Canada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060321/765c9712/attachment.html

From lowekamp at cs.wm.edu  Tue Mar 21 16:30:07 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 21 Mar 2006 15:30:07 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <1fc32c8f0603211323l2e741cebnef4c0fa6bd7d8978@mail.gmail.com>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
	<B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
	<e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
	<e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
	<1fc32c8f0603211323l2e741cebnef4c0fa6bd7d8978@mail.gmail.com>
Message-ID: <e9132f820603211330l52fd232aycf572f0b0a001c50@mail.gmail.com>

Could you please point to where the nightmare occurs?  Our -02 draft
addresses most of these issues and I think extensions have been
discussed for how to implement the remaining issues, and the changes
are really minor.  It works on top of a standard stack.  Plus, since
non-dht nodes don't need to see any of the new params or headers, it
doesn't create anything that those not implementing P2PSIP need to
worry about seeing or supporting.  I don't see the nightmare.

Bruce

On 3/21/06, Michael Slavitch <slavitch at gmail.com> wrote:
> To change SIP to support all those things is going to be a bloody nightmare.


From lowekamp at cs.wm.edu  Tue Mar 21 16:36:59 2006
From: lowekamp at cs.wm.edu (Bruce Lowekamp)
Date: Tue, 21 Mar 2006 15:36:59 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <010b01c64d2d$41c340a0$6400a8c0@comcast.net>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
	<B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
	<e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
	<e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
	<010b01c64d2d$41c340a0$6400a8c0@comcast.net>
Message-ID: <e9132f820603211336p78d72639g497ca0f6005712c7@mail.gmail.com>

The DHT is well-suited for storing URIs.  My thought for storing
arbitrary data (voicemail, configuration, etc) is to have the DHT
store the URI of a configuration file that is accessed via other
standard means (MSRP, etc).  That keeps the DHT as simple as possible.

Bruce

On 3/21/06, Peter Pan <huang-ming.pan at comcast.net> wrote:
> > The exact DHT algorithm we implement, the exact way the we use SIP,
> > the methods, etc I think are all going to need some careful thought.
> > But for P2P SIP, I think SIP is the obvious choice for DHT operations,
> > and I would consider it the default unless a convincing argument is
> > made against it.
>
> Not an argument but a question: how to efficiently implement Kademlia
> STORE_VALUE primitive in SIP INVITE/REGISTER when the stored values can be
> of any length and type?
>
> peter
>
> *** dont bash me, we are of the same country ***
>
> ----- Original Message -----
> From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
> To: "Philip Matthews" <philip_matthews at magma.ca>
> Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
> Sent: Tuesday, March 21, 2006 12:41 PM
> Subject: Re: [p2p-sip] Some comments on Use-cases document
>
>
> > resending since this hasn't been acked by the mailing list or appeared
> > after over 30 minutes.  (I'm sure it will now immediately appear)
> >
> > On 3/21/06, Bruce Lowekamp <lowekamp at cs.wm.edu> wrote:
> > > I don't thnk it has anything to do with use-cases, either
> > >
> > > but
> > >
> > > Building DHT maintenance on top of SIP gives us all of the advantages
> > > of the routing, addressing, naming, and security issues already built
> > > into SIP.  Plus the NAT traversal capabilities of STUN, TURN, and ICE.
> > >  While not perfect for our use, I think with very minimal
> > > modifications (whatever the final protocol is) they will provide a
> > > very good solution.
> > >
> > > A proposal to use something else needs to:
> > > - explain what the shortcomings of SIP are for this purpose
> > > - explain how a new or different solution will provide equivalent
> functionality
> > > - explain how/why the new protocol will be better after resolving the
> > > complexities that SIP has become so complex to address
> > > - make a convincing enough case to justify deploying devices (and
> > > we're frequently talking about very small devices) to implement two
> > > separate protocol stacks.
> > >
> > >
> > > The exact DHT algorithm we implement, the exact way the we use SIP,
> > > the methods, etc I think are all going to need some careful thought.
> > > But for P2P SIP, I think SIP is the obvious choice for DHT operations,
> > > and I would consider it the default unless a convincing argument is
> > > made against it.
> > >
> > > A number of comments have been made stating that NATs must be taken
> > > into account from the beginning.  Again, one of the issues that using
> > > SIP helps address already is NAT traversal.  It's not perfect for
> > > signalling, but the framework is there.
> > >
> > > Bruce
> > >
> > > On 3/21/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > > > On 20-Mar-06, at 21:19 , David Barrett wrote:
> > > >
> > > > > I'd add to that this key question:
> > > > >
> > > > >       "Will we extend SIP, or create a new protocol?"
> > > > >
> > > > > I'm finding it hard to make any headway without understanding the
> > > > > above.
> > > > > Basically, I see two major directions we could go:
> > > > >
> > > > > 1) Extend SIP with overlay-maintenance and resource-location
> messages.
> > > > >
> > > > > 2) Create a new overlay protocol and develop bindings for SIP and
> > > > > ICE (eg,
> > > > > distributed proxy service and STUN/TURN resource-location service).
> > > > >
> > > > > It seems this high-level decision keeps coming up again and again in
> > > > > discussions of the smaller issues.
> > > > >
> > > >
> > > > For what it is worth, I can say that my own views on this subject
> > > > have changed.
> > > >
> > > > For all of last year, I strongly believed that there should be two
> > > > distinct layers:
> > > > a SIP layer and a P2P layer (i.e., option 2). See the arguments I
> > > > wrote in
> > > >    http://www.p2psip.org/drafts/draft-matthews-sipping-p2p-industrial-
> > > > strength-00.txt
> > > > as well as those on Alan Johnston in
> > > >
> http://www.p2psip.org/drafts/draft-johnston-sipping-p2p-ipcom-01.txt
> > > >
> > > > However, in the last few months, I have come to see that there are
> > > > some good reasons
> > > > to put the two layers together into one (i.e., option 1):
> > > >         a) NAT Traversal becomes easier because there is just one
> port,
> > > > rather than two
> > > >            (this assumes that the P2P layer would run on a different
> port
> > > > than SIP)
> > > >         b) Possible performance improvements. With two layers, you
> have to
> > > > first ask
> > > >           "where is user U?" and then send the Invite message. With
> one
> > > > layer, there is
> > > >           the possibility of sending the Invite and having it routed
> to the
> > > > user.
> > > >           (David et al removed this from their latest draft, but this
> was an
> > > > option in the
> > > >           earlier version, and also in the work done by Henning's
> group).
> > > >
> > > > As others have pointed out, there are also drawbacks, so I haven't
> > > > concluded anything
> > > > yet, but I am a lot more open to option 1 than I was a few months ago.
> > > >
> > > >
> > > > - Philip
> > > >
> > > > PS. What this has to do with the use-cases document, however, I am
> > > > not clear on  ;-)
> > > > _______________________________________________
> > > > p2p-sip mailing list
> > > > p2p-sip at cs.columbia.edu
> > > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > > >
> > > >
> > >
> >
> >
> >
>
>


From huang-ming.pan at comcast.net  Tue Mar 21 16:56:56 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Tue, 21 Mar 2006 13:56:56 -0800
Subject: [p2p-sip] Some comments on Use-cases document
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
	<B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
	<e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
	<e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
	<010b01c64d2d$41c340a0$6400a8c0@comcast.net>
	<e9132f820603211336p78d72639g497ca0f6005712c7@mail.gmail.com>
Message-ID: <012801c64d32$5f6e5e50$6400a8c0@comcast.net>

Hi Bruce,

Thanks for sharing the thought of protocol mixing.

Unfortunately, whoever chooses Kademlia often must handle value retrieval at
the same time as key lookup. For example, FIND_VALUE implies FIND_NODE. It
is beyond my imagination to mix SIP and (eg.) MSRP in one single primitive
:-(

By the way, some overlay operations may be parallel and asynchronous. Though
much less than those in SS7, RFC 3261 defines some timers
(T1/.../A/B/C/.../K) in SIP protocol. Will P2P SIP regulate possible
asynchronous/timed activities? It seems essential for interoperability.

peter
----- Original Message -----
From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
To: "Peter Pan" <huang-ming.pan at comcast.net>
Cc: "Philip Matthews" <philip_matthews at magma.ca>; "P2P-SIP"
<p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 1:36 PM
Subject: Re: [p2p-sip] Some comments on Use-cases document


The DHT is well-suited for storing URIs.  My thought for storing
arbitrary data (voicemail, configuration, etc) is to have the DHT
store the URI of a configuration file that is accessed via other
standard means (MSRP, etc).  That keeps the DHT as simple as possible.

Bruce

On 3/21/06, Peter Pan <huang-ming.pan at comcast.net> wrote:
> > The exact DHT algorithm we implement, the exact way the we use SIP,
> > the methods, etc I think are all going to need some careful thought.
> > But for P2P SIP, I think SIP is the obvious choice for DHT operations,
> > and I would consider it the default unless a convincing argument is
> > made against it.
>
> Not an argument but a question: how to efficiently implement Kademlia
> STORE_VALUE primitive in SIP INVITE/REGISTER when the stored values can be
> of any length and type?
>
> peter
>
> *** dont bash me, we are of the same country ***
>
> ----- Original Message -----
> From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
> To: "Philip Matthews" <philip_matthews at magma.ca>
> Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
> Sent: Tuesday, March 21, 2006 12:41 PM
> Subject: Re: [p2p-sip] Some comments on Use-cases document
>
>
> > resending since this hasn't been acked by the mailing list or appeared
> > after over 30 minutes.  (I'm sure it will now immediately appear)
> >
> > On 3/21/06, Bruce Lowekamp <lowekamp at cs.wm.edu> wrote:
> > > I don't thnk it has anything to do with use-cases, either
> > >
> > > but
> > >
> > > Building DHT maintenance on top of SIP gives us all of the advantages
> > > of the routing, addressing, naming, and security issues already built
> > > into SIP.  Plus the NAT traversal capabilities of STUN, TURN, and ICE.
> > >  While not perfect for our use, I think with very minimal
> > > modifications (whatever the final protocol is) they will provide a
> > > very good solution.
> > >
> > > A proposal to use something else needs to:
> > > - explain what the shortcomings of SIP are for this purpose
> > > - explain how a new or different solution will provide equivalent
> functionality
> > > - explain how/why the new protocol will be better after resolving the
> > > complexities that SIP has become so complex to address
> > > - make a convincing enough case to justify deploying devices (and
> > > we're frequently talking about very small devices) to implement two
> > > separate protocol stacks.
> > >
> > >
> > > The exact DHT algorithm we implement, the exact way the we use SIP,
> > > the methods, etc I think are all going to need some careful thought.
> > > But for P2P SIP, I think SIP is the obvious choice for DHT operations,
> > > and I would consider it the default unless a convincing argument is
> > > made against it.
> > >
> > > A number of comments have been made stating that NATs must be taken
> > > into account from the beginning.  Again, one of the issues that using
> > > SIP helps address already is NAT traversal.  It's not perfect for
> > > signalling, but the framework is there.
> > >
> > > Bruce
> > >
> > > On 3/21/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > > > On 20-Mar-06, at 21:19 , David Barrett wrote:
> > > >
> > > > > I'd add to that this key question:
> > > > >
> > > > >       "Will we extend SIP, or create a new protocol?"
> > > > >
> > > > > I'm finding it hard to make any headway without understanding the
> > > > > above.
> > > > > Basically, I see two major directions we could go:
> > > > >
> > > > > 1) Extend SIP with overlay-maintenance and resource-location
> messages.
> > > > >
> > > > > 2) Create a new overlay protocol and develop bindings for SIP and
> > > > > ICE (eg,
> > > > > distributed proxy service and STUN/TURN resource-location
service).
> > > > >
> > > > > It seems this high-level decision keeps coming up again and again
in
> > > > > discussions of the smaller issues.
> > > > >
> > > >
> > > > For what it is worth, I can say that my own views on this subject
> > > > have changed.
> > > >
> > > > For all of last year, I strongly believed that there should be two
> > > > distinct layers:
> > > > a SIP layer and a P2P layer (i.e., option 2). See the arguments I
> > > > wrote in
> > > >
http://www.p2psip.org/drafts/draft-matthews-sipping-p2p-industrial-
> > > > strength-00.txt
> > > > as well as those on Alan Johnston in
> > > >
> http://www.p2psip.org/drafts/draft-johnston-sipping-p2p-ipcom-01.txt
> > > >
> > > > However, in the last few months, I have come to see that there are
> > > > some good reasons
> > > > to put the two layers together into one (i.e., option 1):
> > > >         a) NAT Traversal becomes easier because there is just one
> port,
> > > > rather than two
> > > >            (this assumes that the P2P layer would run on a different
> port
> > > > than SIP)
> > > >         b) Possible performance improvements. With two layers, you
> have to
> > > > first ask
> > > >           "where is user U?" and then send the Invite message. With
> one
> > > > layer, there is
> > > >           the possibility of sending the Invite and having it routed
> to the
> > > > user.
> > > >           (David et al removed this from their latest draft, but
this
> was an
> > > > option in the
> > > >           earlier version, and also in the work done by Henning's
> group).
> > > >
> > > > As others have pointed out, there are also drawbacks, so I haven't
> > > > concluded anything
> > > > yet, but I am a lot more open to option 1 than I was a few months
ago.
> > > >
> > > >
> > > > - Philip
> > > >
> > > > PS. What this has to do with the use-cases document, however, I am
> > > > not clear on  ;-)
> > > > _______________________________________________
> > > > p2p-sip mailing list
> > > > p2p-sip at cs.columbia.edu
> > > > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> > > >
> > > >
> > >
> >
> >
> >
>
>


From sathya at research.panasonic.com  Tue Mar 21 17:16:45 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Tue, 21 Mar 2006 17:16:45 -0500
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <88fb3612.361288fb@pintlmail.MITL.Research.Panasonic.COM>


Hi Bruce -

Let me re-iterate what I said earlier - I am not against the idea of using SIP for the overlay layer YET. But ...

> A proposal to use something else needs to:
> - explain what the shortcomings of SIP are for this purpose
> - explain how a new or different solution will provide equivalent 
> functionality- explain how/why the new protocol will be better after resolving the
> complexities that SIP has become so complex to address
> - make a convincing enough case to justify deploying devices (and we're frequently talking about very small devices) to implement two
> separate protocol stacks.

I don't agree with the above statement. It is  not clear to me why it is the burden of somebody proposing the design non-SIP protocol to justify it. In my mind, its as much the responsibility of somebody proposing the use of SIP to justify such an use. Even more so, because there is no disagreement on the need for a overlay protocol,  the disagreement is whether SIP fits the bill SO well that it can be used with minimal changes. Ofcourse, draft-bryan is a starting point for such an idea, but obviously it leaves many open issues which worries me that we may regret the choice later. With a new protocol we have better freedom to make any necessary design choices without having worry about affecting SIP.

thanks,
Sathya



> 
> On 3/21/06, Philip Matthews <philip_matthews at magma.ca> wrote:
> > On 20-Mar-06, at 21:19 , David Barrett wrote:
> >
> > > I'd add to that this key question:
> > >
> > >       "Will we extend SIP, or create a new protocol?"
> > >
> > > I'm finding it hard to make any headway without understanding the
> > > above.
> > > Basically, I see two major directions we could go:
> > >
> > > 1) Extend SIP with overlay-maintenance and resource-location 
> messages.> >
> > > 2) Create a new overlay protocol and develop bindings for SIP and
> > > ICE (eg,
> > > distributed proxy service and STUN/TURN resource-location 
> service).> >
> > > It seems this high-level decision keeps coming up again and 
> again in
> > > discussions of the smaller issues.
> > >
> >
> > For what it is worth, I can say that my own views on this subject
> > have changed.
> >
> > For all of last year, I strongly believed that there should be two
> > distinct layers:
> > a SIP layer and a P2P layer (i.e., option 2). See the arguments I
> > wrote in
> >    http://www.p2psip.org/drafts/draft-matthews-sipping-p2p-
> industrial-
> > strength-00.txt
> > as well as those on Alan Johnston in
> >    http://www.p2psip.org/drafts/draft-johnston-sipping-p2p-ipcom-
> 01.txt>
> > However, in the last few months, I have come to see that there are
> > some good reasons
> > to put the two layers together into one (i.e., option 1):
> >         a) NAT Traversal becomes easier because there is just 
> one port,
> > rather than two
> >            (this assumes that the P2P layer would run on a 
> different port
> > than SIP)
> >         b) Possible performance improvements. With two layers, 
> you have to
> > first ask
> >           "where is user U?" and then send the Invite message. 
> With one
> > layer, there is
> >           the possibility of sending the Invite and having it 
> routed to the
> > user.
> >           (David et al removed this from their latest draft, but 
> this was an
> > option in the
> >           earlier version, and also in the work done by 
> Henning's group).
> >
> > As others have pointed out, there are also drawbacks, so I haven't
> > concluded anything
> > yet, but I am a lot more open to option 1 than I was a few 
> months ago.
> >
> >
> > - Philip
> >
> > PS. What this has to do with the use-cases document, however, I am
> > not clear on  ;-)
> > _______________________________________________
> > p2p-sip mailing list
> > p2p-sip at cs.columbia.edu
> > https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >
> >
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From sathya at research.panasonic.com  Tue Mar 21 17:19:23 2006
From: sathya at research.panasonic.com (Sathya Narayanan)
Date: Tue, 21 Mar 2006 17:19:23 -0500
Subject: [p2p-sip] Single layer vs two layers
Message-ID: <53217598.75985321@pintlmail.MITL.Research.Panasonic.COM>



----- Original Message -----
From: Bruce Lowekamp <lowekamp at cs.wm.edu>
Date: Tuesday, March 21, 2006 3:37 pm
Subject: Re: [p2p-sip] Single layer vs two layers

> Our -02 draft suggests the idea of making a UA responsible for storing
> replicas by modifying the resource URI and registering each replica
> separately.  There are pros and cons to this, but it does avoid having
> the responsible node also be responsible for maintaining replicas just
> because it received a REGISTER. 

This is exactly my point. The design choice of who must be responsible for replicas is being forced on us by the choice to use SIP. If we have a new protocol design, that decision will be based on what is the best (in terms of reliability and efficiency) for the overlay operations.

thanks,
Sathya 

> I haven't contemplated yet whether we
> can simply use sip-outbound techniques to accomplish replicas, but
> that might be an interesting question.
> 
> Bruce
> 
> On 3/21/06, Sathya Narayanan <sathya at research.panasonic.com> wrote:
> > Even assuming that the semantics of REGISTER message matches the 
> required semantics for node join and resource add of a DHT, how 
> are we going to address replication of resources on DHT. We have 
> to extend the way REGISTER message is used to support whatever 
> form of replication is developed, while AFAIK replication is not 
> part of the REGISTER semantics today. And I am sure there will be 
> other requirements to have a stable overlay network that will need 
> further changes.
> 


From dean.willis at softarmor.com  Tue Mar 21 18:00:01 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 21 Mar 2006 17:00:01 -0600
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <009901c64d15$7dd2b250$6400a8c0@comcast.net>
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<009901c64d15$7dd2b250$6400a8c0@comcast.net>
Message-ID: <A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>


On Mar 21, 2006, at 12:30 PM, Peter Pan wrote:

>> I'm sorry, but if it doesn't address NAT we might as well go home.
>
> I believe any incremental p2p-sip approach a disaster, too. Since all
> technologies involved in p2p-sip have been there integrally proven  
> for years
> (by Skype) and most of them seem ought to be tightly coupled for  
> feasibility
> and interoperability, I believe a "do it once for all and forever"  
> is the
> right approach for the late comers (-: us :-) to attract world  
> attention and
> deroot all weeds in the future.
>
> I wish I may add one more sentence: "I'm sorry, but if it doesn't  
> address
> SECURITY, TOP-LEVEL FEATURES, SPECIFIC OVERLAY(S)... among others  
> we might
> as well go home. The whole point of the overlay is to get around  
> structures
> such as NAT ... have STRONG SECURITY beneath it, have TOP-LEVEL  
> FEATURES
> (voicemail, centrex, ...) above it, and have INTEROPERABILITY  
> surrounding
> it."

one caveat here. Skype does a lot of its work by doing things like  
tunneling over port 443. The IETF is NOT going to accept any RFCs  
that include this sort of stuff. It's just not going to happen. It's  
a religion thing. Don't waste your time suggesting it, and if this is  
a REQUIREMENT for getting your solution to work, you'd be best served  
by taking it someplace else.

--
Dean


From dean.willis at softarmor.com  Tue Mar 21 18:02:51 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 21 Mar 2006 17:02:51 -0600
Subject: [p2p-sip] Single layer vs two layers
In-Reply-To: <e9132f820603211210v51a29c9ai157003571270ae50@mail.gmail.com>
References: <69b45c18.5c1869b4@pintlmail.MITL.Research.Panasonic.COM>
	<e9132f820603211210v51a29c9ai157003571270ae50@mail.gmail.com>
Message-ID: <46BAD6E5-CCE4-41FF-87A8-9648F171CBA5@softarmor.com>


On Mar 21, 2006, at 2:10 PM, Bruce Lowekamp wrote:
>
> -01 used INVITE to search the DHT for the contact and to place the
> call.  There was some concern that this might introduce issues when
> connecting to non-P2P UAs with p2p headers (even if they are just
> options or extension headers).

Despite our conversation on the topic last night, I fail to see how  
delivering a properly formatted SIP INVITE to a node (P2P or  
otherwise) is a problem. If it's a terminal node, it'll ring and  
maybe answer. If not, it'll try to proxy-route based on the request- 
URI, which had best be both syntactically valid and RFC 3263  
meaningful, and it might end up on a P2P UA eventually. Or a  
conventional UA. That's a feature, not a bug.

--
Dean


From dean.willis at softarmor.com  Tue Mar 21 18:05:05 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 21 Mar 2006 17:05:05 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <010b01c64d2d$41c340a0$6400a8c0@comcast.net>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu><B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca><e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
	<e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
	<010b01c64d2d$41c340a0$6400a8c0@comcast.net>
Message-ID: <30338F63-9BF7-4327-97CB-93D49E389F3E@softarmor.com>



On Mar 21, 2006, at 3:20 PM, Peter Pan wrote:

>> The exact DHT algorithm we implement, the exact way the we use SIP,
>> the methods, etc I think are all going to need some careful thought.
>> But for P2P SIP, I think SIP is the obvious choice for DHT  
>> operations,
>> and I would consider it the default unless a convincing argument is
>> made against it.
>
> Not an argument but a question: how to efficiently implement Kademlia
> STORE_VALUE primitive in SIP INVITE/REGISTER when the stored values  
> can be
> of any length and type?


By bounding the P2PSIP store-value primitive as a vector of  
distinguished name and contact URI.

It's not a question of what Kademlia can express, it's a question of  
what P2PSIP requires.

--
Dean

From dean.willis at softarmor.com  Tue Mar 21 18:08:43 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Tue, 21 Mar 2006 17:08:43 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <1fc32c8f0603211323l2e741cebnef4c0fa6bd7d8978@mail.gmail.com>
References: <200603210319.k2L3JIrj001729@cs.columbia.edu>
	<B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca>
	<e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com>
	<e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com>
	<1fc32c8f0603211323l2e741cebnef4c0fa6bd7d8978@mail.gmail.com>
Message-ID: <789D1CFA-10D4-4DF4-9DB1-E8628745D98B@softarmor.com>


On Mar 21, 2006, at 3:23 PM, Michael Slavitch wrote:

> I believe that SIP is now such a monster that any attempt to P2P  
> the SIP protocol itself is going to be a total implementation  
> mess.   A separate overlay that links nicely with SIP but requires  
> no changes to SIP/RTP/SDP apart from understanding different naming  
> syntaxes (ie, something not DNS or numbers) is a lot more clean  
> with a lot less screaming from other working groups that are  
> obsessed with IMS and the CSS7.
>


I don't necessarily disagree, but this after all the P2PSIP BOF we're  
talking about. If we need a P2PVOIP BOF that defines an entirely new  
protocol, draft up a charter and get some sponsoring ADs and see if  
you can get the BOF approved. If it looks like it has traction, I'll  
show up and help out.

Or maybe an OpenSkypeLike open source effort would be better. Way  
cool, but way outside the IETF's scope.

--
Dean

From huang-ming.pan at comcast.net  Tue Mar 21 18:15:26 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Tue, 21 Mar 2006 15:15:26 -0800
Subject: [p2p-sip] NATs and P2P
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<009901c64d15$7dd2b250$6400a8c0@comcast.net>
	<A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>
Message-ID: <014c01c64d3d$5a7b9470$6400a8c0@comcast.net>

hehe... i smell some unfriendliness. i propose no solution or proprietary
stuff. (i wish i had)

what a mess up!!

peter
----- Original Message -----
From: "Dean Willis" <dean.willis at softarmor.com>
To: "Peter Pan" <huang-ming.pan at comcast.net>
Cc: "Michael Slavitch" <slavitch at objectworld.com>; "Brijesh Kumar"
<kumarb at research.panasonic.com>; "Bruce Lowekamp" <lowekamp at cs.wm.edu>;
"Philip Matthews" <philip_matthews at magma.ca>; "P2P-SIP"
<p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 3:00 PM
Subject: Re: [p2p-sip] NATs and P2P


>
> On Mar 21, 2006, at 12:30 PM, Peter Pan wrote:
>
> >> I'm sorry, but if it doesn't address NAT we might as well go home.
> >
> > I believe any incremental p2p-sip approach a disaster, too. Since all
> > technologies involved in p2p-sip have been there integrally proven
> > for years
> > (by Skype) and most of them seem ought to be tightly coupled for
> > feasibility
> > and interoperability, I believe a "do it once for all and forever"
> > is the
> > right approach for the late comers (-: us :-) to attract world
> > attention and
> > deroot all weeds in the future.
> >
> > I wish I may add one more sentence: "I'm sorry, but if it doesn't
> > address
> > SECURITY, TOP-LEVEL FEATURES, SPECIFIC OVERLAY(S)... among others
> > we might
> > as well go home. The whole point of the overlay is to get around
> > structures
> > such as NAT ... have STRONG SECURITY beneath it, have TOP-LEVEL
> > FEATURES
> > (voicemail, centrex, ...) above it, and have INTEROPERABILITY
> > surrounding
> > it."
>
> one caveat here. Skype does a lot of its work by doing things like
> tunneling over port 443. The IETF is NOT going to accept any RFCs
> that include this sort of stuff. It's just not going to happen. It's
> a religion thing. Don't waste your time suggesting it, and if this is
> a REQUIREMENT for getting your solution to work, you'd be best served
> by taking it someplace else.
>
> --
> Dean
>


From huang-ming.pan at comcast.net  Tue Mar 21 18:50:30 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Tue, 21 Mar 2006 15:50:30 -0800
Subject: [p2p-sip] Some comments on Use-cases document
References: <200603210319.k2L3JIrj001729@cs.columbia.edu><B225AB16-9E9C-4A5F-A3C1-97309B78A920@magma.ca><e9132f820603211202k3024d6b1x548400900a75a742@mail.gmail.com><e9132f820603211241m7a43eab4q3f4f3cf3530052c3@mail.gmail.com><1fc32c8f0603211323l2e741cebnef4c0fa6bd7d8978@mail.gmail.com>
	<789D1CFA-10D4-4DF4-9DB1-E8628745D98B@softarmor.com>
Message-ID: <016a01c64d42$3d3fa7c0$6400a8c0@comcast.net>

pls dont give up so soon.

with original initiative in mind of and topdown+bottomup+macro+micro
viewpoints of smart people here from academical and industrial fields, ietf
can stop in time similar funny (foul) plays by some branded softswitch
vendors who approached class5 switches by heavy-weight b2bua utilizing only
basic SIP call control primitives blended with brute forces (some even relay
media between two ua's behind full-cone nat!!) when ietf has taken years of
efforts on standardizing STUN, REFER, "SIP Service Examples", ...etc.

open source (to attackers) is a no go, as i have questioned (but none
answered) earlier.

peter
----- Original Message -----
From: "Dean Willis" <dean.willis at softarmor.com>
To: "Michael Slavitch" <slavitch at gmail.com>
Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
Sent: Tuesday, March 21, 2006 3:08 PM
Subject: Re: [p2p-sip] Some comments on Use-cases document


>
> On Mar 21, 2006, at 3:23 PM, Michael Slavitch wrote:
>
> > I believe that SIP is now such a monster that any attempt to P2P
> > the SIP protocol itself is going to be a total implementation
> > mess.   A separate overlay that links nicely with SIP but requires
> > no changes to SIP/RTP/SDP apart from understanding different naming
> > syntaxes (ie, something not DNS or numbers) is a lot more clean
> > with a lot less screaming from other working groups that are
> > obsessed with IMS and the CSS7.
> >
>
>
> I don't necessarily disagree, but this after all the P2PSIP BOF we're
> talking about. If we need a P2PVOIP BOF that defines an entirely new
> protocol, draft up a charter and get some sponsoring ADs and see if
> you can get the BOF approved. If it looks like it has traction, I'll
> show up and help out.
>
> Or maybe an OpenSkypeLike open source effort would be better. Way
> cool, but way outside the IETF's scope.
>
> --
> Dean
>
>


From kyara543 at yahoo.com  Tue Mar 21 21:08:20 2006
From: kyara543 at yahoo.com (Kyara Yamamoto)
Date: Tue, 21 Mar 2006 18:08:20 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <01f801c64d50$5775d3e0$6400a8c0@comcast.net>
Message-ID: <20060322020820.29502.qmail@web38608.mail.mud.yahoo.com>

Peter:

>From http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html Chord is official overlay of P2P SIP. 

Forget about KAD and your solution here, as long as Bush insists staying in Iraq.

--Yamamoto
> On Mar 21, 2006, at 3:20 PM, Peter Pan wrote:
> 
> >> The exact DHT algorithm we implement, the exact way the we use SIP,
> >> the methods, etc I think are all going to need some careful thought.
> >> But for P2P SIP, I think SIP is the obvious choice for DHT  
> >> operations,
> >> and I would consider it the default unless a convincing argument is
> >> made against it.
> >
> > Not an argument but a question: how to efficiently implement Kademlia
> > STORE_VALUE primitive in SIP INVITE/REGISTER when the stored values  
> > can be
> > of any length and type?
> 
> 
> By bounding the P2PSIP store-value primitive as a vector of  
> distinguished name and contact URI.
> 
> It's not a question of what Kademlia can express, it's a question of  
> what P2PSIP requires.
> 
> --
> Dean



__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060321/ca483d44/attachment.html

From karstbj at yahoo.com  Wed Mar 22 01:43:29 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Tue, 21 Mar 2006 22:43:29 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060322020820.29502.qmail@web38608.mail.mud.yahoo.com>
Message-ID: <20060322064329.24078.qmail@web53012.mail.yahoo.com>



--- Kyara Yamamoto <kyara543 at yahoo.com> wrote:

> Peter:
> 
> >From
>
http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html
> Chord is official overlay of P2P SIP. 
> 
> Forget about KAD and your solution here, as long as
> Bush insists staying in Iraq.

A DHT is essentially a set of nodes that conspire to
resolve names. There are multiple ways to build such
conspiracies -- Kamdelia, Chord, Microsoft's PNRP,
even dynamic DNS. We can be pretty usre that some of
the existing ways will be proven broken in the future,
and that new algorithms will be found. Linking P2P SIP
to is specific conspiracy does not appear very wise.
Can't we build a level of indirection? For example, is
it possible to somehow identify the specific DHT
inside the P2P SIP URL?

-- Karst

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From jh at tutpro.com  Wed Mar 22 03:43:40 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Wed, 22 Mar 2006 10:43:40 +0200
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<009901c64d15$7dd2b250$6400a8c0@comcast.net>
	<A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>
Message-ID: <17441.3644.379784.440053@rautu.tutpro.com>

Dean Willis writes:

 > Skype does a lot of its work by doing things like  
 > tunneling over port 443. The IETF is NOT going to accept any RFCs  
 > that include this sort of stuff. It's just not going to happen. 
 > It's a religion thing.

how about what i suggested:  map SIP into XML-RPC rather than tunnel
over?  what kind of IETF religion would prevent that?

-- juha

From dbarrett at quinthar.com  Wed Mar 22 04:19:09 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Wed, 22 Mar 2006 01:19:09 -0800
Subject: [p2p-sip] Questions for the BOF
Message-ID: <200603220919.k2M9JJcK022034@cs.columbia.edu>

Unfortunately I can't be at the meeting today.  But some issues I think
would be good to decide:

1) Will we extend SIP with overlay functionality, or develop an overlay with
SIP bindings?

Both seem very compelling, and have the hallmarks of the classic build/buy
or rewrite/refactor decision:

The idea of refactoring SIP to naturally support P2P sounds great in theory
-- and I'd wager the total changes at the end would be very small and neat
-- but requires detailed stitching into hundreds of pages of RFCs, and
careful analysis of the consequences.

The idea of just building a new P2P overlay layer and specifying bindings
sounds like more work in theory, but that work in practice will probably be
much easier as the problem scope is more easily defined with a much cleaner
interface.

At the end of the day, my idealistic side says the former, but my pragmatic
side leans toward the latter.


2) Will we commit to a single type of overlay network today, or generalize
on the assumption we'll 'swap in' different overlays over time?

Personally, I think it'll be hard enough to just pick one and be done with
it.  Trying to generalize the problem seems unnecessarily hard.  We have a
round peg, and we're debating whether to make a round hole, or some more
general hole that can fit different shapes of pegs that haven't yet been
designed.  It can be done, but the result is you have a hole that doesn't
neatly fit any peg, and certainly not the one we have today.


3) To what degree will we tackle security considerations in the first rev?

One challenge is that security in a DHT setting is still an open topic for
research.  The IETF typically doesn't do research.  Thus I lean toward not
standardizing a secure solution, because (so far as I know) that secure
solution doesn't yet exist (though I'd be happy to be wrong).  Regardless,
at the very least we should detail precisely what the security weaknesses
are, so we can best focus the research community on how to solve them.  Not
great, but there you have it.


Those are the top issues in my mind.  Have fun without me, and I'm eager to
read the results!

-david


From stiemerling at netlab.nec.de  Wed Mar 22 07:30:24 2006
From: stiemerling at netlab.nec.de (Martin Stiemerling)
Date: Wed, 22 Mar 2006 13:30:24 +0100
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <017d01c64cfa$9f978480$01000001@Brijesh2K>
References: <9151EA5D-BA55-4A69-8F6A-7118492C4283@magma.ca>
	<e9132f820603201758n727d6d25p62f4c4d60dac42d9@mail.gmail.com>
	<017d01c64cfa$9f978480$01000001@Brijesh2K>
Message-ID: <FC97466E-71C1-4157-BE63-6410B9347A6D@netlab.nec.de>


Am 21.03.2006 um 16:17 schrieb Brijesh Kumar:

> I think Bruce's approach makes sense.
>
> The first task should be to come out with a  base specification
> that works. If nothing else, this can be deployed in a hypothetical
> pure IPv6 world, or  in a  limited enterprise universe where
> traversing NAT may not be an issue.

Well, I strongly disagree here! That is the usual mistake people
do when design protocols. They do not care about NATs right
from the beginning and need to fix the protocol spec afterwards.
This fixing is at least factor two of effort compared to thinking
about it right from the beginning.

>
> The second task should be to modify the spec  that works with
> NAT.  Trying to resolve NAT issue from day one is good idea, but
> one can equally treat is as an extension of the core design.  It is  
> just
> a matter of convenience rather than  any hard logic.

Ask all the SIP community what a pain this can be.

   Martin

>
> Cheers,
>
> --bk
>
> ----- Original Message -----
> From: "Bruce Lowekamp" <lowekamp at cs.wm.edu>
> To: "Philip Matthews" <philip_matthews at magma.ca>
> Cc: "P2P-SIP" <p2p-sip at cs.columbia.edu>
> Sent: Monday, March 20, 2006 8:58 PM
> Subject: Re: [p2p-sip] NATs and P2P
>
>
>> Philip,
>>
>> I absolutely agree that we need to have a solution for this type of
>> environment.   IMHO, some of the worries about NATs causing problems
>> have been somewhat overstated.  As you point out, the majority of  
>> NATs
>> in the wild are address independent mapping, and it is relatively
>> trivial to punch holes to establish communications.  But you're also
>> right that, especially for corporate networks, this doesn't solve the
>> entire problem.
>>
>> I spent some time thinking about how to solve the specific problems
>> you pose in terms of the protocol we have been playing with in
>> draft-bryan-sipping-p2p.  We completely ignore the issue of NATs in
>> there (not because we don't think it's a challenge, but just because
>> we want to work on a basic protocol first).  If I was to address some
>> of the problems you pose using that framework (modified chord), then
>> one approach would be to remove the requirement that DHT  
>> operations be
>> iterative and allow them to be recursive.  If DHT operations can be
>> recursive, then the number of connections that a node must maintain
>> through the firewall is limited by the size of the finger table.   
>> That
>> number doesn't have to be very big to have a rather efficient DHT.
>> (never more than 128, and I think in practice the number can be much
>> smaller.  If your guess of 160 is correct, we're more than OK here.)
>>
>> I'm a bit unhappy with that because I really like the iterative
>> approach for a variety of reasons, but certainly the protocol  
>> could be
>> modified that way and a hybrid environment could be established where
>> some nodes are open (or easy to request opens) and support iterative
>> queries with anyone, but others are recursive queries only.  One
>> problem is that the burden of having a finger-table entry to a
>> recursive-only host is placed on the non-recursive node that would
>> otherwise redirect to its finger-table entry, but now must recurse  
>> the
>> request.
>>
>> Right now the scheme would also have problems because, as Eric  
>> rightly
>> points out in a separate thread on this mailing list, we don't have a
>> good way of selecting Identifiers for nodes behind NATs (in
>> particular).  But if you have node certificates or another way of
>> authenticating your nodes, this is less of a problem.
>>
>> I want to make it clear that I'm not ready to propose a specific
>> solution, but I think there are ways to support structured DHT
>> algorithms that are not really very painful and work reasonably well
>> inside these environments.  Clearly support for such an environment
>> needs to be a listed in eventual requirements docs.
>>
>> Bruce
>>
>>
>> On 3/20/06, Philip Matthews <philip_matthews at magma.ca> wrote:
>>> [I guess all this traffic on the mailing list will teach me not  
>>> to go
>>> on vacation
>>> just before an IETF meeting!]
>>>
>>> As I catch up on all the messages about NATs over the past two  
>>> weeks,
>>> it seems to me that many people are thinking only of the "Public P2P
>>> Service Provider"
>>> use-case as described in the use-cases document. In other words,  
>>> a Skype
>>> competitor.
>>>
>>> However, the use-cases document that David Bryan and his co-authors
>>> wrote
>>> identified a number of other use-cases and it seems to me that these
>>> have somewhat
>>> different NAT traversal requirements.
>>>
>>> In particular, in some of these other use-cases, it seems to me that
>>> we CANNOT assume
>>> there are peers with public IP addresses.
>>>
>>> For example, consider the "Presence using Multimedia Consumer
>>> Electronics Devices"
>>> use-case (section 3.1.3) -- essentially a P2P network of multimedia
>>> consumer electronics devices
>>> that need presence information. Who is going to pay the extra money
>>> to give their digital camera (or those
>>> neat 770 tablets that Nokia is demoing here in Dallas?) a public  
>>> IPv4
>>> address?? On the contrary,
>>> devices like this are almost certainly going to have private IP
>>> addresses -- it is very common today for
>>> wireless internet providers to place a big NAT in front of their
>>> entire network and give private addresses
>>> to all their customers.
>>>
>>> Or consider the "IP PBX" use-case -- a IP PBX system for a company
>>> with a number of small branches
>>> scattered throughout the world. Each branch is going to have a  
>>> NAT in
>>> front of its network, and all
>>> the phones in that branch are going to have private IP addresses.
>>> None of the phones are going to
>>> have public IP addresses.
>>>
>>> It is handling the NAT traversal issues for use-cases like these  
>>> that
>>> Eric Cooper and I wrote our
>>> internet-draft on NAT Traversal for P2P:
>>>        http://www.ietf.org/internet-drafts/draft-matthews-p2psip- 
>>> nats-
>>> and-overlays-00.txt
>>>
>>> - Philip
>>>
>>> _______________________________________________
>>> p2p-sip mailing list
>>> p2p-sip at cs.columbia.edu
>>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>>
>>>
>>
>> _______________________________________________
>> p2p-sip mailing list
>> p2p-sip at cs.columbia.edu
>> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>>
>
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


From eunsoo at research.panasonic.com  Wed Mar 22 07:37:06 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Wed, 22 Mar 2006 07:37:06 -0500
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060322020820.29502.qmail@web38608.mail.mud.yahoo.com>
References: <20060322020820.29502.qmail@web38608.mail.mud.yahoo.com>
Message-ID: <442144F2.30204@research.panasonic.com>

No decision has been made as far as I know.
I think we need to select a DHT algorithm (the first one we will use)
sooner or later.
What have been discussed (and roughly agreed, I believe) was that we
should not spend too much effort to compare all kinds of DHTs to pick
the very best one. A number of people believe that DHT algorithms have
pros and cons and thus it would be hard to say which is the best.
Probably we could have a brief comparison among well-known DHT
algorithms and pick one as our starting point. We may find a better one
later and thus the architecture and protocol should be modular enough to
allow easy change of the DHT algorithm.

Certainly Chord is one of the primary candidates. But I understand there
are some other proposals, too.
Thanks.

Eunsoo

Kyara Yamamoto wrote:

> Peter:
>
> From http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html
> Chord is official overlay of P2P SIP.
>
> Forget about KAD and your solution here, as long as Bush insists
> staying in Iraq.
>
> --Yamamoto
>
>     > On Mar 21, 2006, at 3:20 PM, Peter Pan wrote:
>     >
>     > >> The exact DHT algorithm we implement, the exact way the we
>     use SIP,
>     > >> the methods, etc I think are all going to need some careful
>     thought.
>     > >> But for P2P SIP, I think SIP is the obvious choice for DHT
>     > >> operations,
>     > >> and I would consider it the default unless a convincing
>     argument is
>     > >> made against it.
>     > >
>     > > Not an argument but a question: how to efficiently implement
>     Kademlia
>     > > STORE_VALUE primitive in SIP INVITE/REGISTER when the stored
>     values
>     > > can be
>     > > of any length and type?
>     >
>     >
>     > By bounding the P2PSIP store-value primitive as a vector of
>     > distinguished name and contact URI.
>     >
>     > It's not a question of what Kademlia can express, it's a
>     question of
>     > what P2PSIP requires.
>     >
>     > --
>     > Dean
>
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam? Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From eunsoo at research.panasonic.com  Wed Mar 22 08:02:13 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Wed, 22 Mar 2006 08:02:13 -0500
Subject: [p2p-sip] Questions for the BOF
In-Reply-To: <200603220919.k2M9JJcK022034@cs.columbia.edu>
References: <200603220919.k2M9JJcK022034@cs.columbia.edu>
Message-ID: <44214AD5.3000804@research.panasonic.com>

David Barrett wrote:

>Unfortunately I can't be at the meeting today.  But some issues I think
>would be good to decide:
>
>1) Will we extend SIP with overlay functionality, or develop an overlay with
>SIP bindings?
>
>Both seem very compelling, and have the hallmarks of the classic build/buy
>or rewrite/refactor decision:
>
>The idea of refactoring SIP to naturally support P2P sounds great in theory
>-- and I'd wager the total changes at the end would be very small and neat
>-- but requires detailed stitching into hundreds of pages of RFCs, and
>careful analysis of the consequences.
>
>The idea of just building a new P2P overlay layer and specifying bindings
>sounds like more work in theory, but that work in practice will probably be
>much easier as the problem scope is more easily defined with a much cleaner
>interface.
>
>At the end of the day, my idealistic side says the former, but my pragmatic
>side leans toward the latter.
>
>  
>
This is a good and important question.
But I think we can deal with this after the BOF.
BOF should be focused on the charter and deeper design decisions should
be discussed later.
The ad-hoc meeting on Friday will be a good chance to discuss about this
together.

>2) Will we commit to a single type of overlay network today, or generalize
>on the assumption we'll 'swap in' different overlays over time?
>
>Personally, I think it'll be hard enough to just pick one and be done with
>it.  Trying to generalize the problem seems unnecessarily hard.  We have a
>round peg, and we're debating whether to make a round hole, or some more
>general hole that can fit different shapes of pegs that haven't yet been
>designed.  It can be done, but the result is you have a hole that doesn't
>neatly fit any peg, and certainly not the one we have today.
>
>
>  
>
As I wrote in another email, I think we should pick one as the starting
point and be open to change into or add different overlay algorithms
over time.
I hope the discussion in the BOF on this topic should be limited to the
extent affecting the charter and deeper discussions on the implication
to the architecture and so on are left for the ad-hoc meeting.

>3) To what degree will we tackle security considerations in the first rev?
>
>One challenge is that security in a DHT setting is still an open topic for
>research.  The IETF typically doesn't do research.  Thus I lean toward not
>standardizing a secure solution, because (so far as I know) that secure
>solution doesn't yet exist (though I'd be happy to be wrong).  Regardless,
>at the very least we should detail precisely what the security weaknesses
>are, so we can best focus the research community on how to solve them.  Not
>great, but there you have it.
>
>
>  
>
I think at least we need to take a look at what the research community
has done.
Even though I understand your point but I am not so sure whether we
should preclude security measures as a whole from the first rev.
Couldn't we take a middle ground, i.e., adopting known measures while
explicitly stating known and unrsolved security issues?

>Those are the top issues in my mind.  Have fun without me, and I'm eager to
>read the results!
>
>-david
>
>  
>
Could you come to the ad-hoc meeting?

Thanks.

Eunsoo

>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From mike_robnson at yahoo.com  Wed Mar 22 10:35:44 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Wed, 22 Mar 2006 07:35:44 -0800 (PST)
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <17441.3644.379784.440053@rautu.tutpro.com>
Message-ID: <20060322153544.94993.qmail@web38510.mail.mud.yahoo.com>

That sounds interesting idea to me.

-- mike.
-------------------------
Mike Robinson
VoIP Trainer and Consultant

Juha Heinanen <jh at tutpro.com> wrote: Dean Willis writes:

 > Skype does a lot of its work by doing things like  
 > tunneling over port 443. The IETF is NOT going to accept any RFCs  
 > that include this sort of stuff. It's just not going to happen. 
 > It's a religion thing.

how about what i suggested:  map SIP into XML-RPC rather than tunnel
over?  what kind of IETF religion would prevent that?

-- juha
_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


		
---------------------------------
 Yahoo! Mail
 Use Photomail to share photos without annoying attachments.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060322/8c8b6202/attachment.html

From mike_robnson at yahoo.com  Wed Mar 22 11:16:04 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Wed, 22 Mar 2006 08:16:04 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060322064329.24078.qmail@web53012.mail.yahoo.com>
Message-ID: <20060322161604.52512.qmail@web38502.mail.mud.yahoo.com>

Karst,

Could you please elaborate on why do you say " Linking P2P SIP
to a particular DHT is specific conspiracy does not appear very wise.".
What is wrong with Chord that is being suggested?

--mike
-------------------
Mike Robinson
VoIP Trainer and Consultant



Karst Bjorgson <karstbj at yahoo.com> wrote: 

--- Kyara Yamamoto  wrote:

> Peter:
> 
> >From
>
http://www.p2psip.org/drafts/draft-bryan-sipping-p2p-02.html
> Chord is official overlay of P2P SIP. 
> 
> Forget about KAD and your solution here, as long as
> Bush insists staying in Iraq.

A DHT is essentially a set of nodes that conspire to
resolve names. There are multiple ways to build such
conspiracies -- Kamdelia, Chord, Microsoft's PNRP,
even dynamic DNS. We can be pretty usre that some of
the existing ways will be proven broken in the future,
and that new algorithms will be found. Linking P2P SIP
to is specific conspiracy does not appear very wise.
Can't we build a level of indirection? For example, is
it possible to somehow identify the specific DHT
inside the P2P SIP URL?

-- Karst

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


		
---------------------------------
 Yahoo! Mail
 Use Photomail to share photos without annoying attachments.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060322/3498e0a9/attachment.html

From RADHIKA.R.ROY at saic.com  Wed Mar 22 11:27:32 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Wed, 22 Mar 2006 11:27:32 -0500
Subject: [p2p-sip] NATs and P2P
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31D9DEE@0591-ITS-EXMP02.us.saic.com>

Hi, all:

 

SIP into XML-RPC and similar things were proposed in the past (some drafts
were written). Now, probably the time has come to revisit of those
interesting ideas.

 

Best regards,

Radhika

 

  _____  

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
Sent: Wednesday, March 22, 2006 10:36 AM
To: P2P-SIP
Subject: Re: [p2p-sip] NATs and P2P

 

That sounds interesting idea to me.

-- mike.
-------------------------
Mike Robinson
VoIP Trainer and Consultant

Juha Heinanen <jh at tutpro.com> wrote:

Dean Willis writes:

> Skype does a lot of its work by doing things like 
> tunneling over port 443. The IETF is NOT going to accept any RFCs 
> that include this sort of stuff. It's just not going to happen. 
> It's a religion thing.

how about what i suggested: map SIP into XML-RPC rather than tunnel
over? what kind of IETF religion would prevent that?

-- juha
_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip

 

  _____  

Yahoo! Mail
Use
<http://pa.yahoo.com/*http:/us.rd.yahoo.com/evt=38867/*http:/photomail.mail.
yahoo.com>  Photomail to share photos without annoying attachments.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060322/a34ea116/attachment-0001.html

From hgs at cs.columbia.edu  Wed Mar 22 15:56:21 2006
From: hgs at cs.columbia.edu (Henning Schulzrinne)
Date: Wed, 22 Mar 2006 15:56:21 -0500
Subject: [p2p-sip] [Fwd: I-D ACTION:draft-irtf-p2prg-survey-search-00.txt]
Message-ID: <4421B9F5.80000@cs.columbia.edu>


-------------- next part --------------
An embedded message was scrubbed...
From: Internet-Drafts at ietf.org
Subject: I-D ACTION:draft-irtf-p2prg-survey-search-00.txt 
Date: Wed, 22 Mar 2006 15:50:01 -0500
Size: 6510
Url: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060322/bc1c9eb4/draft-irtf-p2prg-survey-search-00.mht

From buford at research.panasonic.com  Wed Mar 22 17:47:46 2006
From: buford at research.panasonic.com (buford@research.panasonic.com)
Date: Wed, 22 Mar 2006 17:47:46 -0500
Subject: [p2p-sip] [Fwd: I-D
 ACTION:draft-irtf-p2prg-survey-search-00.txt]
Message-ID: <a8d93eb9.3eb9a8d9@pintlmail.MITL.Research.Panasonic.COM>

This ID is a version of a journal article that will
appear in Computer Networks.  If you prefer to read
the nicely formatted PDF version, you can find it at:

http://www.eet.unsw.edu.au/~timm/

-------------- next part --------------

-------------- next part --------------
An embedded message was scrubbed...
From: Internet-Drafts at ietf.org
Subject: I-D ACTION:draft-irtf-p2prg-survey-search-00.txt 
Date: Wed, 22 Mar 2006 15:50:01 -0500
Size: 6301
Url: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060322/e1805423/draft-irtf-p2prg-survey-search-00.mht
-------------- next part --------------
_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip

From kyara543 at yahoo.com  Wed Mar 22 20:09:12 2006
From: kyara543 at yahoo.com (Kyara Yamamoto)
Date: Wed, 22 Mar 2006 17:09:12 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <442144F2.30204@research.panasonic.com>
Message-ID: <20060323010912.23253.qmail@web38612.mail.mud.yahoo.com>


>     > By bounding the P2PSIP store-value primitive as a vector of
>     > distinguished name and contact URI.
>     >
>     > It's not a question of what Kademlia can express, it's a
>     question of
>     > what P2PSIP requires.
>     >
Using an *indirect* vector of contact URIs and another protocol (MSRP) to store/retrieve a resource sounds like taking pants off to fart. It takes extra NAT penetration efforts (risk) for the second protocol to *reach* the URIs.

If David's *5x overhead* theory on multiple P2P-SIP ports is valid, a protocol suite with *2x less efficiency* and "2xx more complexity" is far from encouraging. 

"Ask all the SIP community what a pain this can be."

--Yamamoto


		
---------------------------------
Yahoo! Messenger with Voice. Make PC-to-Phone Calls to the US (and 30+ countries) for 2?/min or less.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060322/2c23468e/attachment.html

From aki.niemi at nokia.com  Wed Mar 22 11:49:42 2006
From: aki.niemi at nokia.com (Aki Niemi)
Date: Wed, 22 Mar 2006 18:49:42 +0200
Subject: [p2p-sip] SIP, ICE, and an Overlay Network
In-Reply-To: <44189C36.8000706@cisco.com>
References: <200603150909.k2F997OZ026285@cs.columbia.edu>
	<44189C36.8000706@cisco.com>
Message-ID: <44218026.7060902@nokia.com>


ext Jonathan Rosenberg wrote:
> One solution there is that the 200 OK from the UAS include multiple 
> contact values, each representing a potentially reachable IP address and 
>   a stun username/password. You then, ICE_like, check connectivity to 
> each of those, and whichever one works, you use that as the contact for 
> direct mid-dialog signaling. If none work, you use the proxy for 
> mid-dialog requests too.

Good suggestion.

Isn't this enough to fulfill the "ICE for signaling" requirement also
voiced in today's meeting?

Cheers,
Aki

From dean.willis at softarmor.com  Thu Mar 23 03:48:48 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Thu, 23 Mar 2006 02:48:48 -0600
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <17441.3644.379784.440053@rautu.tutpro.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<009901c64d15$7dd2b250$6400a8c0@comcast.net>
	<A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>
	<17441.3644.379784.440053@rautu.tutpro.com>
Message-ID: <D581A05F-7FCB-4EE3-B6CA-88099C779B70@softarmor.com>


On Mar 22, 2006, at 2:43 AM, Juha Heinanen wrote:

> Dean Willis writes:
>
>> Skype does a lot of its work by doing things like
>> tunneling over port 443. The IETF is NOT going to accept any RFCs
>> that include this sort of stuff. It's just not going to happen.
>> It's a religion thing.
>
> how about what i suggested:  map SIP into XML-RPC rather than tunnel
> over?  what kind of IETF religion would prevent that?
>

The one that keeps lots of generally useful ideas from proceeding and  
that holds as a tenet "Gee, that sortof conflicts with something we  
already have", perhaps?

Might be right, might be wrong, but it's an article of faith.

--
Dean



From eunsoo at research.panasonic.com  Thu Mar 23 23:54:16 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Thu, 23 Mar 2006 23:54:16 -0500
Subject: [p2p-sip] Ad-hoc meeting time and location during IETF65
Message-ID: <44237B78.2090102@research.panasonic.com>

Below is the information about the P2PSIP ad-hoc meeting tomorrow:
Time: 12-2pm, March 24 (Friday)
Room: Cortez

Looking forward to seeing you.
Thanks.

Eunsoo

From jh at tutpro.com  Fri Mar 24 00:43:28 2006
From: jh at tutpro.com (Juha Heinanen)
Date: Fri, 24 Mar 2006 07:43:28 +0200
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <D581A05F-7FCB-4EE3-B6CA-88099C779B70@softarmor.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<009901c64d15$7dd2b250$6400a8c0@comcast.net>
	<A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>
	<17441.3644.379784.440053@rautu.tutpro.com>
	<D581A05F-7FCB-4EE3-B6CA-88099C779B70@softarmor.com>
Message-ID: <17443.34560.590494.585408@harjus.tutpro.com>

Dean Willis writes:

 > The one that keeps lots of generally useful ideas from proceeding and  
 > that holds as a tenet "Gee, that sortof conflicts with something we  
 > already have", perhaps?

there would be no changes to SIP and the mapping to XML-RPC would be
one-to-one.  sure it should be possible to have multiple ways to carry
SIP requests between SIP entities.

-- juha

From dean.willis at softarmor.com  Fri Mar 24 13:28:41 2006
From: dean.willis at softarmor.com (Dean Willis)
Date: Fri, 24 Mar 2006 12:28:41 -0600
Subject: [p2p-sip] NATs and P2P
In-Reply-To: <17443.34560.590494.585408@harjus.tutpro.com>
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<009901c64d15$7dd2b250$6400a8c0@comcast.net>
	<A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>
	<17441.3644.379784.440053@rautu.tutpro.com>
	<D581A05F-7FCB-4EE3-B6CA-88099C779B70@softarmor.com>
	<17443.34560.590494.585408@harjus.tutpro.com>
Message-ID: <3814DE00-6B78-4AAF-9C74-1A4239081E7D@softarmor.com>


On Mar 23, 2006, at 11:43 PM, Juha Heinanen wrote:

> Dean Willis writes:
>
>> The one that keeps lots of generally useful ideas from proceeding and
>> that holds as a tenet "Gee, that sortof conflicts with something we
>> already have", perhaps?
>
> there would be no changes to SIP and the mapping to XML-RPC would be
> one-to-one.  sure it should be possible to have multiple ways to carry
> SIP requests between SIP entities.
>

I see. Actually, someone did a draft mapping of SIP to XML (I think  
it might have been XML-RPC, might have been SOAP) about 5 years ago.  
I thought it was pretty cool, but there was a strong consensus that  
it was "out of scope" for any of the existing working groups. My  
hunch is that  it would still be out of scope, and probably get a  
violent response from the IESG. But that doesn't keep such work from  
proceeding in other venues, and I suspect that there have been  
several academic mapping exercises since.

But this thread is probably getting out-of-scope for P2PSIP, which I  
believe is restricted to making conventional SIP work where the  
locator database is distributed using P2P techniques, not inventing a  
new format for SIP encoding and transport.

--
Dean


From huang-ming.pan at comcast.net  Fri Mar 24 18:16:24 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Fri, 24 Mar 2006 15:16:24 -0800
Subject: [p2p-sip] NATs and P2P
References: <4E2DC1ACE541B94A946B8D8657E372C528C858@photon.objectworld.com>
	<009901c64d15$7dd2b250$6400a8c0@comcast.net>
	<A187B83C-3AF5-4096-83D4-10BD88D48D95@softarmor.com>
	<17441.3644.379784.440053@rautu.tutpro.com>
	<D581A05F-7FCB-4EE3-B6CA-88099C779B70@softarmor.com>
	<17443.34560.590494.585408@harjus.tutpro.com>
	<3814DE00-6B78-4AAF-9C74-1A4239081E7D@softarmor.com>
Message-ID: <011d01c64f98$f938f280$6400a8c0@comcast.net>


XML-RPC might help the fusion of SIP more into Jabber than into P2P.

A Jabber spy seems sneaking in ... :-)

peter
----- Original Message -----
From: "Dean Willis" <dean.willis at softarmor.com>
To: "Juha Heinanen" <jh at tutpro.com>
Cc: "Peter Pan" <huang-ming.pan at comcast.net>; "P2P-SIP"
<p2p-sip at cs.columbia.edu>
Sent: Friday, March 24, 2006 10:28 AM
Subject: Re: [p2p-sip] NATs and P2P


>
> On Mar 23, 2006, at 11:43 PM, Juha Heinanen wrote:
>
> > Dean Willis writes:
> >
> >> The one that keeps lots of generally useful ideas from proceeding and
> >> that holds as a tenet "Gee, that sortof conflicts with something we
> >> already have", perhaps?
> >
> > there would be no changes to SIP and the mapping to XML-RPC would be
> > one-to-one.  sure it should be possible to have multiple ways to carry
> > SIP requests between SIP entities.
> >
>
> I see. Actually, someone did a draft mapping of SIP to XML (I think
> it might have been XML-RPC, might have been SOAP) about 5 years ago.
> I thought it was pretty cool, but there was a strong consensus that
> it was "out of scope" for any of the existing working groups. My
> hunch is that  it would still be out of scope, and probably get a
> violent response from the IESG. But that doesn't keep such work from
> proceeding in other venues, and I suspect that there have been
> several academic mapping exercises since.
>
> But this thread is probably getting out-of-scope for P2PSIP, which I
> believe is restricted to making conventional SIP work where the
> locator database is distributed using P2P techniques, not inventing a
> new format for SIP encoding and transport.
>
> --
> Dean
>


From karstbj at yahoo.com  Sat Mar 25 02:04:31 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Fri, 24 Mar 2006 23:04:31 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060322161604.52512.qmail@web38502.mail.mud.yahoo.com>
Message-ID: <20060325070431.4143.qmail@web53008.mail.yahoo.com>

> Could you please elaborate on why do you say "
> Linking P2P SIP
> to a particular DHT is specific conspiracy does not
> appear very wise.".

Most of the DHT technologies are still at the research
level. We can expect significant progress in security,
handling of locality, amount of management overhead,
etc. Ths, we may well expect that a DHT implementation
that looks attractive now will look terrible in say,
five years. This calls for a level of indirection.

I used the word "conspiracy" to denote the tight
coupling between notes expected in many DHT
technologies. The INSERT primitive, in most systems,
finds the node "closest" to the inserted index, and
asks that node to publish the value. In practice, this
requires a lot of trust: trust that a third party will
publish your data, just because the hash code of their
node-id is closest to the hash code of your name. That
kind of trust is easier to obtain in small groups than
in very large systems. Thus, even if we picked a
single DHT technology, we would still want a level of
indirection to indicate which particular DHT group
should be used.

> What is wrong with Chord that is being suggested?

I would rather be modest here -- Chord is a nice piece
of work, and I enjoyed reading their papers. I am
concerned mostly with two aspects: the possibility for
bad nodes to insert themselves at strategic point in
the "routing tables"; and the amount of overhead
required for table maintenance. That does not mean we
cannot use Chord, but I believe it would be a mistake
to not plan for technical evolution.

-- Karst

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From mike_robnson at yahoo.com  Sun Mar 26 15:35:37 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Sun, 26 Mar 2006 12:35:37 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060325070431.4143.qmail@web53008.mail.yahoo.com>
Message-ID: <20060326203537.32419.qmail@web38510.mail.mud.yahoo.com>

Hello Karst:

Thanks explaining your "consipracy" rationale.. 

After following the  discussion  in BOF and  in ad-hoc meetingin IETF, it appears that certain people are more interested in pushing their personal agendas of p2p sip than caring for technical merits of this group. Now that I understand your point of view, and details a bit better, and it appears to me that p2p sip idea as being proposed  seems to have a real "conspiracy" behind it.

Why are are we trying to standadize some thing in hurry that is so ill-defined and is not yet ready for standardization?  How many research or analytical papers have been published in this area? Perhaps none! Several people, I talked to seem to favour sip over p2p rather than what is being proposed by the current crop of people that are pushing their solution with limited implementation experience. May it is some IPR thingy or something I don't understand well the politics of it. When a company and University work together, I usually become suspicious of their motivation.

 
Karst Bjorgson <karstbj at yahoo.com> wrote: > Could you please elaborate on why do you say "
> Linking P2P SIP
> to a particular DHT is specific conspiracy does not
> appear very wise.".

Most of the DHT technologies are still at the research
level. We can expect significant progress in security,
handling of locality, amount of management overhead,
etc. Ths, we may well expect that a DHT implementation
that looks attractive now will look terrible in say,
five years. This calls for a level of indirection.

Hmm....
Looks like we haven't yet reached a stage where we can really
talk of  a particular DHT being standardized in a IETF protocol.

I used the word "conspiracy" to denote the tight
coupling between notes expected in many DHT
technologies. The INSERT primitive, in most systems,
finds the node "closest" to the inserted index, and
asks that node to publish the value. In practice, this
requires a lot of trust: trust that a third party will
publish your data, just because the hash code of their
node-id is closest to the hash code of your name. That
kind of trust is easier to obtain in small groups than
in very large systems. Thus, even if we picked a
single DHT technology, we would still want a level of
indirection to indicate which particular DHT group
should be used.


I see your point - it appears that technology needs more research
before it can be standardized. I believe there is an IRTF sub-group
researching into DHT implementations and other P2P area.  Formation
of this group should perhaps wait till some more input from IRTF is
available in this direction. Moreover, I didn't see a very compelling use
case that says the p2psip is needed right away. It does not do any thing that cannot be done with sip at least in the enterprise network
that I am familiar with.

I hope area director is reading this list.

> What is wrong with Chord that is being suggested?

I would rather be modest here -- Chord is a nice piece
of work, and I enjoyed reading their papers. I am
concerned mostly with two aspects: the possibility for
bad nodes to insert themselves at strategic point in
the "routing tables"; and the amount of overhead
required for table maintenance. That does not mean we
cannot use Chord, but I believe it would be a mistake
to not plan for technical evolution.

Thanks for putting this information in a clear perspective.
I do agree with your assessment.

--mike
VOIP and SIP Trainer and Consultant


		
---------------------------------
New Yahoo! Messenger with Voice. Call regular phones from your PC for low, low rates.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060326/d334ccc8/attachment.html

From karstbj at yahoo.com  Sun Mar 26 22:45:29 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Sun, 26 Mar 2006 19:45:29 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060326203537.32419.qmail@web38510.mail.mud.yahoo.com>
Message-ID: <20060327034529.92603.qmail@web53002.mail.yahoo.com>


> Why are are we trying to standadize some thing in
> hurry that is so ill-defined and is not yet ready
> for standardization?  How many research or
> analytical papers have been published in this area?
> Perhaps none! 

Well, it is easy to see the potential benefits of a
standardized P2P version of SIP. Power to the people,
end to end architecture, freedom to connect, and all
that sort of things. Definitely interesting, and a
worthy endeavour for the IETF. 

On the other hand, the current approach smells a lot
like "rushing to the solution before we actually
understand the problem". David Bryan's draft
(draft-bryan-sipping-p2p-02) is typical of that
approach: it describes in great detail how to
implement a DHT using SIP formats and algorithms
derived from CHORD, but this degree of detail can only
be useful if actually know what to optimize for. What
about security, privacy, resistance to DOS attacks,
management overhead, selection of supernodes? On there
are certainly a bunch of other problems...

-- Karst


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From bryan at ethernot.org  Sun Mar 26 23:18:59 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Sun, 26 Mar 2006 23:18:59 -0500
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060327034529.92603.qmail@web53002.mail.yahoo.com>
References: <20060326203537.32419.qmail@web38510.mail.mud.yahoo.com>
	<20060327034529.92603.qmail@web53002.mail.yahoo.com>
Message-ID: <8b2769930603262018n20bbde62sf66809a4451203f6@mail.gmail.com>

As has been pointed out before on this list (dozens of times), is
clearly stated in the draft, and I even asked EKR to say in his
presentation at the plenary -- the draft is a strawman. It is there to
prove such a solution is possible. In the earliest days of this
effort, it was not clear such a solution was possible, so we went and
built one, and created the draft to detail what we did. As we work on
our idea, we update the draft.

If you have alternate suggestions or would like to present solutions
to the list of problems you mention (or the even longer list of
problems we list in the draft itself, including the ones you mention),
please do! That is the purpose of the draft.

No one on this list -- most certainly not the authors -- have ever
claimed it is a complete solution, is ready to be deployed, or even
reflects what an IETF group is likely to adopt. Several people in this
thread have implied that is the case. It most certainly is not. It is
there as an example, and I think it is useful to update it as we go to
reflect one possible approach in enough detail to allow people to
possibly implement and play with the ideas.

I intend to continue to update it, but I expect many other
suggestions, alternate drafts, etc. from others in this list as well.
There have been suggestions from Eunsoo, Philip Matthews, Henning and
others from Columbia, and many others. Please feel free to suggest
something different. You can write an alternate draft, or if as you
say, you think that having drafts of protocols is too much detail,
please outline a suggestion in email what you think might be a better
approach, suggest comments for the requirements draft, suggest a new
architecture draft. There are many avenues for you to suggest
something.

Neither I nor any of my coauthors have said our draft is the only way
-- quite to opposite, we have changed to a design that separates
search from invite in response to suggestions in Vancouver.

If you have better ideas -- present them. Repeatedly complaining the
draft has flaws or claiming that there is a conspiracy against
proposed designs using other mechanisms that haven't even been
submitted yet is not helping advance the efforts of this group. Do you
have a suggestion for resistance to DOS attacks? Write it up! An idea
about selecting supernodes? Please suggest! We are very interested in
contributions. Please make them.

David Bryan


On 3/26/06, Karst Bjorgson <karstbj at yahoo.com> wrote:
>
> > Why are are we trying to standadize some thing in
> > hurry that is so ill-defined and is not yet ready
> > for standardization?  How many research or
> > analytical papers have been published in this area?
> > Perhaps none!
>
> Well, it is easy to see the potential benefits of a
> standardized P2P version of SIP. Power to the people,
> end to end architecture, freedom to connect, and all
> that sort of things. Definitely interesting, and a
> worthy endeavour for the IETF.
>
> On the other hand, the current approach smells a lot
> like "rushing to the solution before we actually
> understand the problem". David Bryan's draft
> (draft-bryan-sipping-p2p-02) is typical of that
> approach: it describes in great detail how to
> implement a DHT using SIP formats and algorithms
> derived from CHORD, but this degree of detail can only
> be useful if actually know what to optimize for. What
> about security, privacy, resistance to DOS attacks,
> management overhead, selection of supernodes? On there
> are certainly a bunch of other problems...
>
> -- Karst
>
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>


From karstbj at yahoo.com  Mon Mar 27 00:46:20 2006
From: karstbj at yahoo.com (Karst Bjorgson)
Date: Sun, 26 Mar 2006 21:46:20 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <8b2769930603262018n20bbde62sf66809a4451203f6@mail.gmail.com>
Message-ID: <20060327054620.60580.qmail@web53003.mail.yahoo.com>

> Repeatedly complaining ... that there is a
> conspiracy against proposed designs...

To be clear: I used the word "conspiracy" to describe
the tight relation between nodes in a DHT, as in:
"nodes conspire to store and resolve hashes." I am
sorry that it somehow implied an alternate meaning. I
do not believe that there is inappropriate behavior
going on in the proposed working group.

-- Karst



__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

From bryan at ethernot.org  Mon Mar 27 01:14:31 2006
From: bryan at ethernot.org (David A. Bryan)
Date: Mon, 27 Mar 2006 01:14:31 -0500
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060327054620.60580.qmail@web53003.mail.yahoo.com>
References: <8b2769930603262018n20bbde62sf66809a4451203f6@mail.gmail.com>
	<20060327054620.60580.qmail@web53003.mail.yahoo.com>
Message-ID: <8b2769930603262214h37b8eb34n9a55e24516e9bef5@mail.gmail.com>

Indeed you did not use that term, sorry to imply you did.

The reply was in general to several messages in this thread, and that
particular comment was in response to the concern Mike expressed (and
Kyara seemed to imply earlier) that there was a specific effort  to
exclude ideas besides Chord over SIP. There is not. Do I believe that
approach has merits? Certainly, otherwise I would not have spent time
writing the draft. I currently think it is a very good approach, and
will continue to advocate it as long as I believe it is. I can
certainly be convinced something else is better. Many others on the
list have proposed other mechanisms, and I expect even more.

Nothing at all has been decided, and none of the drafts currently
proposed are polished or enjoy consensus from the (non-existent as
yet) WG. It is very early. All ideas are on the table, and I think I
speak for everyone in saying we have many open problems.

The only things that are off the table are things that simply won't
fly for a group called P2PSIP at the IETF -- things like tunneling
messages over port 80 or fundamentally redesigning SIP simply won't be
allowed in IETF and are out of scope. That is the only constraint
right now.

--David Bryan

On 3/27/06, Karst Bjorgson <karstbj at yahoo.com> wrote:
> > Repeatedly complaining ... that there is a
> > conspiracy against proposed designs...
>
> To be clear: I used the word "conspiracy" to describe
> the tight relation between nodes in a DHT, as in:
> "nodes conspire to store and resolve hashes." I am
> sorry that it somehow implied an alternate meaning. I
> do not believe that there is inappropriate behavior
> going on in the proposed working group.
>
> -- Karst
>
>
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>


From mike_robnson at yahoo.com  Mon Mar 27 10:23:01 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Mon, 27 Mar 2006 07:23:01 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <8b2769930603262214h37b8eb34n9a55e24516e9bef5@mail.gmail.com>
Message-ID: <20060327152301.56487.qmail@web38502.mail.mud.yahoo.com>

Hello David:
 
 Let me be very clear that I have nothing against your draft and I am aware that you have been putting great efforts in developing a version based on p2p sip. As a matter of fact, I applaud *your* efforts in maintaining p2p sip site, but do not necessarily agree, as others have pointed out too, that the technology is at a stage where standardization is not possible or a wise choice yet.  Karst has given many specific issues that range security, node insertion, dynamics of DHT maintenance etc. As a research project - please go ahead and publish your papers, your results, your code, let some of us try it and see if it is really worth standardization, or where we need improvements. SIP took over 10 years to be where it is today and still sip and sipping group is dealing with many many new drafts addressing a lot of new issues or problems. So, is p2p sip in a shape where a standards body should bless it?  Where are the data points to support one choice over others? My opinion is!
  that if
 p2p file transfer applications can survive without standardization for some time and flourish, p2p sip can too. 
 
 My comments were more directed at some presentations that I saw which neither had any specific ideas, but still trying to push p2p sip standardization, and seems to be in more hurry than you. I am just intrigued why would some people familiar with sipping and sip WG work load would be so interested without having any new specific ideas to address these issues to create a new WG? What is that they are looking for in return? I am not saying that there is any thing inappropriate going on, but I am still curious why so much of hurry.
 
 The question that area director and others need to answer - Is a hurriedly crafted standard that may cause a lot of issues later on is better than no standard at all? That is all, folks. 
 
 I need to finish some installation today so may not see my response immediately.
 
 --mike
 VOIP and SIP Trainer and Consultant


"David A. Bryan" <bryan at ethernot.org> wrote: Indeed you did not use that term, sorry to imply you did.

The reply was in general to several messages in this thread, and that
particular comment was in response to the concern Mike expressed (and
Kyara seemed to imply earlier) that there was a specific effort  to
exclude ideas besides Chord over SIP. There is not. Do I believe that
approach has merits? Certainly, otherwise I would not have spent time
writing the draft. I currently think it is a very good approach, and
will continue to advocate it as long as I believe it is. I can
certainly be convinced something else is better. Many others on the
list have proposed other mechanisms, and I expect even more.

Nothing at all has been decided, and none of the drafts currently
proposed are polished or enjoy consensus from the (non-existent as
yet) WG. It is very early. All ideas are on the table, and I think I
speak for everyone in saying we have many open problems.

The only things that are off the table are things that simply won't
fly for a group called P2PSIP at the IETF -- things like tunneling
messages over port 80 or fundamentally redesigning SIP simply won't be
allowed in IETF and are out of scope. That is the only constraint
right now.

--David Bryan

On 3/27/06, Karst Bjorgson  wrote:
> > Repeatedly complaining ... that there is a
> > conspiracy against proposed designs...
>
> To be clear: I used the word "conspiracy" to describe
> the tight relation between nodes in a DHT, as in:
> "nodes conspire to store and resolve hashes." I am
> sorry that it somehow implied an alternate meaning. I
> do not believe that there is inappropriate behavior
> going on in the proposed working group.
>
> -- Karst
>
>
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>


		
---------------------------------
Yahoo! Messenger with Voice. Make PC-to-Phone Calls to the US (and 30+ countries) for 2?/min or less.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060327/be670666/attachment-0001.html

From henry at pulver.com  Mon Mar 27 11:17:48 2006
From: henry at pulver.com (Henry Sinnreich)
Date: Mon, 27 Mar 2006 10:17:48 -0600
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060326203537.32419.qmail@web38510.mail.mud.yahoo.com>
Message-ID: <200603271622.k2RGJDcL011957@cs.columbia.edu>

Mike Robinson writes:

 

>How many research ! or analytical papers have been published in this area? 

I have found by Googling a reference listing 4,800 papers on P2P. Hopefully,
this enough to get us started. :-)

Henry

  _____  

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
Sent: Sunday, March 26, 2006 2:36 PM
To: Karst Bjorgson; Kyara Yamamoto; Peter Pan
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document

 

Hello Karst:

Thanks explaining your "consipracy" rationale.. 

After following the  discussion  in BOF and  in ad-hoc meetingin IETF, it
appears that certain people are more interested in pushing their personal
agendas of p2p sip than caring for technical merits of this group. Now that
I understand your point of view, and details a bit better, and it appears to
me that p2p sip idea as being proposed  seems to have a real "conspiracy"
behind it.

Why are are we trying to standadize some thing in hurry that is so
ill-defined and is not yet ready for standardization?  How many research !
or analytical papers have been published in this area? Perhaps none! Several
people, I talked to seem to favour sip over p2p rather than what is being
proposed by the current crop of people that are pushing their solution with
limited implementation experience. May it is some IPR thingy or something I
don't understand well the politics of it. When a company and University work
together, I usually become suspicious of their motivation.


Karst Bjorgson <karstbj at yahoo.com> wrote:

> Could you please elaborate on why do you say "
> Linking P2P SIP
> to a particular DHT is specific conspiracy does not
> appear very wise.".

Most of the DHT technologies are still at the research
level. We can expect significant progress in security,
handling of locality, amount of management overhead,
etc. Ths, we may well expect that a DHT implementation
that looks attractive now will look terrible in say,
five years. This calls for a level of indirection.

Hmm....
Looks like we haven't yet reached a stage where we can really
talk of  a particular DHT being standardized in a IETF protocol.

I used the word "conspiracy" to denote the tight
coupling between notes expected in many DHT
technologies. The INSERT primitive, in most systems,
finds the node "closest" to the inserted index, and
asks that node to publish the value. In practice, this
requires a lot of trust: trust that a third party will
publish your data, just because the hash code of their
node-id is closest to the hash code of your name. That
kind of trust is easier to obtain! in small groups than
in very large systems. Thus, even if we picked a
single DHT technology, we would still want a level of
indirection to indicate which particular DHT group
should be used.


I see your point - it appears that technology needs more research
before it can be standardized. I believe there is an IRTF sub-group
researching into DHT implementations and other P2P area.  Formation
of this group should perhaps wait till some more input from IRTF is
available in this direction. Moreover, I didn't see a very compelling use
case that says the p2psip is needed right away. It does not do any thing
that cannot be done with sip at least in the enterprise network
that I am familiar with.

I hope area director is reading this list.

> What is wrong with Chord that is being suggested?

I would rather b! e modest here -- Chord is a nice piece
of work, and I enjoyed reading their papers. I am
concerned mostly with two aspects: the possibility for
bad nodes to insert themselves at strategic point in
the "routing tables"; and the amount of overhead
required for table maintenance. That does not mean we
cannot use Chord, but I believe it would be a mistake
to not plan for technical evolution.

Thanks for putting this information in a clear perspective.
I do agree with your assessment.

--mike
VOIP and SIP Trainer and Consultant

 

  _____  

New Yahoo!
<http://us.rd.yahoo.com/mail_us/taglines/postman4/*http:/us.rd.yahoo.com/evt
=39666/*http:/beta.messenger.yahoo.com%20>  Messenger with Voice. Call
regular phones from your PC for low, low rates.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060327/234981fa/attachment.html

From huang-ming.pan at comcast.net  Mon Mar 27 12:23:45 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Mon, 27 Mar 2006 09:23:45 -0800
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <001a01c651c3$341ee3b0$6400a8c0@comcast.net>

Not only on P2P species, if the comparison between Jabber and SIP in
attached paper is not too exaggerating, especially regarding "Bandwidth
Requirements and Scalability Implications", SIP protocol may need revamps
before acting as a P2P "wrapper" protocol.

http://www.jabber.com/index.cgi?CONTENT_ID=55&VMX_TRACKED=YES

Just thought the place should have more open minds for both positive
suggestions and negative challenges, even bitter critiques; otherwise it is
just another piece of tofu under construction like the volunerable Windows.

peter
----- Original Message -----
From: Henry Sinnreich
To: 'Mike Robinson' ; 'Karst Bjorgson' ; 'Kyara Yamamoto' ; 'Peter Pan'
Cc: p2p-sip at cs.columbia.edu
Sent: Monday, March 27, 2006 8:17 AM
Subject: RE: [p2p-sip] Some comments on Use-cases document


Mike Robinson writes:

>How many research ! or analytical papers have been published in this area?
I have found by Googling a reference listing 4,800 papers on P2P. Hopefully,
this enough to get us started. J
Henry



From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
Sent: Sunday, March 26, 2006 2:36 PM
To: Karst Bjorgson; Kyara Yamamoto; Peter Pan
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document

Hello Karst:

Thanks explaining your "consipracy" rationale..

After following the  discussion  in BOF and  in ad-hoc meetingin IETF, it
appears that certain people are more interested in pushing their personal
agendas of p2p sip than caring for technical merits of this group. Now that
I understand your point of view, and details a bit better, and it appears to
me that p2p sip idea as being proposed  seems to have a real "conspiracy"
behind it.

Why are are we trying to standadize some thing in hurry that is so
ill-defined and is not yet ready for standardization?  How many research !
or analytical papers have been published in this area? Perhaps none! Several
people, I talked to seem to favour sip over p2p rather than what is being
proposed by the current crop of people that are pushing their solution with
limited implementation experience. May it is some IPR thingy or something I
don't understand well the politics of it. When a company and University work
together, I usually become suspicious of their motivation.


Karst Bjorgson <karstbj at yahoo.com> wrote:
> Could you please elaborate on why do you say "
> Linking P2P SIP
> to a particular DHT is specific conspiracy does not
> appear very wise.".

Most of the DHT technologies are still at the research
level. We can expect significant progress in security,
handling of locality, amount of management overhead,
etc. Ths, we may well expect that a DHT implementation
that looks attractive now will look terrible in say,
five years. This calls for a level of indirection.

Hmm....
Looks like we haven't yet reached a stage where we can really
talk of  a particular DHT being standardized in a IETF protocol.

I used the word "conspiracy" to denote the tight
coupling between notes expected in many DHT
technologies. The INSERT primitive, in most systems,
finds the node "closest" to the inserted index, and
asks that node to publish the value. In practice, this
requires a lot of trust: trust that a third party will
publish your data, just because the hash code of their
node-id is closest to the hash code of your name. That
kind of trust is easier to obtain! in small groups than
in very large systems. Thus, even if we picked a
single DHT technology, we would still want a level of
indirection to indicate which particular DHT group
should be used.


I see your point - it appears that technology needs more research
before it can be standardized. I believe there is an IRTF sub-group
researching into DHT implementations and other P2P area.  Formation
of this group should perhaps wait till some more input from IRTF is
available in this direction. Moreover, I didn't see a very compelling use
case that says the p2psip is needed right away. It does not do any thing
that cannot be done with sip at least in the enterprise network
that I am familiar with.

I hope area director is reading this list.

> What is wrong with Chord that is being suggested?

I would rather b! e modest here -- Chord is a nice piece
of work, and I enjoyed reading their papers. I am
concerned mostly with two aspects: the possibility for
bad nodes to insert themselves at strategic point in
the "routing tables"; and the amount of overhead
required for table maintenance. That does not mean we
cannot use Chord, but I believe it would be a mistake
to not plan for technical evolution.

Thanks for putting this information in a clear perspective.
I do agree with your assessment.

--mike
VOIP and SIP Trainer and Consultant




New Yahoo! Messenger with Voice. Call regular phones from your PC for low,
low rates.


From mike_robnson at yahoo.com  Mon Mar 27 13:17:00 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Mon, 27 Mar 2006 10:17:00 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060327152301.56487.qmail@web38502.mail.mud.yahoo.com>
Message-ID: <20060327181700.12068.qmail@web38504.mail.mud.yahoo.com>

Oops!
   
   - the second sentence should have read:
   
  As a matter of fact, I applaud *your* efforts in maintaining p2p sip site, but do not necessarily agree, as others have pointed out too, that the technology is at a stage where standardization is possible or a wise choice yet.

  (there was an extra "not" inserted in the half sleepy state)
  
Mike Robinson <mike_robnson at yahoo.com> wrote:
  Hello David:

Let me be very clear that I have nothing against your draft and I am aware that you have been putting great efforts in developing a version based on p2p sip. As a matter of fact, I applaud *your* efforts in maintaining p2p sip site, but do not necessarily agree, as others have pointed out too, that the technology is at a stage where standardization is not possible or a wise choice yet.  Karst has given many specific issues that range security, node insertion, dynamics of DHT maintenance etc. As a research project - please go ahead and publish your papers, your results, your code, let some of us try it and see if it is really worth standardization, or where we need improvements. SIP took over 10 years to be where it is today and still sip and sipping group is dealing with many many new drafts addressing a lot of new issues or problems. So, is p2p sip in a shape where a standards body should bless it?  Where are the data points to support one choice over others? My opinion is !
 that if
 p2p file transfer applications can survive without standardization for some time and flourish, p2p sip can too. 

My comments were more directed at some presentations that I saw which neither had any specific ideas, but still trying to push p2p sip standardization, and seems to be in more hurry than you. I am just intrigued why would some people familiar with sipping and sip WG work load would be so interested without having any new specific ideas to address these issues to create a new WG? What is that they are looking for in return? I am not saying that there is any thing inappropriate going on, but I am still curious why so much of hurry.

The question that area director and others need to answer - Is a hurriedly crafted standard that may cause a lot of issues later on is better than no standard at all? That is all, folks. 

I need to finish some installation today so may not see my response immediately.

--mike
VOIP and SIP Trainer and Consultant


"David A. Bryan" <bryan at ethernot.org> wrote:  Indeed you did not use that term, sorry to imply you did.

The reply was in general to several messages in this thread, and that
particular comment was in response to the concern Mike expressed (and
Kyara seemed to imply earlier) that there was a specific effort to
exclude ideas besides Chord over SIP. There is not. Do I believe that
approach has merits? Certainly, otherwise I would not have spent time
writing the draft. I currently think it is a very good approach, and
will continue to advocate it as long as I believe it is. I can
certainly be convinced something else is better. Many others on the
list have proposed other mechanisms, and I expect even more.

Nothing at all has been decided, and none of the drafts currently
proposed are polished or enjoy consensus from the (non-existent as
yet) WG. It is very early. All ideas are on the table, and I think I
speak for everyone in saying we have many open problems.

The only things that are off the table are things that simply won't
fly for a group called P2PSIP at the IETF -- things like tunneling
messages over port 80 or fundamentally redesigning SIP simply won't be
allowed in IETF and are out of scope. That is the only constraint
right now.

--David Bryan

On 3/27/06, Karst Bjorgson wrote:
> > Repeatedly complaining ... that there is a
> > conspiracy against proposed designs...
>
> To be clear: I used the word "conspiracy" to describe
> the tight relation between nodes in a DHT, as in:
> "nodes conspire to store and resolve hashes." I am
> sorry that it somehow implied an alternate meaning. I
> do not believe that there is inappropriate behavior
> going on in the proposed working group.
>
> -- Karst
>
>
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam? Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>

    
---------------------------------
  Yahoo! Messenger with Voice. Make PC-to-Phone Calls to the US (and 30+ countries) for 2?/min or less.

			
---------------------------------
Yahoo! Messenger with Voice. PC-to-Phone calls for ridiculously low rates.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060327/00aa6c23/attachment.html

From mike_robnson at yahoo.com  Mon Mar 27 13:32:12 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Mon, 27 Mar 2006 10:32:12 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <20060327183212.13955.qmail@web38512.mail.mud.yahoo.com>

 
  Henry:
   
  Do I need to tell that 4800 P2P google references have nothing to do with P2P SIP?
   
  Please let us not obfuscate the issues. You know, as well as, anyone sincere about the value of standards, that the pre-mature standardization leads to bad design as well as failure of the standards. As I mentioned there is no compelling use case presented yet that says we need to "standardize" p2p sip now without even bothering to get output from P2P IRTF group which is looking into core issues of P2P research.  
   
  Many would say that we first need an IETF standard for P2P file sharing before we want to use "P2P" to design P2P SIP. We need more research on P2P SIP before we standardize any particular "conspiracy" (I use conspiracy in the sense of Karst definition of linking design with a particular DHT)
   
  -mike
  VOIP and SIP Trainer and Consultant
   
  
Henry Sinnreich <henry at pulver.com> wrote:
        v\:* {behavior:url(#default#VML);}  o\:* {behavior:url(#default#VML);}  w\:* {behavior:url(#default#VML);}  .shape {behavior:url(#default#VML);}                Mike Robinson writes:
   
  >How many research ! or analytical papers have been published in this area? 
    I have found by Googling a reference listing 4,800 papers on P2P. Hopefully, this enough to get us started. J
  Henry

      
---------------------------------
  
  From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
Sent: Sunday, March 26, 2006 2:36 PM
To: Karst Bjorgson; Kyara Yamamoto; Peter Pan
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document

   
  Hello Karst:

Thanks explaining your "consipracy" rationale.. 

After following the  discussion  in BOF and  in ad-hoc meetingin IETF, it appears that certain people are more interested in pushing their personal agendas of p2p sip than caring for technical merits of this group. Now that I understand your point of view, and details a bit better, and it appears to me that p2p sip idea as being proposed  seems to have a real "conspiracy" behind it.

Why are are we trying to standadize some thing in hurry that is so ill-defined and is not yet ready for standardization?  How many research ! or analytical papers have been published in this area? Perhaps none! Several people, I talked to seem to favour sip over p2p rather than what is being proposed by the current crop of people that are pushing their solution with limited implementation experience. May it is some IPR thingy or something I don't understand well the politics of it. When a company and University work together, I usually become suspicious of their motivation.


Karst Bjorgson <karstbj at yahoo.com> wrote:
  > Could you please elaborate on why do you say "
> Linking P2P SIP
> to a particular DHT is specific conspiracy does not
> appear very wise.".

Most of the DHT technologies are still at the research
level. We can expect significant progress in security,
handling of locality, amount of management overhead,
etc. Ths, we may well expect that a DHT implementation
that looks attractive now will look terrible in say,
five years. This calls for a level of indirection.

Hmm....
Looks like we haven't yet reached a stage where we can really
talk of  a particular DHT being standardized in a IETF protocol.

I used the word "conspiracy" to denote the tight
coupling between notes expected in many DHT
technologies. The INSERT primitive, in most systems,
finds the node "closest" to the inserted index, and
asks that node to publish the value. In practice, this
requires a lot of trust: trust that a third party will
publish your data, just because the hash code of their
node-id is closest to the hash code of your name. That
kind of trust is easier to obtain! in small groups than
in very large systems. Thus, even if we picked a
single DHT technology, we would still want a level of
indirection to indicate which particular DHT group
should be used.


I see your point - it appears that technology needs more research
before it can be standardized. I believe there is an IRTF sub-group
researching into DHT implementations and other P2P area.  Formation
of this group should perhaps wait till some more input from IRTF is
available in this direction. Moreover, I didn't see a very compelling use
case that says the p2psip is needed right away. It does not do any thing that cannot be done with sip at least in the enterprise network
that I am familiar with.

I hope area director is reading this list.

> What is wrong with Chord that is being suggested?

I would rather b! e modest here -- Chord is a nice piece
of work, and I enjoyed reading their papers. I am
concerned mostly with two aspects: the possibility for
bad nodes to insert themselves at strategic point in
the "routing tables"; and the amount of overhead
required for table maintenance. That does not mean we
cannot use Chord, but I believe it would be a mistake
to not plan for technical evolution.

Thanks for putting this information in a clear perspective.
I do agree with your assessment.

--mike
VOIP and SIP Trainer and Consultant
   
    
---------------------------------
  
  New Yahoo! Messenger with Voice. Call regular phones from your PC for low, low rates.



			
---------------------------------
Yahoo! Messenger with Voice. PC-to-Phone calls for ridiculously low rates.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060327/0331583d/attachment-0001.html

From dbarrett at quinthar.com  Mon Mar 27 14:40:07 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Mon, 27 Mar 2006 11:40:07 -0800
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060327183212.13955.qmail@web38512.mail.mud.yahoo.com>
Message-ID: <200603271940.k2RJeIcK002639@cs.columbia.edu>

I guess one view is the IETF should forever be relegated to the janitorial
role of cleaning up after the party's over.  But I believe the IETF can
participate in and even promote the party through standardization.  I'd like
the IETF to do more than simply formalize de facto standards, but actually
create the standards that attract the party.  

 

In the case of P2P-SIP, I entirely agree there are multiple, valid, proven
ways to solve each of these problems - none of which is the clear victor.
And I also agree there are outstanding issues that have no proven solutions.

 

If we pick amongst the best practices today, and skip the problems to which
we have no clear solution, I personally believe that'll give us the most
immediate value, while paving the road for future improvements. I don't
believe we'll ever meet any higher standard to get started than has already
been met. 

 

-david 

 

  _____  

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
Sent: Monday, March 27, 2006 10:32 AM
To: henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document

 

 

Henry:

 

Do I need to tell that 4800 P2P google references have nothing to do with
P2P SIP?

 

Please let us not obfuscate the issues. You know, as well as, anyone sincere
about the value of standards, that the pre-mature standardization leads to
bad design as well as failure of the standards. As I mentioned there is no
compelling use case presented yet that says we need to "standardize" p2p sip
now without even bothering to get output from P2P IRTF group which is
looking into core issues of P2P research.  

 

Many would say that we first need an IETF standard for P2P file sharing
before we want to use "P2P" to design P2P SIP. We need more research on P2P
SIP before we standardize any particular "conspiracy" (I use conspiracy in
the sense of Karst definition of linking design with a particular DHT)

 

-mike

V! OIP and SIP Trainer and Consultant

 


Henry Sinnreich <henry at pulver.com> wrote:

Mike Robinson writes:

 

>How many research ! or analytical papers have been published in this area? 

I have found by Googling a reference listing 4,800 papers on P2P. Hopefully,
this enough to get us started. :-)

Henry


  _____  


From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
Sent: Sunday, March 26, 2006 2:36 PM
To: Karst Bjorgson; Kyara Yamamoto; Peter Pan
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document

 

! Hello Karst:

Thanks explaining your "consipracy" rationale.. 

After following the  discussion  in BOF and  in ad-hoc meetingin IETF, it
appears that certain people are more interested in pushing their personal
agendas of p2p sip than caring for technical merits of this group. Now that
I understand your point of view, and details a bit better, and it appears to
me that p2p sip idea as being proposed  seems to have a real "conspiracy"
behind it.

Why are are we trying to standadize some thing in hurry that is so
ill-defined and is not yet ready for standardization?  How many research !
or analytical papers have been published in this area? Perhaps none! Several
people, I talked to seem to favour sip over p2p rather than what is being
proposed by the current crop of peo! ple that are pushing their solution
with limited implementation experience. May it is some IPR thingy or
something I don't understand well the politics of it. When a company and
University work together, I usually become suspicious of their motivation.


Karst Bjorgson <karstbj at yahoo.com> wrote:

> Could you please elaborate on why do you say "
> Linking P2P SIP
> to a particular DHT is specific conspiracy does not
> appear very wise.".

Most of the DHT technologies are still at the research
level. We can expect significant progress in security,
handling of locality, amount of management overhead,
etc. Ths, we may well expect that a DHT implementation
that looks attractive now wil! l look terrible in say,
five years. This calls for a level of indirection.

Hmm....
Looks like we haven't yet reached a stage where we can really
talk of  a particular DHT being standardized in a IETF protocol.

I used the word "conspiracy" to denote the tight
coupling between notes expected in many DHT
technologies. The INSERT primitive, in most systems,
finds the node "closest" to the inserted index, and
asks that node to publish the value. In practice, this
requires a lot of trust: trust that a third party will
publish your data, just because the hash code of their
node-id is closest to the hash code of your name. That
kind of trust is easier to obtain! in small groups than
in very large systems. Thus, even if we picked a
single DHT technology, we would still want a level of
indirection to indicate which particular DHT group
should be used.


I see your point - it appears that technology needs more research
before it can be standardized. I believe there is an IRTF sub-group
researching into DHT implementations and other P2P area.  Formation
of this group should perhaps wait till some more input from IRTF is
available in this direction. Moreover, I didn't see a very compelling use
case that says the p2psip is needed right away. It does not do any thing
that cannot be done with sip at least in the enterprise network
that I am familiar with.

I hope area director is reading this list.

> What is wrong with Chord that is being suggested?

I would rather b! e modest here -- Chord is a nice piece
of work, and I enjoyed reading their papers. I am
concerned mostly with two aspects: the possibility for
bad nodes to insert themselves at strategic point in
the "routing tables"; and the amount of overhead
requ! ired for table maintenance. That does not mean we
cannot use Chord, but I believe it would be a mistake
to not plan for technical evolution.

Thanks for putting this information in a clear perspective.
I do agree with your assessment.

--mike
VOIP and SIP Trainer and Consultant

 


  _____  


New Ya!
<http://us.rd.yahoo.com/mail_us/taglines/postman4/*http:/us.rd.yahoo.com/evt
=39666/*http:/beta.messenger.yahoo.com%20>  hoo! Messenger with Voice. Call
regular phones from your PC for low, low rates.

 

  _____  

Yahoo!
<http://us.rd.yahoo.com/mail_us/taglines/postman3/*http:/us.rd.yahoo.com/evt
=39666/*http:/beta.messenger.yahoo.com>  Messenger with Voice. PC-to-Phone
calls for ridiculously low rates.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060327/ab31fe7f/attachment.html

From RADHIKA.R.ROY at saic.com  Mon Mar 27 14:51:38 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Mon, 27 Mar 2006 14:51:38 -0500
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31D9DF3@0591-ITS-EXMP02.us.saic.com>

I agree with David.

 

It is also NOT harm to linkage a particular design with a particular DHT, if
this is the good choice for a given time. However, other proposals are also
welcome. It is an open to all of us.

 

Standardizations are always done throughout the world if there is a
consensus to go for something that is good. With respect P2P-SIP, there are
a lot of things to be solved. However, if something provides immediate
value, as David and many others of this list have opined, what is the
problem for standardization? A better standard in the future can always
supersede an inferior standard of the past.

 

Best regards,

Radhika

 

  _____  

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David Barrett
Sent: Monday, March 27, 2006 2:40 PM
To: 'Mike Robinson'; henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto';
'Peter Pan'
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document

 

I guess one view is the IETF should forever be relegated to the janitorial
role of cleaning up after the party's over.  But I believe the IETF can
participate in and even promote the party through standardization.  I'd like
the IETF to do more than simply formalize de facto standards, but actually
create the standards that attract the party.  

 

In the case of P2P-SIP, I entirely agree there are multiple, valid, proven
ways to solve each of these problems - none of which is the clear victor.
And I also agree there are outstanding issues that have no proven solutions.

 

If we pick amongst the best practices today, and skip the problems to which
we have no clear solution, I personally believe that'll give us the most
immediate value, while paving the road for future improvements. I don't
believe we'll ever meet any higher standard to get started than has already
been met. 

 

-david 

 

  _____  

From: p2p-sip-bounces at cs.columbia.edu
[mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
Sent: Monday, March 27, 2006 10:32 AM
To: henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document

 

 

Henry:

 

Do I need to tell that 4800 P2P google references have nothing to do with
P2P SIP?

 

Please let us not obfuscate the issues. You know, as well as, anyone sincere
about the value of standards, that the pre-mature standardization leads to
bad design as well as failure of the standards. As I mentioned there is no
compelling use case presented yet that says we need to "standardize" p2p sip
now without even bothering to get output from P2P IRTF group which is
looking into core issues of P2P research.  

 

Many would say that we first need an IETF standard for P2P file sharing
before we want to use "P2P" to design P2P SIP. We need more research on P2P
SIP before we standardize any particular "conspiracy" (I use conspiracy in
the sense of Karst definition of linking design with a particular DHT)

 

-mike

V! OIP and SIP Trainer and Consultant

 


Henry Sinnreich <henry at pulver.com> wrote:

Mike Robinson writes:

 

>How many research ! or analytical papers have been published in this area? 

I have found by Googling a reference listing 4,800 papers on P2P. Hopefully,
this enough to get us started. :-)

Henry

rates.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060327/5075e129/attachment-0001.html

From gregd at research.panasonic.com  Mon Mar 27 17:45:11 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Mon, 27 Mar 2006 17:45:11 -0500
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31D9DF3@0591-ITS-EXMP02.us.saic.com>
References: <A879F72922C16E4F80EA2F34EBF2AAA31D9DF3@0591-ITS-EXMP02.us.saic.com>
Message-ID: <44286AF7.1010402@research.panasonic.com>

Hi Radhika,

Roy, Radhika R. wrote:
> I agree with David.
> 
>  
> 
> It is also NOT harm to linkage a particular design with a particular 
> DHT, if this is the good choice for a given time. However, other 
> proposals are also welcome. It is an open to all of us.
> 
>  
> 
> Standardizations are always done throughout the world if there is a 
> consensus to go for something that is good. With respect P2P-SIP, there 
> are a lot of things to be solved. However, if something provides 
> immediate value, as David and many others of this list have opined, what 
> is the problem for standardization? A better standard in the future can 
> always supersede an inferior standard of the past.
> 

I agree that it may be possible to upgrade or replace technologies or
algorithms later, but there are two caveats:

1) The base system has to be built in a way which allows upgrades and 
extensions.

2) Once there is a deployed base, it is tricky to avoid forklift 
upgrades (e.g. we see this today with IPv6).

So long as provisions are made to ensure compatibility with these goals,
it may be possible to create something now with extension or upgrade
in the future.

Greg


From mike_robnson at yahoo.com  Mon Mar 27 19:14:15 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Mon, 27 Mar 2006 16:14:15 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <A879F72922C16E4F80EA2F34EBF2AAA31D9DF3@0591-ITS-EXMP02.us.saic.com>
Message-ID: <20060328001415.97808.qmail@web38513.mail.mud.yahoo.com>


Radhika:

You agree that there are many problems  that are not resolved yet, and no solution is  available. That means more research needs to be done. Research belongs in the labs, Universities or in IRTF. 

It is not clear how you can change a DHT and still be compatible since SIP messages and headers fields are so much dependent on the type of underlying DHT used. The data maintained by each DHT is different and so are node insertion and removal procedures. I believe the security issues of each DHT types would also be different, so when someone says that they can replace DHT later on, they are just being too simplistic.

Do you really want to create OSPF versus IS-IS kind of two incompatible solutions?

--mike


"Roy, Radhika R." <RADHIKA.R.ROY at saic.com> wrote:          v\:* {behavior:url(#default#VML);} o\:* {behavior:url(#default#VML);} w\:* {behavior:url(#default#VML);} .shape {behavior:url(#default#VML);}           I agree with David.
   
  It is also NOT harm to linkage a particular design with a particular DHT, if this is the good choice for a given time. However, other proposals are also welcome. It is an open to all of us.
   
  Standardizations are always done throughout the world if there is a consensus to go for something that is good. With respect P2P-SIP, there are a lot of things to be solved. However, if something provides immediate value, as David and many others of this list have opined, what is the problem for standardization? A better standard in the future can always supersede an inferior standard of the past.
   
  Best regards,
  Radhika
   
      
---------------------------------
  
  From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David Barrett
 Sent: Monday, March 27, 2006 2:40 PM
 To: 'Mike Robinson'; henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
 Cc: p2p-sip at cs.columbia.edu
 Subject: Re: [p2p-sip] Some comments on Use-cases document
  
   
  I guess one view is the IETF should forever be relegated to the janitorial role of cleaning up after the party???s over.  But I believe the IETF can participate in and even promote the party through standardization.  I???d like the IETF to do more than simply formalize de facto standards, but actually create the standards that attract the party.  
   
  In the case of P2P-SIP, I entirely agree there are multiple, valid, proven ways to solve each of these problems ??? none of which is the clear victor.  And I also agree there are outstanding issues that have no proven solutions.
   
  If we pick amongst the best practices today, and skip the problems to which we have no clear solution, I personally believe that???ll give us the most immediate value, while paving the road for future improvements. I don???t believe we???ll ever meet any higher standard to get started than has already been met. 
   
  -david 
   
        
---------------------------------
  
  From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
 Sent: Monday, March 27, 2006 10:32 AM
 To: henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
 Cc: p2p-sip at cs.columbia.edu
 Subject: Re: [p2p-sip] Some comments on Use-cases document
  
   
     
  
    Henry:
  
     
  
    Do I need to tell that 4800 P2P google references have nothing to do with P2P SIP?
  
     
  
    Please let us not obfuscate the issues. You know, as well as, anyone sincere about the value of standards, that the pre-mature standardization leads to bad design as well as failure of the standards. As I mentioned there is no compelling use case presented yet that says we need to "standardize" p2p sip now without even bothering to get output from P2P IRTF group which is looking into core issues of P2P research.  
  
     
  
    Many would say that we first need an IETF standard for P2P file sharing before we want to use "P2P" to design P2P SIP. We need more research on P2P SIP before we standardize any particular "conspiracy" (I use conspiracy in the sense of Karst definition of linking design with a particular DHT)
  
     
  
    -mike
  
    V! OIP and SIP Trainer and Consultant
  
     
  
    
 Henry Sinnreich <henry at pulver.com> wrote:
  
      Mike Robinson writes:
  
     
  
    >How many research ! or analytical papers have been published in this area? 
  
      I have found by Googling a reference listing 4,800 papers on P2P. Hopefully, this enough to get us started. J
  
    Henry
  
  
    rates.
  
  
  

		
---------------------------------
Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls.  Great rates starting at 1&cent;/min.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060327/aa7807a4/attachment.html

From eunsoo at research.panasonic.com  Mon Mar 27 21:19:09 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Mon, 27 Mar 2006 21:19:09 -0500
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060328001415.97808.qmail@web38513.mail.mud.yahoo.com>
References: <20060328001415.97808.qmail@web38513.mail.mud.yahoo.com>
Message-ID: <44289D1D.7080605@research.panasonic.com>

Mike Robinson wrote:

>
> Radhika:
>
> You agree that there are many problems that are not resolved yet, and 
> no solution is available. That means more research needs to be done. 
> Research belongs in the labs, Universities or in IRTF.
>
The P2P technologies are continusouly evolving and still various issues 
remain open. There was a good presentation about DHT in the technical 
plenary of IETF65, which pointed out many security issues of DHTs. 
However, the conclusion of the presentation was not "don't use DHT" but 
rather "use DHT with caution" to be simplified in my understanding.
A P2P-based VoIP service provider has more subscribers than any other 
VoIP providers currently. Other commercial developments related to P2P 
SIP are being released via news media. I think this is a quite strong 
signal calling for P2P SIP standardization efforts. The number of people 
who have been showing interest in P2P SIP last several IETF meetings is 
another signal, I think.

Novertheless, in general, I guess it is a quite subjective matter 
whether a particular technology is ready for standardization or its 
standardization is worth the effort. But I guess you can agree that it 
is not a good idea waiting for all the problems to be solved before 
standardization of anything.
Many protocols or technologies have been evolving even after the first 
version of their standards. It does not mean standardization of those 
protocols was a mistake or worthless.
Considering all the time we spent so far on discussions about various 
apsects of P2P SIP, I think we have passed the stage to say "there are 
lots of problems unsolved". Would it be too much if I ask for more 
concrete discussions like what tools we can develop, what use case can 
be supported in which manner, and what is the difficulty for a certain 
use case?

> It is not clear how you can change a DHT and still be compatible since 
> SIP messages and headers fields are so much dependent on the type of 
> underlying DHT used.

I thought you attended the P2P SIP ad-hoc meeting in Dallas but maybe I 
was confused.
What you are describing above is based on a certain approach. Please 
check the presentation about a P2P SIP architecture made during the 
ad-hoc meeting in Dallas. The presentation is available in the IETF65 
agenda and presentation page. You can find a different approach that 
enables a modular design in which the underlying DHT or an overlay 
algorithm can be changed without affecting the SIP messages.

> The data maintained by each DHT is different and so are node insertion 
> and removal procedures. I believe the security issues of each DHT 
> types would also be different, so when someone says that they can 
> replace DHT later on, they are just being too simplistic.
>
> Do you really want to create OSPF versus IS-IS kind of two 
> incompatible solutions?
>
I think what Radhika meant was that different P2P SIP domains could use 
different DHTs. OSPF and IS-IS won't be together in the same domain but 
each can be used in different domains. Same approach for P2P SIP. Again 
please check the P2P SIP architecture presentation in the ad-hoc meeting.

I respect your concern regarding P2P SIP standardization. Such concern 
was expressed back in the first P2P SIP ad-hoc meeting in March last 
year. We have come a long way since then to have a proposed charter and 
the BOF in Dallas.

I hope we move forward.
Thanks.

Eunsoo
ps)I think David made really good points.

>
> */"Roy, Radhika R." <RADHIKA.R.ROY at saic.com>/* wrote:
>
>     I agree with David.
>     It is also NOT harm to link! age a particular design with a
>     particular DHT, if this is the good choice for a given time.
>     However, other proposals are also welcome. It is an open to all of us.
>     Standardizations are always done throughout the world if there is
>     a consensus to go for something that is good. With respect
>     P2P-SIP, there are a lot of things to be solved. However, if
>     something provides immediate value, as David and many others of
>     this list have opined, what is the problem for standardization? A
>     better standard in the future can always supersede an inferior
>     standard of the past.
>     Best regards,
>     Radhika
>     ------------------------------------------------------------------------
>     *From:* p2p-sip-bounces at cs.columbia.edu
>     [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *David Barrett
>     *Sent:* Monday, March 27, 2006 2:40 PM
>     *To:* 'Mike Robinson'; henry at pulver.com; 'Karst Bjorgson'; 'Kyara
>     Yamamoto'; 'Peter Pan'
>     *Cc:* p2p-sip at cs.columbia.edu
>     *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     I guess one view is the IETF should forever be relegated to the j!
>     anitorial role of cleaning up after the party???s over. But I
>     believe the IETF can participate in and even promote the party
>     through standardization. I???d like the IETF to do more than
>     simply formalize de facto standards, but actually create the
>     standards that attract the party.
>     In the case of P2P-SIP, I entirely agree there are multiple,
>     valid, proven ways to solve each of these problems ??? none of
>     which is the clear victor. And I also agree there are outstanding
>     issues that have no proven solutions.
>     If we pick amongst the best practices today, and skip the problems
>     to which we have no clear solution, I personally believe that???ll
>     give us the most immediate value, while paving the road for future
>     improvements. I don???t believe we???ll ever meet any higher
>     standard to get started than has already been met.
>     -david
>     ------------------------------------------------------------------------
>     *From:* p2p-sip-bounces at cs.columbia.edu
>     [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *Mike Robinson
>     *Sent:* Monday, March 27, 2006 10:32 AM
>     *To:* henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter
>     Pan'
>     *Cc:* p2p-sip at cs.columbia.edu
>     *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     Henry:
>     Do I need to tell that 48! 00 P2P google references have nothing
>     to do with P2P SIP?
>     Please let us not obfuscate the issues. You know, as well as,
>     anyone sincere about the value of standards, that the pre-mature
>     standardization leads to bad design as well as failure of the
>     standards. As I mentioned there is no compelling use case
>     presented yet that says we need to "standardize" p2p sip now
>     without even bothering to get output from P2P IRTF group which is
>     looking into core issues of P2P research.
>     Many would say that we first need an IETF standard for P2P file
>     sharing before we want to use "P2P" to design P2P SIP. We need
>     more research on P2P SIP before we standardize any particular
>     "conspiracy" (I use conspiracy in the sense of Karst definition of
>     linking design with a particular DHT)
>     -mike
>     V! OIP and SIP Trainer and Consultant
>
>     */Henry Sinnreich <henry at pulver.com>/* wrote:
>
>         Mike Robinson writes:
>>How many research ! or analytical papers have been published
>         in this area?
>         I have found by Googling a reference listing 4,800 papers on
>         P2P. Hopefully, this enough to get us started. J
>         Henry
>
>     rates.
>
>
> ------------------------------------------------------------------------
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great 
> rates starting at 1?/min. 
> <http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/evt=39666/*http://beta.messenger.yahoo.com> 
>
>
>
> <http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/evt=39666/*http://beta.messenger.yahoo.com>
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> <http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/evt=39666/*http://beta.messenger.yahoo.com>
>


From ksankar at cisco.com  Mon Mar 27 21:29:33 2006
From: ksankar at cisco.com (Krishna Sankar (ksankar))
Date: Mon, 27 Mar 2006 18:29:33 -0800
Subject: [p2p-sip] Some comments on Use-cases document <KS> Slightly
	long e-mail </KS>
Message-ID: <03CD385D379ECE4988B9C88834382EA701BB1DFD@xmb-sjc-22e.amer.cisco.com>

Folks,

Had a chance to read thru some of the P2P SIP material - during a long layoff at the airports ! Some thoughts, the observations are not linear. I will try to make it short & crisp (as much as I can ;o)) I am still internalizing the material and so pl bear with my ignorance.

a) There are two dimensions - scenarios/motivation and feature sets.

b) First we need to bound the trajectory and locus of P2P SIP - the use case document holistically mentions lots of ideas, but we need to crisply classify (and prioritize) them in terms of likely hood, short term pain points and essential feature set.

For example, if we are enabling hastily formed networks for first responders or communication in disaster scenario or many other similar situations, we might have a case for developing a P2P substrate. 

We can, then, opt for DHT, because it enables a much valued pragmatic need. DHT is mature enough to function at this level. The question is not whether a technology is mature in absolute terms, but is it good enough to enable a pain point we are facing today.

c) The next thing is to figure out the whys, whats and finally hows, of whatever we are enabling by the P2P SIP substrate.

d) Then we also need to see what feature set we need - like anonymity (for example leveraging primitives from tor) and traffic shaping (for resource constraint devices - may be need to look at sensor networks) Are we any way connected with domains like Ad-Hoc networks ? Do we need to embed primitives from the various usage patterns ?

The scenarios vs. feature set is very important for the mechanics and mechanisms we embed - using a P2P in normal internet would require strong anonymity primitives while in a hastily formed network might require bandwidth conservation and resource maximization with less chatty protocols.

e) In any case, our goal should be an orthogonally extensible framework not a tightly coupled set of messages, thus trying our best for flexibility in the face of future technological and algorithmic innovations

f) Are we (i.e. P2P SIP) in the control path or in the data path ? The requirement document has both and that puzzles me, because once a P2P substrate is in the data path, we need to figure out strong anonymity as well as traffic shaping ! People might be charged with $ for the bandwidth they are brokering. Then we will have to resort to BitTorrent-like reward mechanisms, fair bandwidth shaping et al. One idea is to have Tor-like super nodes.

g) As we read thru the documents, we also can see that we will need discovery of capabilities (because all nodes will not have all capabilities) and associated semantics. 

h) As it was mentioned in an earlier e-mail, Jabber, in some sense has a P2P SIP substrate. By the time we finish, we might resemble XMPP more than one ways (I am not an expert, am open to be enlightened here)

i) The choke of Chord seems a little arbitrary. Most probably the writers have done lots of background work, but it is not obvious from the documents. I think we need to first make sure we separate any DHT mechanism and then standardize the interfaces, protocols and the mechanics. Pluggable P2P is a good way to go.

j) Having said that Chord might be a good choice to start with, we really cannot be in the abstract-land forever !

k) First I thought the "approach to SIP registration .." only supports Chord, then I saw the dht-param. While I am at it, the document needs more thought in many areas including the concept of overlays (for example would we allow multiple overlays, are they hierarchical, why, or would we say a node can participate only in one overlay except for gateway nodes?), trust model, anonymity primitives and bandwidth shaping.

l) Back to the main feature, IMHO we should develop a P2P framework for a set of well bound, pragmatic, very relevant scenarios, and then have another document that implements the framework using Chord. The framework has to be very robust - our challenge would not be the protocol or message format, but the security, anonymity and related mechanics. 

m) I do think P2P *is* mature enough to be the work of IETF standardization, not all of it, but enough to solve some of the current problems. Especially if we base our work from experience in implementing P2P SIP. For example Section 15 of the "P2P approach ..." says there are several different implementations. Let us abstract what worked and what didn't.

n) We also should choose 2 or 3 P2P mechanisms and implement them to validate the potential framework.

Let me stop before I run out of alphabets for outline ...

Cheers & thanks for the patience to read this thru

<k/>
-----------------------------------------------------------------------------
       |            |        Krishna Sankar, Distinguished Engineer
      :|:          :|:          
     :|||:        :|||:      - Disruptive Technologies & Architectures
 ..:|||||||:.....:|||||||:.. - Security, Mobility & Integration
     Cisco  Systems Inc      - Mobile-Asymmetric-Composable-Embedded Networks
     (Ph) 408-853-8475       ksankar at cisco.com
-----------------------------------------------------------------------------
"Not all those who wander are lost ..." -J.R.R.Tolkien
-----------------------------------------------------------------------------
 

________________________________

	From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
	Sent: Monday, March 27, 2006 4:14 PM
	To: Roy, Radhika R.; p2p-sip at cs.columbia.edu
	Subject: Re: [p2p-sip] Some comments on Use-cases document
	
	

	Radhika:
	
	You agree that there are many problems  that are not resolved yet, and no solution is  available. That means more research needs to be done. Research belongs in the labs, Universities or in IRTF. 
	
	It is not clear how you can change a DHT and still be compatible since SIP messages and headers fields are so much dependent on the type of underlying DHT used. The data maintained by each DHT is different and so are node insertion and removal procedures. I believe the security issues of each DHT types would also be different, so when someone says that they can replace DHT later on, they are just being too simplistic.
	
	Do you really want to create OSPF versus IS-IS kind of two incompatible solutions?
	
	--mike
	
	
	"Roy, Radhika R." <RADHIKA.R.ROY at saic.com> wrote: 

		I agree with David.
				It is also NOT harm to link! age a particular design with a particular DHT, if this is the good choice for a given time. However, other proposals are also welcome. It is an open to all of us.
				Standardizations are always done throughout the world if there is a consensus to go for something that is good. With respect P2P-SIP, there are a lot of things to be solved. However, if something provides immediate value, as David and many others of this list have opined, what is the problem for standardization? A better standard in the future can always supersede an inferior standard of the past.
				Best regards,
		Radhika
				________________________________

				From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David Barrett
		Sent: Monday, March 27, 2006 2:40 PM
		To: 'Mike Robinson'; henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
		Cc: p2p-sip at cs.columbia.edu
		Subject: Re: [p2p-sip] Some comments on Use-cases document
				I guess one view is the IETF should forever be relegated to the j! anitorial role of cleaning up after the party???s over.  But I believe the IETF can participate in and even promote the party through standardization.  I???d like the IETF to do more than simply formalize de facto standards, but actually create the standards that attract the party.  
				In the case of P2P-SIP, I entirely agree there are multiple, valid, proven ways to solve each of these problems ??? none of which is the clear victor.  And I also agree there are outstanding issues that have no proven solutions.
				If we pick amongst the best practices today, and skip the problems to which we have no clear solution, I personally believe that???ll give us the most immediate value, while paving the road for future improvements. I don???t believe we???ll ever meet any higher standard to get started than has already been met. 
				-david 
				________________________________

				From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
		Sent: Monday, March 27, 2006 10:32 AM
		To: henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
		Cc: p2p-sip at cs.columbia.edu
		Subject: Re: [p2p-sip] Some comments on Use-cases document
						Henry:
				Do I need to tell that 48! 00 P2P google references have nothing to do with P2P SIP?
				Please let us not obfuscate the issues. You know, as well as, anyone sincere about the value of standards, that the pre-mature standardization leads to bad design as well as failure of the standards. As I mentioned there is no compelling use case presented yet that says we need to "standardize" p2p sip now without even bothering to get output from P2P IRTF group which is looking into core issues of P2P research.  
				Many would say that we first need an IETF standard for P2P file sharing before we want to use "P2P" to design P2P SIP. We need more research on P2P SIP before we standardize any particular "conspiracy" (I use conspiracy in the sense of Karst definition of linking design with a particular DHT)
				-mike
		V! OIP and SIP Trainer and Consultant
				
		Henry Sinnreich <henry at pulver.com> wrote:

			Mike Robinson writes:
						>How many research ! or analytical papers have been published in this area? 
			I have found by Googling a reference listing 4,800 papers on P2P. Hopefully, this enough to get us started. J
			Henry

		rates.


	________________________________

	Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great rates starting at 1?/min. <http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/evt=39666/*http://beta.messenger.yahoo.com> 



From edpimentl at gmail.com  Mon Mar 27 22:02:16 2006
From: edpimentl at gmail.com (Edpimentl)
Date: Mon, 27 Mar 2006 22:02:16 -0500
Subject: [p2p-sip] Some comments on Use-cases document <KS>
 Slightly	long e-mail </KS>
In-Reply-To: <03CD385D379ECE4988B9C88834382EA701BB1DFD@xmb-sjc-22e.amer.cisco.com>
References: <03CD385D379ECE4988B9C88834382EA701BB1DFD@xmb-sjc-22e.amer.cisco.com>
Message-ID: <4428A738.6000005@gmail.com>

Hello Et Al,

I do agree with Mike R. views, however it comes a time, when perfect is
not good enough.

Think about what would have happen if we waited for a blazing dual core
Intel processor instead of the 4bit zilog to go to the moon.

As Henry S. pointed out, today we have over 4k different papers/docs on P2P

As I understand a number of individuals/groups are beginning to
experiment with http://overlayweaver.sourceforge.net/
since it supports Chord, Kademlia, Pastry, Tapestry and OpenDHT

Net net, the pragmatic approach would be to agree on a framework that
would allow the introduction of  better P2P SIP infrastructure.

Respectfully,
Ed Pimentel

Mail:   edpimentl[at]gmail[dot]com
Web:    http://agileco.net
WebLog: http://edpimentel.com  
IM:     [AOL | Jabber | Yahoo | MSN ] edpimentl
Voip:   [SKype | AgileSIP | GoogleTalk | FWD(14272) ] edpimentl

http://GooGaYa.com (Multimodal Messaging, Presence, LBS, Social Network
and Context Aware Services)
http://ViaSIP.com

Krishna Sankar (ksankar) wrote:

>Folks,
>
>Had a chance to read thru some of the P2P SIP material - during a long layoff at the airports ! Some thoughts, the observations are not linear. I will try to make it short & crisp (as much as I can ;o)) I am still internalizing the material and so pl bear with my ignorance.
>
>a) There are two dimensions - scenarios/motivation and feature sets.
>
>b) First we need to bound the trajectory and locus of P2P SIP - the use case document holistically mentions lots of ideas, but we need to crisply classify (and prioritize) them in terms of likely hood, short term pain points and essential feature set.
>
>For example, if we are enabling hastily formed networks for first responders or communication in disaster scenario or many other similar situations, we might have a case for developing a P2P substrate. 
>
>We can, then, opt for DHT, because it enables a much valued pragmatic need. DHT is mature enough to function at this level. The question is not whether a technology is mature in absolute terms, but is it good enough to enable a pain point we are facing today.
>
>c) The next thing is to figure out the whys, whats and finally hows, of whatever we are enabling by the P2P SIP substrate.
>
>d) Then we also need to see what feature set we need - like anonymity (for example leveraging primitives from tor) and traffic shaping (for resource constraint devices - may be need to look at sensor networks) Are we any way connected with domains like Ad-Hoc networks ? Do we need to embed primitives from the various usage patterns ?
>
>The scenarios vs. feature set is very important for the mechanics and mechanisms we embed - using a P2P in normal internet would require strong anonymity primitives while in a hastily formed network might require bandwidth conservation and resource maximization with less chatty protocols.
>
>e) In any case, our goal should be an orthogonally extensible framework not a tightly coupled set of messages, thus trying our best for flexibility in the face of future technological and algorithmic innovations
>
>f) Are we (i.e. P2P SIP) in the control path or in the data path ? The requirement document has both and that puzzles me, because once a P2P substrate is in the data path, we need to figure out strong anonymity as well as traffic shaping ! People might be charged with $ for the bandwidth they are brokering. Then we will have to resort to BitTorrent-like reward mechanisms, fair bandwidth shaping et al. One idea is to have Tor-like super nodes.
>
>g) As we read thru the documents, we also can see that we will need discovery of capabilities (because all nodes will not have all capabilities) and associated semantics. 
>
>h) As it was mentioned in an earlier e-mail, Jabber, in some sense has a P2P SIP substrate. By the time we finish, we might resemble XMPP more than one ways (I am not an expert, am open to be enlightened here)
>
>i) The choke of Chord seems a little arbitrary. Most probably the writers have done lots of background work, but it is not obvious from the documents. I think we need to first make sure we separate any DHT mechanism and then standardize the interfaces, protocols and the mechanics. Pluggable P2P is a good way to go.
>
>j) Having said that Chord might be a good choice to start with, we really cannot be in the abstract-land forever !
>
>k) First I thought the "approach to SIP registration .." only supports Chord, then I saw the dht-param. While I am at it, the document needs more thought in many areas including the concept of overlays (for example would we allow multiple overlays, are they hierarchical, why, or would we say a node can participate only in one overlay except for gateway nodes?), trust model, anonymity primitives and bandwidth shaping.
>
>l) Back to the main feature, IMHO we should develop a P2P framework for a set of well bound, pragmatic, very relevant scenarios, and then have another document that implements the framework using Chord. The framework has to be very robust - our challenge would not be the protocol or message format, but the security, anonymity and related mechanics. 
>
>m) I do think P2P *is* mature enough to be the work of IETF standardization, not all of it, but enough to solve some of the current problems. Especially if we base our work from experience in implementing P2P SIP. For example Section 15 of the "P2P approach ..." says there are several different implementations. Let us abstract what worked and what didn't.
>
>n) We also should choose 2 or 3 P2P mechanisms and implement them to validate the potential framework.
>
>Let me stop before I run out of alphabets for outline ...
>
>Cheers & thanks for the patience to read this thru
>
><k/>
>-----------------------------------------------------------------------------
>       |            |        Krishna Sankar, Distinguished Engineer
>      :|:          :|:          
>     :|||:        :|||:      - Disruptive Technologies & Architectures
> ..:|||||||:.....:|||||||:.. - Security, Mobility & Integration
>     Cisco  Systems Inc      - Mobile-Asymmetric-Composable-Embedded Networks
>     (Ph) 408-853-8475       ksankar at cisco.com
>-----------------------------------------------------------------------------
>"Not all those who wander are lost ..." -J.R.R.Tolkien
>-----------------------------------------------------------------------------
> 
>
>________________________________
>
>	From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
>	Sent: Monday, March 27, 2006 4:14 PM
>	To: Roy, Radhika R.; p2p-sip at cs.columbia.edu
>	Subject: Re: [p2p-sip] Some comments on Use-cases document
>	
>	
>
>	Radhika:
>	
>	You agree that there are many problems  that are not resolved yet, and no solution is  available. That means more research needs to be done. Research belongs in the labs, Universities or in IRTF. 
>	
>	It is not clear how you can change a DHT and still be compatible since SIP messages and headers fields are so much dependent on the type of underlying DHT used. The data maintained by each DHT is different and so are node insertion and removal procedures. I believe the security issues of each DHT types would also be different, so when someone says that they can replace DHT later on, they are just being too simplistic.
>	
>	Do you really want to create OSPF versus IS-IS kind of two incompatible solutions?
>	
>	--mike
>	
>	
>	"Roy, Radhika R." <RADHIKA.R.ROY at saic.com> wrote: 
>
>		I agree with David.
>				It is also NOT harm to link! age a particular design with a particular DHT, if this is the good choice for a given time. However, other proposals are also welcome. It is an open to all of us.
>				Standardizations are always done throughout the world if there is a consensus to go for something that is good. With respect P2P-SIP, there are a lot of things to be solved. However, if something provides immediate value, as David and many others of this list have opined, what is the problem for standardization? A better standard in the future can always supersede an inferior standard of the past.
>				Best regards,
>		Radhika
>				________________________________
>
>				From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David Barrett
>		Sent: Monday, March 27, 2006 2:40 PM
>		To: 'Mike Robinson'; henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
>		Cc: p2p-sip at cs.columbia.edu
>		Subject: Re: [p2p-sip] Some comments on Use-cases document
>				I guess one view is the IETF should forever be relegated to the j! anitorial role of cleaning up after the party???s over.  But I believe the IETF can participate in and even promote the party through standardization.  I???d like the IETF to do more than simply formalize de facto standards, but actually create the standards that attract the party.  
>				In the case of P2P-SIP, I entirely agree there are multiple, valid, proven ways to solve each of these problems ??? none of which is the clear victor.  And I also agree there are outstanding issues that have no proven solutions.
>				If we pick amongst the best practices today, and skip the problems to which we have no clear solution, I personally believe that???ll give us the most immediate value, while paving the road for future improvements. I don???t believe we???ll ever meet any higher standard to get started than has already been met. 
>				-david 
>				________________________________
>
>				From: p2p-sip-bounces at cs.columbia.edu [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of Mike Robinson
>		Sent: Monday, March 27, 2006 10:32 AM
>		To: henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter Pan'
>		Cc: p2p-sip at cs.columbia.edu
>		Subject: Re: [p2p-sip] Some comments on Use-cases document
>						Henry:
>				Do I need to tell that 48! 00 P2P google references have nothing to do with P2P SIP?
>				Please let us not obfuscate the issues. You know, as well as, anyone sincere about the value of standards, that the pre-mature standardization leads to bad design as well as failure of the standards. As I mentioned there is no compelling use case presented yet that says we need to "standardize" p2p sip now without even bothering to get output from P2P IRTF group which is looking into core issues of P2P research.  
>				Many would say that we first need an IETF standard for P2P file sharing before we want to use "P2P" to design P2P SIP. We need more research on P2P SIP before we standardize any particular "conspiracy" (I use conspiracy in the sense of Karst definition of linking design with a particular DHT)
>				-mike
>		V! OIP and SIP Trainer and Consultant
>				
>		Henry Sinnreich <henry at pulver.com> wrote:
>
>			Mike Robinson writes:
>						>How many research ! or analytical papers have been published in this area? 
>			I have found by Googling a reference listing 4,800 papers on P2P. Hopefully, this enough to get us started. J
>			Henry
>
>		rates.
>
>
>	________________________________
>
>	Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great rates starting at 1?/min. <http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/evt=39666/*http://beta.messenger.yahoo.com> 
>
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From dbarrett at quinthar.com  Tue Mar 28 03:17:33 2006
From: dbarrett at quinthar.com (David Barrett)
Date: Tue, 28 Mar 2006 00:17:33 -0800
Subject: [p2p-sip] P2P-SIP-1.0
Message-ID: <200603280817.k2S8HbcK017561@cs.columbia.edu>

So decision #1 is whether to do something now, or go home and reconvene in
12 months.  I vote we do something now.

Decision #2 is whether we're going to expand SIP to support some list of P2P
features, or build a separate P2P system that SIP can leverage.  I think the
latter stands a higher chance of success.

Decision #3 is what comprises this list.  Decentralized SIP proxy is
certainly in.  I vote we do nothing more than that which is required to
create an overlay network with bindings for use as a decentralized SIP
proxy.

Decision #4 is whether to use a specific algorithm (Chord, Kademlia, etc),
or more abstract representation (ala overlayweaver) for the distributed SIP
proxy.  I vote we list the axes of comparison, rate each option, and then
pick one and run with it.


Thus my proposal is we evaluate DHT options, build a generalized DHT layer,
establish clear bindings whereby SIP clients can use this DHT as a proxy,
and then iterate.  This is the P2P-SIP-1.0 plan.  This presumes there will
be a 2.0 immediately following it, and a 3.0 immediately following that.  

What is your proposal?

-david


From enrico.marocco at telecomitalia.it  Tue Mar 28 04:13:12 2006
From: enrico.marocco at telecomitalia.it (Enrico Marocco)
Date: Tue, 28 Mar 2006 11:13:12 +0200
Subject: [p2p-sip] Some comments on Use-cases document <KS>
	Slightly	long e-mail </KS>
In-Reply-To: <03CD385D379ECE4988B9C88834382EA701BB1DFD@xmb-sjc-22e.amer.cisco.com>
References: <03CD385D379ECE4988B9C88834382EA701BB1DFD@xmb-sjc-22e.amer.cisco.com>
Message-ID: <4428FE28.4010903@telecomitalia.it>

Some comments about some thoughts:

> b) First we need to bound the trajectory and locus of P2P SIP

I really appreciate the proposed roadmap (8/06->usecases,
1/07->architecture, 7/07->proto); it allows a pretty pipeline and is
likely achievable.  Which issues do you feel have not been properly
addressed?

> d) Then we also need to see what feature set we need - like anonymity
> (for example leveraging primitives from tor) and traffic shaping (for
> resource constraint devices - may be need to look at sensor networks)

The main architectural idea is to allow different feature sets for
different networks.  Actually the requirements in different environments
are unlikely the same and, in many cases, need centralization (e.g. C/S
SIP) for being satisfied.  So far, the main consensus is on defining an
architecture focused on interconnection between possibly different P2P
and C/S islands; and obviously the interconnection must be the one
already defined for traditional SIP (RFC 3263).

> Are we any way connected with domains like Ad-Hoc networks ?

Sure!  It would be great to set up instantaneus gaming sessions without
needing access points and stuff like that.. and eventually I will be
able to join "Quake Parties" during my long airport waits ;-)

> h) As it was mentioned in an earlier e-mail, Jabber, in some sense has a
> P2P SIP substrate. By the time we finish, we might resemble XMPP more
> than one ways

Jabber has servers and clients strictly separated; it is as P2P as email
is.  OTOH SIP has enpoints assembling UACs and UASs and that's why it
could be considered /almost/ P2P.

> I think we need to first make sure we separate any DHT mechanism and
> then standardize the interfaces, protocols and the mechanics. Pluggable
> P2P is a good way to go.

This option as the remaining items on your mail hit the consensus of the
most part of the group.

Enrico

--------------------------------------------------------------------

CONFIDENTIALITY NOTICE

This message and its attachments are addressed solely to the persons above and may contain confidential information. If you have received the message in error, be informed that any use of the content hereof is prohibited. Please return it immediately to the sender and delete the message. Should you have any questions, please contact us by replying to webmaster at telecomitalia.it.

        Thank you

                                        www.telecomitalia.it

--------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060328/231805a6/attachment-0001.html

From RADHIKA.R.ROY at saic.com  Tue Mar 28 08:17:22 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Tue, 28 Mar 2006 08:17:22 -0500
Subject: [p2p-sip] P2P-SIP-1.0
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31E9691@0591-ITS-EXMP02.us.saic.com>

David:
 
I agree with your proposals, and these are the right set of things to do.
Let us do it now.
 
Best regards,
Radhika

  _____  

From: p2p-sip-bounces at cs.columbia.edu on behalf of David Barrett
Sent: Tue 3/28/2006 3:17 AM
To: p2p-sip at cs.columbia.edu
Subject: [p2p-sip] P2P-SIP-1.0



So decision #1 is whether to do something now, or go home and reconvene in 
12 months.  I vote we do something now. 

Decision #2 is whether we're going to expand SIP to support some list of P2P

features, or build a separate P2P system that SIP can leverage.  I think the

latter stands a higher chance of success. 

Decision #3 is what comprises this list.  Decentralized SIP proxy is 
certainly in.  I vote we do nothing more than that which is required to 
create an overlay network with bindings for use as a decentralized SIP 
proxy. 

Decision #4 is whether to use a specific algorithm (Chord, Kademlia, etc), 
or more abstract representation (ala overlayweaver) for the distributed SIP 
proxy.  I vote we list the axes of comparison, rate each option, and then 
pick one and run with it. 


Thus my proposal is we evaluate DHT options, build a generalized DHT layer, 
establish clear bindings whereby SIP clients can use this DHT as a proxy, 
and then iterate.  This is the P2P-SIP-1.0 plan.  This presumes there will 
be a 2.0 immediately following it, and a 3.0 immediately following that.  

What is your proposal? 

-david 

_______________________________________________ 
p2p-sip mailing list 
p2p-sip at cs.columbia.edu 
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
<https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip>  


From ksankar at cisco.com  Tue Mar 28 10:24:59 2006
From: ksankar at cisco.com (Krishna Sankar (ksankar))
Date: Tue, 28 Mar 2006 07:24:59 -0800
Subject: [p2p-sip] FW:  P2P-SIP-1.0
Message-ID: <03CD385D379ECE4988B9C88834382EA701BB1E69@xmb-sjc-22e.amer.cisco.com>

 
Agreed on all. 
#1, YES. #2, is fine. 
On #3, I assume you are saying we will be in the control path.
On #4, I think we should try the overlayweaver, at least try to abstract
the P2P mechanics  as much as possible.

Cheers
<k/> 

> -----Original Message-----
> From: p2p-sip-bounces at cs.columbia.edu 
> [mailto:p2p-sip-bounces at cs.columbia.edu] On Behalf Of David Barrett
> Sent: Tuesday, March 28, 2006 12:18 AM
> To: p2p-sip at cs.columbia.edu
> Subject: [p2p-sip] P2P-SIP-1.0
> 
> So decision #1 is whether to do something now, or go home and 
> reconvene in
> 12 months.  I vote we do something now.
> 
> Decision #2 is whether we're going to expand SIP to support some list 
> of P2P features, or build a separate P2P system that SIP can leverage.

> I think the latter stands a higher chance of success.
> 
> Decision #3 is what comprises this list.  Decentralized SIP proxy is 
> certainly in.  I vote we do nothing more than that which is required 
> to create an overlay network with bindings for use as a decentralized 
> SIP proxy.
> 
> Decision #4 is whether to use a specific algorithm (Chord, Kademlia, 
> etc), or more abstract representation (ala
> overlayweaver) for the distributed SIP proxy.  I vote we list the axes

> of comparison, rate each option, and then pick one and run with it.
> 
> 
> Thus my proposal is we evaluate DHT options, build a generalized DHT 
> layer, establish clear bindings whereby SIP clients can use this DHT 
> as a proxy, and then iterate.  This is the P2P-SIP-1.0 plan.  This 
> presumes there will be a 2.0 immediately following it, and a 3.0 
> immediately following that.
> 
> What is your proposal?
> 
> -david
> 
> _______________________________________________
> p2p-sip mailing list
> p2p-sip at cs.columbia.edu
> https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 


From ksankar at cisco.com  Tue Mar 28 10:41:29 2006
From: ksankar at cisco.com (Krishna Sankar (ksankar))
Date: Tue, 28 Mar 2006 07:41:29 -0800
Subject: [p2p-sip] Some comments on Use-cases document <KS>
	Slightly	long e-mail </KS>
Message-ID: <03CD385D379ECE4988B9C88834382EA701BB1E72@xmb-sjc-22e.amer.cisco.com>

Enrico,

	Thanks for taking time to reply and clarify the thoughts.

>> likely achievable. Which issues do you feel have not been properly
addressed?

	Re. the usecases, they are excellent as a goal, but we will have
to select and prioritize them. Also, as I had mentioned, we need a
feature set matrix. Let me take time to think through and may be add a
preliminary feature matrix.


>> architecture focused on interconnection between possibly different
P2P and C/S islands; and obviously the 
>> interconnection must be the one already defined for traditional SIP
(RFC 3263).

	The messages and protocols for the connection part are
relatively easier, but the rest (like anonymity and bandwidth shaping,
to name a few) would need lot more work. But if we bound the problem and
say (as David mentioned in his last e-mail) first we will attempt
control path primitives, we have a workable set. As the problem
simplifies to maintaining the various tables (which itself is not
trivial), we can at least add the required security, anonymity and other
primitives. Also we can attempt to make the mechanics and mechanisms
independent of any specific P2P domain.

Cheers
<k/>
________________________________

	From: Enrico Marocco [mailto:enrico.marocco at telecomitalia.it] 
	Sent: Tuesday, March 28, 2006 1:13 AM
	To: Krishna Sankar (ksankar)
	Cc: Mike Robinson; Roy, Radhika R.; p2p-sip at cs.columbia.edu
	Subject: Re: [p2p-sip] Some comments on Use-cases document <KS>
Slightly long e-mail </KS>
	
	
	 
	Some comments about some thoughts:
	
	> b) First we need to bound the trajectory and locus of P2P SIP
	
	I really appreciate the proposed roadmap (8/06->usecases,
	1/07->architecture, 7/07->proto); it allows a pretty pipeline
and is
	likely achievable. Which issues do you feel have not been
properly
	addressed?
	
	> d) Then we also need to see what feature set we need - like
anonymity
	> (for example leveraging primitives from tor) and traffic
shaping (for
	> resource constraint devices - may be need to look at sensor
networks)
	
	The main architectural idea is to allow different feature sets
for
	different networks. Actually the requirements in different
environments
	are unlikely the same and, in many cases, need centralization
(e.g. C/S
	SIP) for being satisfied. So far, the main consensus is on
defining an
	architecture focused on interconnection between possibly
different P2P
	and C/S islands; and obviously the interconnection must be the
one
	already defined for traditional SIP (RFC 3263).
	
	> Are we any way connected with domains like Ad-Hoc networks ?
	
	Sure! It would be great to set up instantaneus gaming sessions
without
	needing access points and stuff like that.. and eventually I
will be
	able to join "Quake Parties" during my long airport waits ;-)
	
	> h) As it was mentioned in an earlier e-mail, Jabber, in some
sense has a
	> P2P SIP substrate. By the time we finish, we might resemble
XMPP more
	> than one ways
	
	Jabber has servers and clients strictly separated; it is as P2P
as email
	is. OTOH SIP has enpoints assembling UACs and UASs and that's
why it
	could be considered /almost/ P2P.
	
	> I think we need to first make sure we separate any DHT
mechanism and
	> then standardize the interfaces, protocols and the mechanics.
Pluggable
	> P2P is a good way to go.
	
	This option as the remaining items on your mail hit the
consensus of the
	most part of the group.
	
	Enrico
	
	
	
--------------------------------------------------------------------
	CONFIDENTIALITY NOTICE
	This message and its attachments are addressed solely to the
persons
	above and may contain confidential information. If you have
received
	the message in error, be informed that any use of the content
hereof
	is prohibited. Please return it immediately to the sender and
delete
	the message. Should you have any questions, please contact us by
	replying to webmaster at telecomitalia.it
<mailto:webmaster at telecomitalia.it> .
	        Thank you
	                                        www.telecomitalia.it
<http://www.telecomitalia.it> 
	
--------------------------------------------------------------------


From gregd at research.panasonic.com  Tue Mar 28 11:07:38 2006
From: gregd at research.panasonic.com (Greg Daley)
Date: Tue, 28 Mar 2006 11:07:38 -0500
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060328001415.97808.qmail@web38513.mail.mud.yahoo.com>
References: <20060328001415.97808.qmail@web38513.mail.mud.yahoo.com>
Message-ID: <44295F4A.50602@research.panasonic.com>

Hi Mike

Mike Robinson wrote:
> 
> Radhika:
> 
> You agree that there are many problems  that are not resolved yet, and 
> no solution is  available. That means more research needs to be done. 
> Research belongs in the labs, Universities or in IRTF.

Thank goodness Universities are partnering with industry then,
to focus the effort on practical solutions ;)

I actually differ on this point, in that I do think there
are valid prototypes, as well as examples of structured
overlay network deployments which meet quite a few of our
needs.

> It is not clear how you can change a DHT and still be compatible since 
> SIP messages and headers fields are so much dependent on the type of 
> underlying DHT used. The data maintained by each DHT is different and so 
> are node insertion and removal procedures. I believe the security issues 
> of each DHT types would also be different, so when someone says that 
> they can replace DHT later on, they are just being too simplistic.

I think you're assuming a particular model here.  If the SIP
messages are the container for the peer-to-peer messages, then
you may be right.

Replacing the overlay algorithm isn't trivial, but neither should
it be the focus of all our efforts (ensuring that the hooks are there
is more likely to be useful).

> Do you really want to create OSPF versus IS-IS kind of two incompatible 
> solutions?

Please be aware that the incompatibility you mention is _not_ a problem,
as it's firmly tied to administrative boundaries.  Therefore,
within the domain, it is an administrative decision as to which protocol
to deploy, that best suits the situation.  Forwarding between domains
isn't affected.

At this stage, draft-shim-sipping-p2p-arch-00.txt proposes that
we use CS/SIP to cross admin domain boundaries. Perhaps this is
equivalent to the differentiation seen in the interior
routing/exterior routing problem?

What we may want to be able to do with peer-to-peer functions
in the future is create the kind of distinction found between
OSPF and BGP (rather than OSPF and ISIS).  Please remember we
started with RIP, and only had to tweak the forwarding algorithms
later to do VLSM.

Greg
P2P-SIP conspiracy member.

From eunsoo at research.panasonic.com  Tue Mar 28 11:09:55 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Tue, 28 Mar 2006 11:09:55 -0500
Subject: [p2p-sip] Some comments on Use-cases document
	<KS>	Slightly	longe-mail </KS>
In-Reply-To: <03CD385D379ECE4988B9C88834382EA701BB1E72@xmb-sjc-22e.amer.cisc
	o.com>
References: <03CD385D379ECE4988B9C88834382EA701BB1E72@xmb-sjc-22e.amer.cisco
	.com>
Message-ID: <44295FD3.3030308@research.panasonic.com>

Krishna Sankar (ksankar) wrote:

>Enrico,
>
>	Thanks for taking time to reply and clarify the thoughts.
>
>  
>
>>>likely achievable. Which issues do you feel have not been properly
>>>      
>>>
>addressed?
>
>	Re. the usecases, they are excellent as a goal, but we will have
>to select and prioritize them. Also, as I had mentioned, we need a
>feature set matrix. Let me take time to think through and may be add a
>preliminary feature matrix.
>
>  
>
I think we should distinguish the use cases in the use cases draft and 
the use cases listed in the proposed charter.
The use cases draft was intended to collect all possible use cases 
rather than to identify the immediate target use cases.
For example, the use cases draft talks about anonymous communication but 
the proposed charter puts anonymous communication as a possible future 
work item.
The use cases draft will help us in understanding what other uses are 
possible while we design the protocol with a limited set of immediate 
target use cases in mind.
Now we should focus on the use cases listed in the charter because those 
are what were proposed as the immediate target use cases.
David picked one of the three use cases in the proposed charter.
It seems to me that the three use cases are all reasonably solvable and 
we can identify and design tools for them.

>  
>
>>>architecture focused on interconnection between possibly different
>>>      
>>>
>P2P and C/S islands; and obviously the 
>  
>
>>>interconnection must be the one already defined for traditional SIP
>>>      
>>>
>(RFC 3263).
>
>	The messages and protocols for the connection part are
>relatively easier, but the rest (like anonymity and bandwidth shaping,
>to name a few) would need lot more work. But if we bound the problem and
>say (as David mentioned in his last e-mail) first we will attempt
>control path primitives, we have a workable set. As the problem
>simplifies to maintaining the various tables (which itself is not
>trivial), we can at least add the required security, anonymity and other
>primitives. Also we can attempt to make the mechanics and mechanisms
>independent of any specific P2P domain.
>
>
>
>  
>
Thanks.

Eunsoo

>________________________________
>
>	From: Enrico Marocco [mailto:enrico.marocco at telecomitalia.it] 
>	Sent: Tuesday, March 28, 2006 1:13 AM
>	To: Krishna Sankar (ksankar)
>	Cc: Mike Robinson; Roy, Radhika R.; p2p-sip at cs.columbia.edu
>	Subject: Re: [p2p-sip] Some comments on Use-cases document <KS>
>Slightly long e-mail </KS>
>	
>	
>	 
>	Some comments about some thoughts:
>	
>	> b) First we need to bound the trajectory and locus of P2P SIP
>	
>	I really appreciate the proposed roadmap (8/06->usecases,
>	1/07->architecture, 7/07->proto); it allows a pretty pipeline
>and is
>	likely achievable. Which issues do you feel have not been
>properly
>	addressed?
>	
>	> d) Then we also need to see what feature set we need - like
>anonymity
>	> (for example leveraging primitives from tor) and traffic
>shaping (for
>	> resource constraint devices - may be need to look at sensor
>networks)
>	
>	The main architectural idea is to allow different feature sets
>for
>	different networks. Actually the requirements in different
>environments
>	are unlikely the same and, in many cases, need centralization
>(e.g. C/S
>	SIP) for being satisfied. So far, the main consensus is on
>defining an
>	architecture focused on interconnection between possibly
>different P2P
>	and C/S islands; and obviously the interconnection must be the
>one
>	already defined for traditional SIP (RFC 3263).
>	
>	> Are we any way connected with domains like Ad-Hoc networks ?
>	
>	Sure! It would be great to set up instantaneus gaming sessions
>without
>	needing access points and stuff like that.. and eventually I
>will be
>	able to join "Quake Parties" during my long airport waits ;-)
>	
>	> h) As it was mentioned in an earlier e-mail, Jabber, in some
>sense has a
>	> P2P SIP substrate. By the time we finish, we might resemble
>XMPP more
>	> than one ways
>	
>	Jabber has servers and clients strictly separated; it is as P2P
>as email
>	is. OTOH SIP has enpoints assembling UACs and UASs and that's
>why it
>	could be considered /almost/ P2P.
>	
>	> I think we need to first make sure we separate any DHT
>mechanism and
>	> then standardize the interfaces, protocols and the mechanics.
>Pluggable
>	> P2P is a good way to go.
>	
>	This option as the remaining items on your mail hit the
>consensus of the
>	most part of the group.
>	
>	Enrico
>	
>	
>	
>--------------------------------------------------------------------
>	CONFIDENTIALITY NOTICE
>	This message and its attachments are addressed solely to the
>persons
>	above and may contain confidential information. If you have
>received
>	the message in error, be informed that any use of the content
>hereof
>	is prohibited. Please return it immediately to the sender and
>delete
>	the message. Should you have any questions, please contact us by
>	replying to webmaster at telecomitalia.it
><mailto:webmaster at telecomitalia.it> .
>	        Thank you
>	                                        www.telecomitalia.it
><http://www.telecomitalia.it> 
>	
>--------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>  
>


From ksankar at cisco.com  Tue Mar 28 11:33:48 2006
From: ksankar at cisco.com (Krishna Sankar (ksankar))
Date: Tue, 28 Mar 2006 08:33:48 -0800
Subject: [p2p-sip] Some comments on Use-cases document <KS>Slightly long
	e-mail </KS>
Message-ID: <03CD385D379ECE4988B9C88834382EA701BB1EA6@xmb-sjc-22e.amer.cisco.com>

Yep, makes sense. 

Just as a clarification, by anonymity, I meant the security mechanisms
against eavesdropping, mangling and other malicious schemes as well as
defense against profiling and traffic pattern hatching. Basically
anonymity to the P2P SIP substrate. In fact many of these approximate to
full anonymity anyway. So we will need these even in the first go
around.  

Cheers
<k/>

> -----Original Message-----
> From: Eunsoo Shim [mailto:eunsoo at research.panasonic.com] 
> Sent: Tuesday, March 28, 2006 8:10 AM
> To: Krishna Sankar (ksankar)
> Cc: Enrico Marocco; p2p-sip at cs.columbia.edu; Roy, Radhika R.
> Subject: Re: [p2p-sip] Some comments on Use-cases document 
> <KS> Slightly longe-mail </KS>
> 
> Krishna Sankar (ksankar) wrote:
> 
> >Enrico,
> >
> >	Thanks for taking time to reply and clarify the thoughts.
> >
> >  
> >
> >>>likely achievable. Which issues do you feel have not been properly
> >>>      
> >>>
> >addressed?
> >
> >	Re. the usecases, they are excellent as a goal, but we 
> will have to 
> >select and prioritize them. Also, as I had mentioned, we 
> need a feature 
> >set matrix. Let me take time to think through and may be add a 
> >preliminary feature matrix.
> >
> >  
> >
> I think we should distinguish the use cases in the use cases 
> draft and the use cases listed in the proposed charter.
> The use cases draft was intended to collect all possible use 
> cases rather than to identify the immediate target use cases.
> For example, the use cases draft talks about anonymous 
> communication but the proposed charter puts anonymous 
> communication as a possible future work item.
> The use cases draft will help us in understanding what other 
> uses are possible while we design the protocol with a limited 
> set of immediate target use cases in mind.
> Now we should focus on the use cases listed in the charter 
> because those are what were proposed as the immediate target 
> use cases.
> David picked one of the three use cases in the proposed charter.
> It seems to me that the three use cases are all reasonably 
> solvable and we can identify and design tools for them.
> 
> >  
> >
> >>>architecture focused on interconnection between possibly different
> >>>      
> >>>
> >P2P and C/S islands; and obviously the
> >  
> >
> >>>interconnection must be the one already defined for traditional SIP
> >>>      
> >>>
> >(RFC 3263).
> >
> >	The messages and protocols for the connection part are 
> relatively 
> >easier, but the rest (like anonymity and bandwidth shaping, 
> to name a 
> >few) would need lot more work. But if we bound the problem 
> and say (as 
> >David mentioned in his last e-mail) first we will attempt 
> control path 
> >primitives, we have a workable set. As the problem simplifies to 
> >maintaining the various tables (which itself is not 
> trivial), we can at 
> >least add the required security, anonymity and other 
> primitives. Also 
> >we can attempt to make the mechanics and mechanisms 
> independent of any 
> >specific P2P domain.
> >
> >
> >
> >  
> >
> Thanks.
> 
> Eunsoo
> 
> >________________________________
> >
> >	From: Enrico Marocco [mailto:enrico.marocco at telecomitalia.it] 
> >	Sent: Tuesday, March 28, 2006 1:13 AM
> >	To: Krishna Sankar (ksankar)
> >	Cc: Mike Robinson; Roy, Radhika R.; p2p-sip at cs.columbia.edu
> >	Subject: Re: [p2p-sip] Some comments on Use-cases document <KS> 
> >Slightly long e-mail </KS>
> >	
> >	
> >	 
> >	Some comments about some thoughts:
> >	
> >	> b) First we need to bound the trajectory and locus of P2P SIP
> >	
> >	I really appreciate the proposed roadmap (8/06->usecases,
> >	1/07->architecture, 7/07->proto); it allows a pretty 
> pipeline and is
> >	likely achievable. Which issues do you feel have not 
> been properly
> >	addressed?
> >	
> >	> d) Then we also need to see what feature set we need - like 
> >anonymity
> >	> (for example leveraging primitives from tor) and 
> traffic shaping 
> >(for
> >	> resource constraint devices - may be need to look at sensor
> >networks)
> >	
> >	The main architectural idea is to allow different 
> feature sets for
> >	different networks. Actually the requirements in different 
> >environments
> >	are unlikely the same and, in many cases, need 
> centralization (e.g. 
> >C/S
> >	SIP) for being satisfied. So far, the main consensus is 
> on defining an
> >	architecture focused on interconnection between 
> possibly different P2P
> >	and C/S islands; and obviously the interconnection must 
> be the one
> >	already defined for traditional SIP (RFC 3263).
> >	
> >	> Are we any way connected with domains like Ad-Hoc networks ?
> >	
> >	Sure! It would be great to set up instantaneus gaming 
> sessions without
> >	needing access points and stuff like that.. and 
> eventually I will be
> >	able to join "Quake Parties" during my long airport waits ;-)
> >	
> >	> h) As it was mentioned in an earlier e-mail, Jabber, 
> in some sense 
> >has a
> >	> P2P SIP substrate. By the time we finish, we might 
> resemble XMPP 
> >more
> >	> than one ways
> >	
> >	Jabber has servers and clients strictly separated; it 
> is as P2P as 
> >email
> >	is. OTOH SIP has enpoints assembling UACs and UASs and 
> that's why it
> >	could be considered /almost/ P2P.
> >	
> >	> I think we need to first make sure we separate any 
> DHT mechanism and
> >	> then standardize the interfaces, protocols and the mechanics.
> >Pluggable
> >	> P2P is a good way to go.
> >	
> >	This option as the remaining items on your mail hit the 
> consensus of 
> >the
> >	most part of the group.
> >	
> >	Enrico
> >	
> >	
> >	
> >--------------------------------------------------------------------
> >	CONFIDENTIALITY NOTICE
> >	This message and its attachments are addressed solely 
> to the persons
> >	above and may contain confidential information. If you 
> have received
> >	the message in error, be informed that any use of the 
> content hereof
> >	is prohibited. Please return it immediately to the 
> sender and delete
> >	the message. Should you have any questions, please contact us by
> >	replying to webmaster at telecomitalia.it 
> ><mailto:webmaster at telecomitalia.it> .
> >	        Thank you
> >	                                        www.telecomitalia.it 
> ><http://www.telecomitalia.it>
> >	
> >--------------------------------------------------------------------
> >
> >_______________________________________________
> >p2p-sip mailing list
> >p2p-sip at cs.columbia.edu
> >https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> >  
> >
> 


From ksankar at cisco.com  Wed Mar 29 22:32:25 2006
From: ksankar at cisco.com (Krishna Sankar (ksankar))
Date: Wed, 29 Mar 2006 19:32:25 -0800
Subject: [p2p-sip] Questions on #5 "A P2P SIP Arch - 2 layer approach".
Message-ID: <03CD385D379ECE4988B9C88834382EA701C3882A@xmb-sjc-22e.amer.cisco.com>

Folks,

	Have some quick questions on slide #5 "A P2P SIP Arch - 2 layer
approach". I assume:

	a)	This is could be a network topology, definitely not a
logical topology.
	b)	ONs & SNs are all SIP end-points and will be hashed as
per Chord algorithm.
	c)	So even though there could be a visibility and
reachability association between specific SNs and ONs, the P2P substrate
will not reflect this, as the hash could place them in different areas
in the ring. In fact, if the hash didn't, it will degenerate to a
routing table (not that that is bad, but that is not what we want to
achieve at a macro level)
	d)	Hopefully this diagram is not overloaded with NAT
traversal as well. In that case, the SN will represent the ONs which
hang off of it. In this case the hashing will happen on the ON name but
the IP address will be that of the associated SN. The SN knows how to
get to the Ones it represents.
	e)	Extending a) further, the Chord ring representation of
this diagram will be very different.

Cheers
<k/>


From mike_robnson at yahoo.com  Thu Mar 30 00:18:09 2006
From: mike_robnson at yahoo.com (Mike Robinson)
Date: Wed, 29 Mar 2006 21:18:09 -0800 (PST)
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <44289D1D.7080605@research.panasonic.com>
Message-ID: <20060330051809.17959.qmail@web38502.mail.mud.yahoo.com>


Eunsoo wrote

>A P2P-based VoIP service provider has more subscribers than any other 
> VoIP providers currently. 

>Other commercial developments related to P2P SIP are being released via > news media.

What does it really prove?  Let us see.

Unfortunately, you are  equating Skype  VoIP over  P2P with your version of p2psip. "P2P VOIP (as in skype) works" is neither a proof nor a validation of your p2p sip design since skype is not based on p2p sip. Besides,   it  is not a pure p2p solution in a strict sense. Skype documentation also lists a number of optimizations that go far beyond signaling a session to get the better voice quality. Please see their site for more details.

Actually, without having some support from an infrastructure provider (or always connected nodes in the public domain) it may not be even possible to run a pure p2p voip system beyond your organization. You have issues of NAT traversal, authentication and  certification management  that cannot be simply washed away. Both Krishna Sankar and Karst raised a number of technical issues.  

May be what IETF needs a VOIP over P2P forum, rather than P2P SIP group.
In the p2p sip, all you are using p2p overlay is for resolving URI into end point 
IP addresses.  That can be done in many ways, and SIP can be used in the unmodified way as it is. It may be possible that you don't even need to tinker much with SIP (johnson's draft discusses one such possibility). 

In addition, using  p2p overlay to resolve URIs does not solve  voice quality issues, and I think Krishna Sankar articulated those issues very well. So, if you believe in skype, that means that you need voip over p2p group that looks into broader issues than the proposed p2p sip group.

--mike




Eunsoo Shim <eunsoo at research.panasonic.com> wrote: Mike Robinson wrote:

>
> Radhika:
>
> You agree that there are many problems that are not resolved yet, and 
> no solution is available. That means more research needs to be done. 
> Research belongs in the labs, Universities or in IRTF.
>
The P2P technologies are continusouly evolving and still various issues 
remain open. There was a good presentation about DHT in the technical 
plenary of IETF65, which pointed out many security issues of DHTs. 
However, the conclusion of the presentation was not "don't use DHT" but 
rather "use DHT with caution" to be simplified in my understanding.
A P2P-based VoIP service provider has more subscribers than any other 
VoIP providers currently. Other commercial developments related to P2P 
SIP are being released via news media. I think this is a quite strong 
signal calling for P2P SIP standardization efforts. The number of people 
who have been showing interest in P2P SIP last several IETF meetings is 
another signal, I think.

Novertheless, in general, I guess it is a quite subjective matter 
whether a particular technology is ready for standardization or its 
standardization is worth the effort. But I guess you can agree that it 
is not a good idea waiting for all the problems to be solved before 
standardization of anything.
Many protocols or technologies have been evolving even after the first 
version of their standards. It does not mean standardization of those 
protocols was a mistake or worthless.
Considering all the time we spent so far on discussions about various 
apsects of P2P SIP, I think we have passed the stage to say "there are 
lots of problems unsolved". Would it be too much if I ask for more 
concrete discussions like what tools we can develop, what use case can 
be supported in which manner, and what is the difficulty for a certain 
use case?

> It is not clear how you can change a DHT and still be compatible since 
> SIP messages and headers fields are so much dependent on the type of 
> underlying DHT used.

I thought you attended the P2P SIP ad-hoc meeting in Dallas but maybe I 
was confused.
What you are describing above is based on a certain approach. Please 
check the presentation about a P2P SIP architecture made during the 
ad-hoc meeting in Dallas. The presentation is available in the IETF65 
agenda and presentation page. You can find a different approach that 
enables a modular design in which the underlying DHT or an overlay 
algorithm can be changed without affecting the SIP messages.

> The data maintained by each DHT is different and so are node insertion 
> and removal procedures. I believe the security issues of each DHT 
> types would also be different, so when someone says that they can 
> replace DHT later on, they are just being too simplistic.
>
> Do you really want to create OSPF versus IS-IS kind of two 
> incompatible solutions?
>
I think what Radhika meant was that different P2P SIP domains could use 
different DHTs. OSPF and IS-IS won't be together in the same domain but 
each can be used in different domains. Same approach for P2P SIP. Again 
please check the P2P SIP architecture presentation in the ad-hoc meeting.

I respect your concern regarding P2P SIP standardization. Such concern 
was expressed back in the first P2P SIP ad-hoc meeting in March last 
year. We have come a long way since then to have a proposed charter and 
the BOF in Dallas.

I hope we move forward.
Thanks.

Eunsoo
ps)I think David made really good points.

>
> */"Roy, Radhika R." /* wrote:
>
>     I agree with David.
>     It is also NOT harm to link! age a particular design with a
>     particular DHT, if this is the good choice for a given time.
>     However, other proposals are also welcome. It is an open to all of us.
>     Standardizations are always done throughout the world if there is
>     a consensus to go for something that is good. With respect
>     P2P-SIP, there are a lot of things to be solved. However, if
>     something provides immediate value, as David and many others of
>     this list have opined, what is the problem for standardization? A
>     better standard in the future can always supersede an inferior
>     standard of the past.
>     Best regards,
>     Radhika
>     ------------------------------------------------------------------------
>     *From:* p2p-sip-bounces at cs.columbia.edu
>     [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *David Barrett
>     *Sent:* Monday, March 27, 2006 2:40 PM
>     *To:* 'Mike Robinson'; henry at pulver.com; 'Karst Bjorgson'; 'Kyara
>     Yamamoto'; 'Peter Pan'
>     *Cc:* p2p-sip at cs.columbia.edu
>     *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     I guess one view is the IETF should forever be relegated to the j!
>     anitorial role of cleaning up after the party???s over. But I
>     believe the IETF can participate in and even promote the party
>     through standardization. I???d like the IETF to do more than
>     simply formalize de facto standards, but actually create the
>     standards that attract the party.
>     In the case of P2P-SIP, I entirely agree there are multiple,
>     valid, proven ways to solve each of these problems ??? none of
>     which is the clear victor. And I also agree there are outstanding
>     issues that have no proven solutions.
>     If we pick amongst the best practices today, and skip the problems
>     to which we have no clear solution, I personally believe that???ll
>     give us the most immediate value, while paving the road for future
>     improvements. I don???t believe we???ll ever meet any higher
>     standard to get started than has already been met.
>     -david
>     ------------------------------------------------------------------------
>     *From:* p2p-sip-bounces at cs.columbia.edu
>     [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *Mike Robinson
>     *Sent:* Monday, March 27, 2006 10:32 AM
>     *To:* henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter
>     Pan'
>     *Cc:* p2p-sip at cs.columbia.edu
>     *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     Henry:
>     Do I need to tell that 48! 00 P2P google references have nothing
>     to do with P2P SIP?
>     Please let us not obfuscate the issues. You know, as well as,
>     anyone sincere about the value of standards, that the pre-mature
>     standardization leads to bad design as well as failure of the
>     standards. As I mentioned there is no compelling use case
>     presented yet that says we need to "standardize" p2p sip now
>     without even bothering to get output from P2P IRTF group which is
>     looking into core issues of P2P research.
>     Many would say that we first need an IETF standard for P2P file
>     sharing before we want to use "P2P" to design P2P SIP. We need
>     more research on P2P SIP before we standardize any particular
>     "conspiracy" (I use conspiracy in the sense of Karst definition of
>     linking design with a particular DHT)
>     -mike
>     V! OIP and SIP Trainer and Consultant
>
>     */Henry Sinnreich /* wrote:
>
>         Mike Robinson writes:
>>How many research ! or analytical papers have been published
>         in this area?
>         I have found by Googling a reference listing 4,800 papers on
>         P2P. Hopefully, this enough to get us started. J
>         Henry
>
>     rates.
>
>
> ------------------------------------------------------------------------
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great 
> rates starting at 1???/min. 
>  
>
>
>
> 
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
> 
>



		
---------------------------------
Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls.  Great rates starting at 1&cent;/min.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060330/ac5466c5/attachment.html

From huang-ming.pan at comcast.net  Thu Mar 30 13:20:48 2006
From: huang-ming.pan at comcast.net (Peter Pan)
Date: Thu, 30 Mar 2006 10:20:48 -0800
Subject: [p2p-sip] Some comments on Use-cases document
References: <20060330051809.17959.qmail@web38502.mail.mud.yahoo.com>
Message-ID: <002001c65426$ac686110$6400a8c0@comcast.net>

it has been suffocating after the proposed decisions #1 - #4 :-(

very curious if anyone here ever assign himself with a simpler p2p project
like cloning dango skype in p2p-sip, image himself the only militia, and
walk his mind (&hands) thru entire design cycle toward commercializability?

maybe a down-to-earth approach can help uncover problems in current p2p-sip
direction.

egs,

* do sip retransmission scheme and backoff timers fit well in p2p arena?
(some p2p msgs need no retransmission)
* what if a sip/udp packet exceeds MTU with appended p2p info (eg,
finger/k-bucket entries)?
* can p2p-sip keep incremental paces (and "religion") after strong security
and distributed databases are added to the overlayS in later phases?
* even if it imaginably can, will it be imaginably efficient and
competitive? (h323 and Jabber fans criticize protocol efficiency of sip even
with sigcomp)
* autonomous certificate management...etc

me feel the p2p-sip topic definitely not SOSIMPLE...

peter
----- Original Message -----
From: Mike Robinson
To: Eunsoo Shim
Cc: p2p-sip at cs.columbia.edu
Sent: Wednesday, March 29, 2006 9:18 PM
Subject: Re: [p2p-sip] Some comments on Use-cases document



Eunsoo wrote

>A P2P-based VoIP service provider has more subscribers than any other
> VoIP providers currently.

>Other commercial developments related to P2P SIP are being released via >
news media.

What does it really prove?  Let us see.

Unfortunately, you are  equating Skype  VoIP over  P2P with your version of
p2psip. "P2P VOIP (as in skype) works" is neither a proof nor a validation
of your p2p sip design since skype is not based on p2p sip. Besides,   it
is not a pure p2p solution in a strict sense. Skype documentation also lists
a number of optimizations that go far beyond signaling a session to get the
better voice quality. Please see their site for more details.

Actually, without having some support from an infrastructure provider (or
always connected nodes in the public domain) it may not be even possible to
run a pure p2p voip system beyond your organization. You have issues! of NAT
traversal, authentication and  certification management  that cannot be
simply washed away. Both Krishna Sankar and Karst raised a number of
technical issues.

May be what IETF needs a VOIP over P2P forum, rather than P2P SIP group.
In the p2p sip, all you are using p2p overlay is for resolving URI into end
point
IP addresses.  That can be done in many ways, and SIP can be used in the
unmodified way as it is. It may be possible that you don't even need to
tinker much with SIP (johnson's draft discusses one such possibility).

In addition, using  p2p overlay to resolve URIs does not solve  voice
quality issues, and I think Krishna Sankar articulated those issues very
well. So, if you believe in skype, that means that you need voip over p2p
group that looks into broader issues than the proposed p2p sip group.

--mike




Eunsoo Shim <eunsoo at research.panasonic.com> wrote:
Mike Robinson wrote:

>
> Radhika:
>
> You agree that there are many problems that are not resolved yet, and
> no solution is available. That means more research needs to be done.
> Research belongs in the labs, Universities or in IRTF.
>
The P2P technologies are continusouly evolving and still various issues
remain open. There was a good presentation about DHT in the technical
plenary of IETF65, which pointed out many security issues of DHTs.
However, the conclusion of the presentation was not "don't use DHT" but
rather "use DHT with caution" to be simplified in my understanding.
A P2P-based VoIP service provider has more subscribers than any other
VoIP providers currently. Other commercial developments related to P2P
SIP are being released via news media. I think this is a quite strong
signal calling for P2! P SIP standardization efforts. The number of people
who have been showing interest in P2P SIP last several IETF meetings is
another signal, I think.

Novertheless, in general, I guess it is a quite subjective matter
whether a particular technology is ready for standardization or its
standardization is worth the effort. But I guess you can agree that it
is not a good idea waiting for all the problems to be solved before
standardization of anything.
Many protocols or technologies have been evolving even after the first
version of their standards. It does not mean standardization of those
protocols was a mistake or worthless.
Considering all the time we spent so far on discussions about various
apsects of P2P SIP, I think we have passed the stage to say "there are
lots of problems unsolved". Would it be too much if I ask for more
concrete discussions like what tools we can develop, what use case can
be supported in which ma! nner, and what is the difficulty for a certain
use case?

> It is not clear how you can change a DHT and still be compatible since
> SIP messages and headers fields are so much dependent on the type of
> underlying DHT used.

I thought you attended the P2P SIP ad-hoc meeting in Dallas but maybe I
was confused.
What you are describing above is based on a certain approach. Please
check the presentation about a P2P SIP architecture made during the
ad-hoc meeting in Dallas. The presentation is available in the IETF65
agenda and presentation page. You can find a different approach that
enables a modular design in which the underlying DHT or an overlay
algorithm can be changed without affecting the SIP messages.

> The data maintained by each DHT is different and so are node insertion
> and removal procedures. I believe the security issues of each DHT
> types would also be different, so when someone says tha! t they can
> replace DHT later on, they are just being too simplistic.
>
> Do you really want to create OSPF versus IS-IS kind of two
> incompatible solutions?
>
I think what Radhika meant was that different P2P SIP domains could use
different DHTs. OSPF and IS-IS won't be together in the same domain but
each can be used in different domains. Same approach for P2P SIP. Again
please check the P2P SIP architecture presentation in the ad-hoc meeting.

I respect your concern regarding P2P SIP standardization. Such concern
was expressed back in the first P2P SIP ad-hoc meeting in March last
year. We have come a long way since then to have a proposed charter and
the BOF in Dallas.

I hope we move forward.
Thanks.

Eunsoo
ps)I think David made really good points.

>
> */"Roy, Radhika R." /* wrote:
>
> I agree with David.
> It is also NOT harm! to link! age a particular design with a
> particular DHT, if this is the good choice for a given time.
> However, other proposals are also welcome. It is an open to all of us.
> Standardizations are always done throughout the world if there is
> a consensus to go for something that is good. With respect
> P2P-SIP, there are a lot of things to be solved. However, if
> something provides immediate value, as David and many others of
> this list have opined, what is the problem for standardization? A
> better standard in the future can always supersede an inferior
> standard of the past.
> Best regards,
> Radhika
> ------------------------------------------------------------------------
> *From:* p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *David Barrett
> *Sent:* Monday, March 27, 2006 2:40 PM> *To:* 'Mike Robinson';
henry at pulver.com; 'Karst Bjorgson'; 'Kyara
> Yamamoto'; 'Peter Pan'
> *Cc:* p2p-sip at cs.columbia.edu
> *Subject:* Re: [p2p-sip] Some comments on Use-cases document
> I guess one view is the IETF should forever be relegated to the j!
> anitorial role of cleaning up after the party???s over. But I
> believe the IETF can participate in and even promote the party
> through standardization. I???d like the IETF to do more than
> simply formalize de facto standards, but actually create the
> standards that attract the party.
> In the case of P2P-SIP, I entirely agree there are multiple,
> valid, proven ways to solve each of these problems ??? none of
> which is the clear victor. And I also agree there are outstanding
> issues that have no proven solutions.
> If we pick amongst the best practices today, and skip the problems
> to which we have no clear solution, I personally believe that???ll
> give us the most immediate value, while paving the road for future
> improvements. I don???t believe we???ll ever meet any higher
> standard to get started than has already been met.
> -david
> ------------------------------------------------------------------------
> *From:* p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *Mike Robinson
> *Sent:* Monday, March 27, 2006 10:32 AM
> *To:* henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter
> Pan'
> *Cc:* p2p-sip at cs.columbia.edu
> *Subject:* Re: [p2p-sip] Some comments on Use-cases document
> Henry:
> Do I need to tell that 48! 00 P2P google references have nothing
> to do with P2P SIP?
> Please let us not obfuscate the issues. You ! know, as well as,
> anyone sincere about the value of standards, that the pre-mature
> standardization leads to bad design as well as failure of the
> standards. As I mentioned there is no compelling use case
> presented yet that says we need to "standardize" p2p sip now
> without even bothering to get output from P2P IRTF group which is
> looking into core issues of P2P research.
> Many would say that we first need an IETF standard for P2P file
> sharing before we want to use "P2P" to design P2P SIP. We need
> more research on P2P SIP before we standardize any particular
> "conspiracy" (I use conspiracy in the sense of Karst definition of
> linking design with a particular DHT)
> -mike
> V! OIP and SIP Trainer and Consultant
>
> */Henry Sinnreich /* wrote:
>
> Mike Robinson writes:
>>How many! research ! or analytical papers have been published
> in this area?
> I have found by Googling a reference listing 4,800 papers on
> P2P. Hopefully, this enough to get us started. J
> Henry
>
> rates.
>
>
> ------------------------------------------------------------------------
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great
> rates starting at 1???/min.
>
>
>
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>






Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great rates
starting at 1?/min.


From eunsoo at research.panasonic.com  Thu Mar 30 16:28:56 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Thu, 30 Mar 2006 16:28:56 -0500
Subject: [p2p-sip] Some comments on Use-cases document
In-Reply-To: <20060330051809.17959.qmail@web38502.mail.mud.yahoo.com>
References: <20060330051809.17959.qmail@web38502.mail.mud.yahoo.com>
Message-ID: <442C4D98.6040904@research.panasonic.com>

Mike Robinson wrote:

>
> Eunsoo wrote
>
> >A P2P-based VoIP service provider has more subscribers than any other
> > VoIP providers currently.
>
> >Other commercial developments related to P2P SIP are being released 
> via > news media.
>
> What does it really prove? Let us see.
>
> Unfortunately, you are equating Skype VoIP over P2P with your version 
> of p2psip. "P2P VOIP (as in skype) works" is neither a proof nor a 
> validation of your p2p sip design since skype is not based on p2p sip. 
> Besides, it is not a pure p2p solution in a strict sense.

I guess you mean something without any kind of central entity by 'pure 
p2p solution in a strict sense'. If you review discussions in the past 
P2P SIP ad-hoc meetings and in the mailing list, you may find out that 
the group is quite open for assuming some limited central entities, in 
particular, for authentication. As the proposed charter suggests, many 
people think a distributed identity management is a hard problem and the 
group should not try to standardize a protocol for that w.r.t. P2P SIP.

> Skype documentation also lists a number of optimizations that go far 
> beyond signaling a session to get the better voice quality. Please see 
> their site for more details.
>
> Actually, without having some support from an infrastructure provider 
> (or always connected nodes in the public domain) it may not be even 
> possible to run a pure p2p voip system beyond your organization. You 
> have issues! o! f NAT traversal, authentication and certification 
> management that cannot be simply washed away. Both Krishna Sankar and 
> Karst raised a number of technical issues.

All those issues have been raised and discussed in the group from the 
beginning.

>
> May be what IETF needs a VOIP over P2P forum, rather than P2P SIP group.

I guess you mean a group to design a complete P2P VoIP system by 'a VoIP 
over P2P forum'. You may notice that IETF WGs do not design a complete 
system but design protocols that are components of systems or tools. It 
is (should be) applied to the P2P SIP group, too.

> In the p2p sip, all you are using p2p overlay is for resolving URI 
> into end point
> IP addresses.

You are close. Basically the P2P overlay provides a resource location 
service for SIP. The resource could be UA (or user), TURN server, and so 
on.

> That can be done in many ways, and SIP can be used in the unmodified 
> way as it is.

That's the approach a number of people proposed.

> It may be possible that you don't even need to tinker much with SIP 
> (johnson's draft discusses one such possibility).
>
> In addition, using p2p overlay to resolve URIs does not solve voice 
> quality issues,

Right. Neither SIP or P2P overlay protocol will solve voice quality 
issues. The group is not aimed to address voice quality issues.
For example, for voice quality, codec selection is important. The SDP 
can be used to notify the type of the codec. But you got to develop an 
algorithm to select a codec, which is not what an IETF WG does 
typically. You may need a better voice codec. It is not dealt with in 
the IETF, either. Please don't expect an IETF WG will address/do all 
those things.
This group is not supposed to address all kind of issues related to 
SIP-based P2P real-time media applications. The group should develop a 
few tools/components by which people can build a system.

> and I think Krishna Sankar articulated those issues very well. So, if 
> you believe in skype, that means that you need voip over p2p group 
> that looks into broader issues than the proposed p2p sip group.
>
Again, I understand there are broader issues than what is covered by the 
proposed charter. Actually that is the purpose of the charter --- 
narrowing down and defining a limited scope/deliverables.

Thanks.

Eunsoo

>
>
>
>
> */Eunsoo Shim <eunsoo at research.panasonic.com>/* wrote:
>
>     Mike Robinson wrote:
>
>     >
>     > Radhika:
>     >
>     > You agree that there are many problems that are not resolved
>     yet, and
>     > no solution is available. That means more research needs to be
>     done.
>     > Research belongs in the labs, Universities or in IRTF.
>     >
>     The P2P technologies are continusouly evolving and still various
>     issues
>     remain open. There was a good presentation about DHT in the technical
>     plenary of IETF65, which pointed out many security issues of DHTs.
>     However, the conclusion of the presentation was not "don't use
>     DHT" but
>     rather "use DHT with caution" to be simplified in my understanding.
>     A P2P-based VoIP service provider has more subscribers than any other
>     VoIP providers currently. Other commercial developments related to
>     P2P
>     SIP are being released via news media. I think this is a quite strong
>     signal calling for P2! P ! SIP standardization efforts. The number
>     of people
>     who have been showing interest in P2P SIP last several IETF
>     meetings is
>     another signal, I think.
>
>     Novertheless, in general, I guess it is a quite subjective matter
>     whether a particular technology is ready for standardization or its
>     standardization is worth the effort. But I guess you can agree
>     that it
>     is not a good idea waiting for all the problems to be solved before
>     standardization of anything.
>     Many protocols or technologies have been evolving even after the
>     first
>     version of their standards. It does not mean standardization of those
>     protocols was a mistake or worthless.
>     Considering all the time we spent so far on discussions about various
>     apsects of P2P SIP, I think we have passed the stage to say "there
>     are
>     lots of problems unsolved". Would it be too much if I ask for more
>     concrete discussions like what tools we can develop, what use case
>     can
>     be supported in which ma! nn! er, and what is the difficulty for a
>     certain
>     use case?
>
>     > It is not clear how you can change a DHT and still be compatible
>     since
>     > SIP messages and headers fields are so much dependent on the
>     type of
>     > underlying DHT used.
>
>     I thought you attended the P2P SIP ad-hoc meeting in Dallas but
>     maybe I
>     was confused.
>     What you are describing above is based on a certain approach. Please
>     check the presentation about a P2P SIP architecture made during the
>     ad-hoc meeting in Dallas. The presentation is available in the IETF65
>     agenda and presentation page. You can find a different approach that
>     enables a modular design in which the underlying DHT or an overlay
>     algorithm can be changed without affecting the SIP messages.
>
>     > The data maintained by each DHT is different and so are node
>     insertion
>     > and removal procedures. I believe the security issues of each DHT
>     > types would also be different, so when someone says tha! t !
>     they can
>     > replace DHT later on, they are just being too simplistic.
>     >
>     > Do you really want to create OSPF versus IS-IS kind of two
>     > incompatible solutions?
>     >
>     I think what Radhika meant was that different P2P SIP domains
>     could use
>     different DHTs. OSPF and IS-IS won't be together in the same
>     domain but
>     each can be used in different domains. Same approach for P2P SIP.
>     Again
>     please check the P2P SIP architecture presentation in the ad-hoc
>     meeting.
>
>     I respect your concern regarding P2P SIP standardization. Such
>     concern
>     was expressed back in the first P2P SIP ad-hoc meeting in March last
>     year. We have come a long way since then to have a proposed
>     charter and
>     the BOF in Dallas.
>
>     I hope we move forward.
>     Thanks.
>
>     Eunsoo
>     ps)I think David made really good points.
>
>     >
>     > */"Roy, Radhika R." /* wrote:
>     >
>     > I agree with David.
>     > It is also NOT harm! t! o link! age a particular design with a
>     > particular DHT, if this is the good choice for a given time.
>     > However, other proposals are also welcome. It is an open to all
>     of us.
>     > Standardizations are always done throughout the world if there is
>     > a consensus to go for something that is good. With respect
>     > P2P-SIP, there are a lot of things to be solved. However, if
>     > something provides immediate value, as David and many others of
>     > this list have opined, what is the problem for standardization? A
>     > better standard in the future can always supersede an inferior
>     > standard of the past.
>     > Best regards,
>     > Radhika
>     >
>     ------------------------------------------------------------------------
>     > *From:* p2p-sip-bounces at cs.columbia.edu
>     > [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *David
>     Barrett
>     > *Sent:* Monday, March 27, 2006 2:40 PM> *To:* 'Mike Robinson';
>     henry at pulver.com; 'Karst Bjorgson'; 'Kyara
>     > Yamamoto'; 'Peter Pan'
>     > *Cc:* p2p-sip at cs.columbia.edu
>     > *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     > I guess one view is the IETF should forever be relegated to the j!
>     > anitorial role of cleaning up after the party???s over. But I
>     > believe the IETF can participate in and even promote the party
>     > through standardization. I???d like the IETF to do more than
>     > simply formalize de facto standards, but actually create the
>     > standards that attract the party.
>     > In the case of P2P-SIP, I entirely agree there are multiple,
>     > valid, proven ways to solve each of these problems ??? none of
>     > which is the clear victor. And I also agree there are outstanding
>     > issues that have no proven solutions.
>     > If we pick amongst the best practices today, and skip the problems
>     > to which we have no clear solution, I personally believe that???ll
>     > give us the most immediate value, while paving the road for future
>     > improvements. I don???t believe we???ll ever meet any higher
>     > standard to get started than has already been met.
>     > -david
>     >
>     ------------------------------------------------------------------------
>     > *From:* p2p-sip-bounces at cs.columbia.edu
>     > [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *Mike
>     Robinson
>     > *Sent:* Monday, March 27, 2006 10:32 AM
>     > *To:* henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter
>     > Pan'
>     > *Cc:* p2p-sip at cs.columbia.edu
>     > *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     > Henry:
>     > Do I need to tell that 48! 00 P2P google references have nothing
>     > to do with P2P SIP?
>     > Please let us not obfuscate the issues. You ! kn! ow, as well as,
>     > anyone sincere about the value of standards, that the pre-mature
>     > standardization leads to bad design as well as failure of the
>     > standards. As I mentioned there is no compelling use case
>     > presented yet that says we need to "standardize" p2p sip now
>     > without even bothering to get output from P2P IRTF group which is
>     > looking into core issues of P2P research.
>     > Many would say that we first need an IETF standard for P2P file
>     > sharing before we want to use "P2P" to design P2P SIP. We need
>     > more research on P2P SIP before we standardize any particular
>     > "conspiracy" (I use conspiracy in the sense of Karst definition of
>     > linking design with a particular DHT)
>     > -mike
>     > V! OIP and SIP Trainer and Consultant
>     >
>     > */Henry Sinnreich /* wrote:
>     >
>     > Mike Robinson writes:
>     >>How many! r! esearch ! or analytical papers have been published
>     > in this area?
>     > I have found by Googling a reference listing 4,800 papers on
>     > P2P. Hopefully, this enough to get us started. J
>     > Henry
>     >
>     > rates.
>     >
>     >
>     >
>     ------------------------------------------------------------------------
>     > Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls.
>     Great
>     > rates starting at 1???/min.
>     >
>     >
>     >
>     >
>     >
>     >------------------------------------------------------------------------
>     >
>     >_______________________________________________
>     >p2p-sip mailing list
>     >p2p-sip at cs.columbia.edu
>     >https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>     >
>     >
>
>
> ------------------------------------------------------------------------
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great 
> rates starting at 1?/min. 
> <http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/evt=39666/*http://beta.messenger.yahoo.com>



From eunsoo at research.panasonic.com  Thu Mar 30 16:38:40 2006
From: eunsoo at research.panasonic.com (Eunsoo Shim)
Date: Thu, 30 Mar 2006 16:38:40 -0500
Subject: [p2p-sip] Questions on #5 "A P2P SIP Arch - 2 layer approach".
In-Reply-To: <03CD385D379ECE4988B9C88834382EA701C3882A@xmb-sjc-22e.amer.cisco.com>
References: <03CD385D379ECE4988B9C88834382EA701C3882A@xmb-sjc-22e.amer.cisco.com>
Message-ID: <442C4FE0.4020404@research.panasonic.com>

Krishna Sankar (ksankar) wrote:

>Folks,
>
>	Have some quick questions on slide #5 "A P2P SIP Arch - 2 layer
>approach". I assume:
>
>	a)	This is could be a network topology, definitely not a
>logical topology.
>	b)	ONs & SNs are all SIP end-points and will be hashed as
>per Chord algorithm.
>  
>
ON and SN are NOT SIP entities.
ONs do not participate in DHT.

>	c)	So even though there could be a visibility and
>reachability association between specific SNs and ONs, the P2P substrate
>will not reflect this, as the hash could place them in different areas
>in the ring. In fact, if the hash didn't, it will degenerate to a
>routing table (not that that is bad, but that is not what we want to
>achieve at a macro level)
>  
>
ONs are not in the (Chord) ring.

>	d)	Hopefully this diagram is not overloaded with NAT
>traversal as well. In that case, the SN will represent the ONs which
>hang off of it. In this case the hashing will happen on the ON name but
>the IP address will be that of the associated SN. The SN knows how to
>get to the Ones it represents.
>  
>
The diagram is not overloaded with NAT traversal.

>	e)	Extending a) further, the Chord ring representation of
>this diagram will be very different.
>  
>
Please see the answers in the above.

Thanks.

Eunsoo

From RADHIKA.R.ROY at saic.com  Thu Mar 30 17:20:58 2006
From: RADHIKA.R.ROY at saic.com (Roy, Radhika R.)
Date: Thu, 30 Mar 2006 17:20:58 -0500
Subject: [p2p-sip] Some comments on Use-cases document
Message-ID: <A879F72922C16E4F80EA2F34EBF2AAA31E96A7@0591-ITS-EXMP02.us.saic.com>

Inline [RRR]

  _____  

From: p2p-sip-bounces at cs.columbia.edu on behalf of Eunsoo Shim
Sent: Thu 3/30/2006 4:28 PM
To: Mike Robinson
Cc: p2p-sip at cs.columbia.edu
Subject: Re: [p2p-sip] Some comments on Use-cases document



Mike Robinson wrote: 

> 
> Eunsoo wrote 
> 
> It may be possible that you don't even need to tinker much with SIP 
> (johnson's draft discusses one such possibility). 
> 
> In addition, using p2p overlay to resolve URIs does not solve voice 
> quality issues, 

Right. Neither SIP or P2P overlay protocol will solve voice quality 
issues. The group is not aimed to address voice quality issues. 
For example, for voice quality, codec selection is important. The SDP 
can be used to notify the type of the codec. But you got to develop an 
algorithm to select a codec, which is not what an IETF WG does 
typically. You may need a better voice codec. It is not dealt with in 
the IETF, either. Please don't expect an IETF WG will address/do all 
those things. 
This group is not supposed to address all kind of issues related to 
SIP-based P2P real-time media applications. The group should develop a 
few tools/components by which people can build a system. 

[RRR] In addition what Eunsoo explained, the QoS issues for real-time
applications (e.g., audio, video) are being addressed in IETF NSIS WG in
addition what DiffServ and RSVP deal with.

[RRR] p2p-sip is not supposed to anything new what SIP is already is doing
for QoS (e.g., SIP signaling messages are being extended to invoke lower
layer QoS signaling messages at the time of the call setup). So, the
voice/video quality is a non-issue for p2p-sip.

Thanks. 

Eunsoo 

> 
> 
> 
> 
> */Eunsoo Shim <eunsoo at research.panasonic.com>/* wrote: 
> 
>     Mike Robinson wrote: 
> 
>     > 
>     > Radhika: 
>     > 
>     > You agree that there are many problems that are not resolved 
>     yet, and 
>     > no solution is available. That means more research needs to be 
>     done. 
>     > Research belongs in the labs, Universities or in IRTF. 
>     > 
>     The P2P technologies are continusouly evolving and still various 
>     issues 
>     remain open. There was a good presentation about DHT in the technical 
>     plenary of IETF65, which pointed out many security issues of DHTs. 
>     However, the conclusion of the presentation was not "don't use 
>     DHT" but 
>     rather "use DHT with caution" to be simplified in my understanding. 
>     A P2P-based VoIP service provider has more subscribers than any other 
>     VoIP providers currently. Other commercial developments related to 
>     P2P 
>     SIP are being released via news media. I think this is a quite strong 
>     signal calling for P2! P ! SIP standardization efforts. The number 
>     of people 
>     who have been showing interest in P2P SIP last several IETF 
>     meetings is 
>     another signal, I think. 
> 
>     Novertheless, in general, I guess it is a quite subjective matter 
>     whether a particular technology is ready for standardization or its 
>     standardization is worth the effort. But I guess you can agree 
>     that it 
>     is not a good idea waiting for all the problems to be solved before 
>     standardization of anything. 
>     Many protocols or technologies have been evolving even after the 
>     first 
>     version of their standards. It does not mean standardization of those 
>     protocols was a mistake or worthless. 
>     Considering all the time we spent so far on discussions about various 
>     apsects of P2P SIP, I think we have passed the stage to say "there 
>     are 
>     lots of problems unsolved". Would it be too much if I ask for more 
>     concrete discussions like what tools we can develop, what use case 
>     can 
>     be supported in which ma! nn! er, and what is the difficulty for a 
>     certain 
>     use case? 
> 
>     > It is not clear how you can change a DHT and still be compatible 
>     since 
>     > SIP messages and headers fields are so much dependent on the 
>     type of 
>     > underlying DHT used. 
> 
>     I thought you attended the P2P SIP ad-hoc meeting in Dallas but 
>     maybe I 
>     was confused. 
>     What you are describing above is based on a certain approach. Please 
>     check the presentation about a P2P SIP architecture made during the 
>     ad-hoc meeting in Dallas. The presentation is available in the IETF65 
>     agenda and presentation page. You can find a different approach that 
>     enables a modular design in which the underlying DHT or an overlay 
>     algorithm can be changed without affecting the SIP messages. 
> 
>     > The data maintained by each DHT is different and so are node 
>     insertion 
>     > and removal procedures. I believe the security issues of each DHT 
>     > types would also be different, so when someone says tha! t ! 
>     they can 
>     > replace DHT later on, they are just being too simplistic. 
>     > 
>     > Do you really want to create OSPF versus IS-IS kind of two 
>     > incompatible solutions? 
>     > 
>     I think what Radhika meant was that different P2P SIP domains 
>     could use 
>     different DHTs. OSPF and IS-IS won't be together in the same 
>     domain but 
>     each can be used in different domains. Same approach for P2P SIP. 
>     Again 
>     please check the P2P SIP architecture presentation in the ad-hoc 
>     meeting. 
> 
>     I respect your concern regarding P2P SIP standardization. Such 
>     concern 
>     was expressed back in the first P2P SIP ad-hoc meeting in March last 
>     year. We have come a long way since then to have a proposed 
>     charter and 
>     the BOF in Dallas. 
> 
>     I hope we move forward. 
>     Thanks. 
> 
>     Eunsoo 
>     ps)I think David made really good points. 
> 
>     > 
>     > */"Roy, Radhika R." /* wrote: 
>     > 
>     > I agree with David. 
>     > It is also NOT harm! t! o link! age a particular design with a 
>     > particular DHT, if this is the good choice for a given time. 
>     > However, other proposals are also welcome. It is an open to all 
>     of us. 
>     > Standardizations are always done throughout the world if there is 
>     > a consensus to go for something that is good. With respect 
>     > P2P-SIP, there are a lot of things to be solved. However, if 
>     > something provides immediate value, as David and many others of 
>     > this list have opined, what is the problem for standardization? A 
>     > better standard in the future can always supersede an inferior 
>     > standard of the past. 
>     > Best regards, 
>     > Radhika 
>     > 
>
------------------------------------------------------------------------ 
>     > *From:* p2p-sip-bounces at cs.columbia.edu 
>     > [mailto:p2p-sip-bounces at cs.columbia.edu
<mailto:p2p-sip-bounces at cs.columbia.edu> ] *On Behalf Of *David 
>     Barrett 
>     > *Sent:* Monday, March 27, 2006 2:40 PM> *To:* 'Mike Robinson'; 
>     henry at pulver.com; 'Karst Bjorgson'; 'Kyara 
>     > Yamamoto'; 'Peter Pan' 
>     > *Cc:* p2p-sip at cs.columbia.edu 
>     > *Subject:* Re: [p2p-sip] Some comments on Use-cases document 
>     > I guess one view is the IETF should forever be relegated to the j! 
>     > anitorial role of cleaning up after the party?EUR(tm)s over. But I 
>     > believe the IETF can participate in and even promote the party 
>     > through standardization. I?EUR(tm)d like the IETF to do more than 
>     > simply formalize de facto standards, but actually create the 
>     > standards that attract the party. 
>     > In the case of P2P-SIP, I entirely agree there are multiple, 
>     > valid, proven ways to solve each of these problems ?EUR" none of 
>     > which is the clear victor. And I also agree there are outstanding 
>     > issues that have no proven solutions. 
>     > If we pick amongst the best practices today, and skip the problems 
>     > to which we have no clear solution, I personally believe
that?EUR(tm)ll 
>     > give us the most immediate value, while paving the road for future 
>     > improvements. I don?EUR(tm)t believe we?EUR(tm)ll ever meet any
higher 
>     > standard to get started than has already been met. 
>     > -david 
>     > 
>
------------------------------------------------------------------------ 
>     > *From:* p2p-sip-bounces at cs.columbia.edu 
>     > [mailto:p2p-sip-bounces at cs.columbia.edu
<mailto:p2p-sip-bounces at cs.columbia.edu> ] *On Behalf Of *Mike 
>     Robinson 
>     > *Sent:* Monday, March 27, 2006 10:32 AM 
>     > *To:* henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter 
>     > Pan' 
>     > *Cc:* p2p-sip at cs.columbia.edu 
>     > *Subject:* Re: [p2p-sip] Some comments on Use-cases document 
>     > Henry: 
>     > Do I need to tell that 48! 00 P2P google references have nothing 
>     > to do with P2P SIP? 
>     > Please let us not obfuscate the issues. You ! kn! ow, as well as, 
>     > anyone sincere about the value of standards, that the pre-mature 
>     > standardization leads to bad design as well as failure of the 
>     > standards. As I mentioned there is no compelling use case 
>     > presented yet that says we need to "standardize" p2p sip now 
>     > without even bothering to get output from P2P IRTF group which is 
>     > looking into core issues of P2P research. 
>     > Many would say that we first need an IETF standard for P2P file 
>     > sharing before we want to use "P2P" to design P2P SIP. We need 
>     > more research on P2P SIP before we standardize any particular 
>     > "conspiracy" (I use conspiracy in the sense of Karst definition of 
>     > linking design with a particular DHT) 
>     > -mike 
>     > V! OIP and SIP Trainer and Consultant 
>     > 
>     > */Henry Sinnreich /* wrote: 
>     > 
>     > Mike Robinson writes: 
>     >>How many! r! esearch ! or analytical papers have been published 
>     > in this area? 
>     > I have found by Googling a reference listing 4,800 papers on 
>     > P2P. Hopefully, this enough to get us started. J 
>     > Henry 
>     > 
>     > rates. 
>     > 
>     > 
>     > 
>
------------------------------------------------------------------------ 
>     > Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. 
>     Great 
>     > rates starting at 1???/min. 
>     > 
>     > 
>     > 
>     > 
>     > 
>
>------------------------------------------------------------------------ 
>     > 
>     >_______________________________________________ 
>     >p2p-sip mailing list 
>     >p2p-sip at cs.columbia.edu 
>     >https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
<https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip>  
>     > 
>     > 
> 
> 
> ------------------------------------------------------------------------ 
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great 
> rates starting at 1?/min. 
>
<http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/ev
t=39666/*http://beta.messenger.yahoo.com
<http://us.rd.yahoo.com/mail_us/taglines/postman7/*http://us.rd.yahoo.com/ev
t=39666/*http://beta.messenger.yahoo.com> > 


_______________________________________________ 
p2p-sip mailing list 
p2p-sip at cs.columbia.edu 
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
<https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip>  


From kyara543 at yahoo.com  Thu Mar 30 19:36:24 2006
From: kyara543 at yahoo.com (Kyara Yamamoto)
Date: Thu, 30 Mar 2006 16:36:24 -0800 (PST)
Subject: [p2p-sip] p2p-sip Digest, Vol 8, Issue 61
In-Reply-To: <mailman.16270.1143754605.16379.p2p-sip@cs.columbia.edu>
Message-ID: <20060331003624.98768.qmail@web38613.mail.mud.yahoo.com>

Can you dictate details about *down-to-earth* approach? 

We, scientists, leads, researchers  and professors are mainly for protocol design. Engineers realize protocols. We do not endorse realization. Realization takes time/resources and those with only academic value like SOSIMPLE or VOCAL is of tiny use to industry and consumers.

--Kyara

p2p-sip-request at cs.columbia.edu wrote:maybe a down-to-earth approach can help uncover problems in current p2p-sip
direction.

egs,

* do sip retransmission scheme and backoff timers fit well in p2p arena?
(some p2p msgs need no retransmission)
* what if a sip/udp packet exceeds MTU with appended p2p info (eg,
finger/k-bucket entries)?
* can p2p-sip keep incremental paces (and "religion") after strong security
and distributed databases are added to the overlayS in later phases?
* even if it imaginably can, will it be imaginably efficient and
competitive? (h323 and Jabber fans criticize protocol efficiency of sip even
with sigcomp)
* autonomous certificate management...etc

me feel the p2p-sip topic definitely not SOSIMPLE...

peter
----- Original Message -----
From: Mike Robinson
To: Eunsoo Shim
Cc: p2p-sip at cs.columbia.edu
Sent: Wednesday, March 29, 2006 9:18 PM
Subject: Re: [p2p-sip] Some comments on Use-cases document



Eunsoo wrote

>A P2P-based VoIP service provider has more subscribers than any other
> VoIP providers currently.

>Other commercial developments related to P2P SIP are being released via >
news media.

What does it really prove?  Let us see.

Unfortunately, you are  equating Skype  VoIP over  P2P with your version of
p2psip. "P2P VOIP (as in skype) works" is neither a proof nor a validation
of your p2p sip design since skype is not based on p2p sip. Besides,   it
is not a pure p2p solution in a strict sense. Skype documentation also lists
a number of optimizations that go far beyond signaling a session to get the
better voice quality. Please see their site for more details.

Actually, without having some support from an infrastructure provider (or
always connected nodes in the public domain) it may not be even possible to
run a pure p2p voip system beyond your organization. You have issues! of NAT
traversal, authentication and  certification management  that cannot be
simply washed away. Both Krishna Sankar and Karst raised a number of
technical issues.

May be what IETF needs a VOIP over P2P forum, rather than P2P SIP group.
In the p2p sip, all you are using p2p overlay is for resolving URI into end
point
IP addresses.  That can be done in many ways, and SIP can be used in the
unmodified way as it is. It may be possible that you don't even need to
tinker much with SIP (johnson's draft discusses one such possibility).

In addition, using  p2p overlay to resolve URIs does not solve  voice
quality issues, and I think Krishna Sankar articulated those issues very
well. So, if you believe in skype, that means that you need voip over p2p
group that looks into broader issues than the proposed p2p sip group.

--mike




Eunsoo Shim  wrote:
Mike Robinson wrote:

>
> Radhika:
>
> You agree that there are many problems that are not resolved yet, and
> no solution is available. That means more research needs to be done.
> Research belongs in the labs, Universities or in IRTF.
>
The P2P technologies are continusouly evolving and still various issues
remain open. There was a good presentation about DHT in the technical
plenary of IETF65, which pointed out many security issues of DHTs.
However, the conclusion of the presentation was not "don't use DHT" but
rather "use DHT with caution" to be simplified in my understanding.
A P2P-based VoIP service provider has more subscribers than any other
VoIP providers currently. Other commercial developments related to P2P
SIP are being released via news media. I think this is a quite strong
signal calling for P2! P SIP standardization efforts. The number of people
who have been showing interest in P2P SIP last several IETF meetings is
another signal, I think.

Novertheless, in general, I guess it is a quite subjective matter
whether a particular technology is ready for standardization or its
standardization is worth the effort. But I guess you can agree that it
is not a good idea waiting for all the problems to be solved before
standardization of anything.
Many protocols or technologies have been evolving even after the first
version of their standards. It does not mean standardization of those
protocols was a mistake or worthless.
Considering all the time we spent so far on discussions about various
apsects of P2P SIP, I think we have passed the stage to say "there are
lots of problems unsolved". Would it be too much if I ask for more
concrete discussions like what tools we can develop, what use case can
be supported in which ma! nner, and what is the difficulty for a certain
use case?

> It is not clear how you can change a DHT and still be compatible since
> SIP messages and headers fields are so much dependent on the type of
> underlying DHT used.

I thought you attended the P2P SIP ad-hoc meeting in Dallas but maybe I
was confused.
What you are describing above is based on a certain approach. Please
check the presentation about a P2P SIP architecture made during the
ad-hoc meeting in Dallas. The presentation is available in the IETF65
agenda and presentation page. You can find a different approach that
enables a modular design in which the underlying DHT or an overlay
algorithm can be changed without affecting the SIP messages.

> The data maintained by each DHT is different and so are node insertion
> and removal procedures. I believe the security issues of each DHT
> types would also be different, so when someone says tha! t they can
> replace DHT later on, they are just being too simplistic.
>
> Do you really want to create OSPF versus IS-IS kind of two
> incompatible solutions?
>
I think what Radhika meant was that different P2P SIP domains could use
different DHTs. OSPF and IS-IS won't be together in the same domain but
each can be used in different domains. Same approach for P2P SIP. Again
please check the P2P SIP architecture presentation in the ad-hoc meeting.

I respect your concern regarding P2P SIP standardization. Such concern
was expressed back in the first P2P SIP ad-hoc meeting in March last
year. We have come a long way since then to have a proposed charter and
the BOF in Dallas.

I hope we move forward.
Thanks.

Eunsoo
ps)I think David made really good points.

>
> */"Roy, Radhika R." /* wrote:
>
> I agree with David.
> It is also NOT harm! to link! age a particular design with a
> particular DHT, if this is the good choice for a given time.
> However, other proposals are also welcome. It is an open to all of us.
> Standardizations are always done throughout the world if there is
> a consensus to go for something that is good. With respect
> P2P-SIP, there are a lot of things to be solved. However, if
> something provides immediate value, as David and many others of
> this list have opined, what is the problem for standardization? A
> better standard in the future can always supersede an inferior
> standard of the past.
> Best regards,
> Radhika
> ------------------------------------------------------------------------
> *From:* p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *David Barrett
> *Sent:* Monday, March 27, 2006 2:40 PM> *To:* 'Mike Robinson';
henry at pulver.com; 'Karst Bjorgson'; 'Kyara
> Yamamoto'; 'Peter Pan'
> *Cc:* p2p-sip at cs.columbia.edu
> *Subject:* Re: [p2p-sip] Some comments on Use-cases document
> I guess one view is the IETF should forever be relegated to the j!
> anitorial role of cleaning up after the party????????s over. But I
> believe the IETF can participate in and even promote the party
> through standardization. I????????d like the IETF to do more than
> simply formalize de facto standards, but actually create the
> standards that attract the party.
> In the case of P2P-SIP, I entirely agree there are multiple,
> valid, proven ways to solve each of these problems ???????? none of
> which is the clear victor. And I also agree there are outstanding
> issues that have no proven solutions.
> If we pick amongst the best practices today, and skip the problems
> to which we have no clear solution, I personally believe that????????ll
> give us the most immediate value, while paving the road for future
> improvements. I don????????t believe we????????ll ever meet any higher
> standard to get started than has already been met.
> -david
> ------------------------------------------------------------------------
> *From:* p2p-sip-bounces at cs.columbia.edu
> [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *Mike Robinson
> *Sent:* Monday, March 27, 2006 10:32 AM
> *To:* henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter
> Pan'
> *Cc:* p2p-sip at cs.columbia.edu
> *Subject:* Re: [p2p-sip] Some comments on Use-cases document
> Henry:
> Do I need to tell that 48! 00 P2P google references have nothing
> to do with P2P SIP?
> Please let us not obfuscate the issues. You ! know, as well as,
> anyone sincere about the value of standards, that the pre-mature
> standardization leads to bad design as well as failure of the
> standards. As I mentioned there is no compelling use case
> presented yet that says we need to "standardize" p2p sip now
> without even bothering to get output from P2P IRTF group which is
> looking into core issues of P2P research.
> Many would say that we first need an IETF standard for P2P file
> sharing before we want to use "P2P" to design P2P SIP. We need
> more research on P2P SIP before we standardize any particular
> "conspiracy" (I use conspiracy in the sense of Karst definition of
> linking design with a particular DHT)
> -mike
> V! OIP and SIP Trainer and Consultant
>
> */Henry Sinnreich /* wrote:
>
> Mike Robinson writes:
>>How many! research ! or analytical papers have been published
> in this area?
> I have found by Googling a reference listing 4,800 papers on
> P2P. Hopefully, this enough to get us started. J
> Henry
>
> rates.
>
>
> ------------------------------------------------------------------------
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great
> rates starting at 1??????/min.
>
>
>
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>p2p-sip mailing list
>p2p-sip at cs.columbia.edu
>https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>
>






Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great rates
starting at 1??/min.


From: Eunsoo Shim <eunsoo at research.panasonic.com>
CC: p2p-sip at cs.columbia.edu
To: Mike Robinson <mike_robnson at yahoo.com>
Date: Thu, 30 Mar 2006 16:28:56 -0500
Subject: Re: [p2p-sip] Some comments on Use-cases document

 Mike Robinson wrote:

>
> Eunsoo wrote
>
> >A P2P-based VoIP service provider has more subscribers than any other
> > VoIP providers currently.
>
> >Other commercial developments related to P2P SIP are being released 
> via > news media.
>
> What does it really prove? Let us see.
>
> Unfortunately, you are equating Skype VoIP over P2P with your version 
> of p2psip. "P2P VOIP (as in skype) works" is neither a proof nor a 
> validation of your p2p sip design since skype is not based on p2p sip. 
> Besides, it is not a pure p2p solution in a strict sense.

I guess you mean something without any kind of central entity by 'pure 
p2p solution in a strict sense'. If you review discussions in the past 
P2P SIP ad-hoc meetings and in the mailing list, you may find out that 
the group is quite open for assuming some limited central entities, in 
particular, for authentication. As the proposed charter suggests, many 
people think a distributed identity management is a hard problem and the 
group should not try to standardize a protocol for that w.r.t. P2P SIP.

> Skype documentation also lists a number of optimizations that go far 
> beyond signaling a session to get the better voice quality. Please see 
> their site for more details.
>
> Actually, without having some support from an infrastructure provider 
> (or always connected nodes in the public domain) it may not be even 
> possible to run a pure p2p voip system beyond your organization. You 
> have issues! o! f NAT traversal, authentication and certification 
> management that cannot be simply washed away. Both Krishna Sankar and 
> Karst raised a number of technical issues.

All those issues have been raised and discussed in the group from the 
beginning.

>
> May be what IETF needs a VOIP over P2P forum, rather than P2P SIP group.

I guess you mean a group to design a complete P2P VoIP system by 'a VoIP 
over P2P forum'. You may notice that IETF WGs do not design a complete 
system but design protocols that are components of systems or tools. It 
is (should be) applied to the P2P SIP group, too.

> In the p2p sip, all you are using p2p overlay is for resolving URI 
> into end point
> IP addresses.

You are close. Basically the P2P overlay provides a resource location 
service for SIP. The resource could be UA (or user), TURN server, and so 
on.

> That can be done in many ways, and SIP can be used in the unmodified 
> way as it is.

That's the approach a number of people proposed.

> It may be possible that you don't even need to tinker much with SIP 
> (johnson's draft discusses one such possibility).
>
> In addition, using p2p overlay to resolve URIs does not solve voice 
> quality issues,

Right. Neither SIP or P2P overlay protocol will solve voice quality 
issues. The group is not aimed to address voice quality issues.
For example, for voice quality, codec selection is important. The SDP 
can be used to notify the type of the codec. But you got to develop an 
algorithm to select a codec, which is not what an IETF WG does 
typically. You may need a better voice codec. It is not dealt with in 
the IETF, either. Please don't expect an IETF WG will address/do all 
those things.
This group is not supposed to address all kind of issues related to 
SIP-based P2P real-time media applications. The group should develop a 
few tools/components by which people can build a system.

> and I think Krishna Sankar articulated those issues very well. So, if 
> you believe in skype, that means that you need voip over p2p group 
> that looks into broader issues than the proposed p2p sip group.
>
Again, I understand there are broader issues than what is covered by the 
proposed charter. Actually that is the purpose of the charter --- 
narrowing down and defining a limited scope/deliverables.

Thanks.

Eunsoo

>
>
>
>
> */Eunsoo Shim /* wrote:
>
>     Mike Robinson wrote:
>
>     >
>     > Radhika:
>     >
>     > You agree that there are many problems that are not resolved
>     yet, and
>     > no solution is available. That means more research needs to be
>     done.
>     > Research belongs in the labs, Universities or in IRTF.
>     >
>     The P2P technologies are continusouly evolving and still various
>     issues
>     remain open. There was a good presentation about DHT in the technical
>     plenary of IETF65, which pointed out many security issues of DHTs.
>     However, the conclusion of the presentation was not "don't use
>     DHT" but
>     rather "use DHT with caution" to be simplified in my understanding.
>     A P2P-based VoIP service provider has more subscribers than any other
>     VoIP providers currently. Other commercial developments related to
>     P2P
>     SIP are being released via news media. I think this is a quite strong
>     signal calling for P2! P ! SIP standardization efforts. The number
>     of people
>     who have been showing interest in P2P SIP last several IETF
>     meetings is
>     another signal, I think.
>
>     Novertheless, in general, I guess it is a quite subjective matter
>     whether a particular technology is ready for standardization or its
>     standardization is worth the effort. But I guess you can agree
>     that it
>     is not a good idea waiting for all the problems to be solved before
>     standardization of anything.
>     Many protocols or technologies have been evolving even after the
>     first
>     version of their standards. It does not mean standardization of those
>     protocols was a mistake or worthless.
>     Considering all the time we spent so far on discussions about various
>     apsects of P2P SIP, I think we have passed the stage to say "there
>     are
>     lots of problems unsolved". Would it be too much if I ask for more
>     concrete discussions like what tools we can develop, what use case
>     can
>     be supported in which ma! nn! er, and what is the difficulty for a
>     certain
>     use case?
>
>     > It is not clear how you can change a DHT and still be compatible
>     since
>     > SIP messages and headers fields are so much dependent on the
>     type of
>     > underlying DHT used.
>
>     I thought you attended the P2P SIP ad-hoc meeting in Dallas but
>     maybe I
>     was confused.
>     What you are describing above is based on a certain approach. Please
>     check the presentation about a P2P SIP architecture made during the
>     ad-hoc meeting in Dallas. The presentation is available in the IETF65
>     agenda and presentation page. You can find a different approach that
>     enables a modular design in which the underlying DHT or an overlay
>     algorithm can be changed without affecting the SIP messages.
>
>     > The data maintained by each DHT is different and so are node
>     insertion
>     > and removal procedures. I believe the security issues of each DHT
>     > types would also be different, so when someone says tha! t !
>     they can
>     > replace DHT later on, they are just being too simplistic.
>     >
>     > Do you really want to create OSPF versus IS-IS kind of two
>     > incompatible solutions?
>     >
>     I think what Radhika meant was that different P2P SIP domains
>     could use
>     different DHTs. OSPF and IS-IS won't be together in the same
>     domain but
>     each can be used in different domains. Same approach for P2P SIP.
>     Again
>     please check the P2P SIP architecture presentation in the ad-hoc
>     meeting.
>
>     I respect your concern regarding P2P SIP standardization. Such
>     concern
>     was expressed back in the first P2P SIP ad-hoc meeting in March last
>     year. We have come a long way since then to have a proposed
>     charter and
>     the BOF in Dallas.
>
>     I hope we move forward.
>     Thanks.
>
>     Eunsoo
>     ps)I think David made really good points.
>
>     >
>     > */"Roy, Radhika R." /* wrote:
>     >
>     > I agree with David.
>     > It is also NOT harm! t! o link! age a particular design with a
>     > particular DHT, if this is the good choice for a given time.
>     > However, other proposals are also welcome. It is an open to all
>     of us.
>     > Standardizations are always done throughout the world if there is
>     > a consensus to go for something that is good. With respect
>     > P2P-SIP, there are a lot of things to be solved. However, if
>     > something provides immediate value, as David and many others of
>     > this list have opined, what is the problem for standardization? A
>     > better standard in the future can always supersede an inferior
>     > standard of the past.
>     > Best regards,
>     > Radhika
>     >
>     ------------------------------------------------------------------------
>     > *From:* p2p-sip-bounces at cs.columbia.edu
>     > [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *David
>     Barrett
>     > *Sent:* Monday, March 27, 2006 2:40 PM> *To:* 'Mike Robinson';
>     henry at pulver.com; 'Karst Bjorgson'; 'Kyara
>     > Yamamoto'; 'Peter Pan'
>     > *Cc:* p2p-sip at cs.columbia.edu
>     > *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     > I guess one view is the IETF should forever be relegated to the j!
>     > anitorial role of cleaning up after the party???s over. But I
>     > believe the IETF can participate in and even promote the party
>     > through standardization. I???d like the IETF to do more than
>     > simply formalize de facto standards, but actually create the
>     > standards that attract the party.
>     > In the case of P2P-SIP, I entirely agree there are multiple,
>     > valid, proven ways to solve each of these problems ??? none of
>     > which is the clear victor. And I also agree there are outstanding
>     > issues that have no proven solutions.
>     > If we pick amongst the best practices today, and skip the problems
>     > to which we have no clear solution, I personally believe that???ll
>     > give us the most immediate value, while paving the road for future
>     > improvements. I don???t believe we???ll ever meet any higher
>     > standard to get started than has already been met.
>     > -david
>     >
>     ------------------------------------------------------------------------
>     > *From:* p2p-sip-bounces at cs.columbia.edu
>     > [mailto:p2p-sip-bounces at cs.columbia.edu] *On Behalf Of *Mike
>     Robinson
>     > *Sent:* Monday, March 27, 2006 10:32 AM
>     > *To:* henry at pulver.com; 'Karst Bjorgson'; 'Kyara Yamamoto'; 'Peter
>     > Pan'
>     > *Cc:* p2p-sip at cs.columbia.edu
>     > *Subject:* Re: [p2p-sip] Some comments on Use-cases document
>     > Henry:
>     > Do I need to tell that 48! 00 P2P google references have nothing
>     > to do with P2P SIP?
>     > Please let us not obfuscate the issues. You ! kn! ow, as well as,
>     > anyone sincere about the value of standards, that the pre-mature
>     > standardization leads to bad design as well as failure of the
>     > standards. As I mentioned there is no compelling use case
>     > presented yet that says we need to "standardize" p2p sip now
>     > without even bothering to get output from P2P IRTF group which is
>     > looking into core issues of P2P research.
>     > Many would say that we first need an IETF standard for P2P file
>     > sharing before we want to use "P2P" to design P2P SIP. We need
>     > more research on P2P SIP before we standardize any particular
>     > "conspiracy" (I use conspiracy in the sense of Karst definition of
>     > linking design with a particular DHT)
>     > -mike
>     > V! OIP and SIP Trainer and Consultant
>     >
>     > */Henry Sinnreich /* wrote:
>     >
>     > Mike Robinson writes:
>     >>How many! r! esearch ! or analytical papers have been published
>     > in this area?
>     > I have found by Googling a reference listing 4,800 papers on
>     > P2P. Hopefully, this enough to get us started. J
>     > Henry
>     >
>     > rates.
>     >
>     >
>     >
>     ------------------------------------------------------------------------
>     > Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls.
>     Great
>     > rates starting at 1???/min.
>     >
>     >
>     >
>     >
>     >
>     >------------------------------------------------------------------------
>     >
>     >_______________________________________________
>     >p2p-sip mailing list
>     >p2p-sip at cs.columbia.edu
>     >https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip
>     >
>     >
>
>
> ------------------------------------------------------------------------
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great 
> rates starting at 1?/min. 
> 



From: Eunsoo Shim <eunsoo at research.panasonic.com>
CC: p2p-sip at cs.columbia.edu
To: "Krishna Sankar (ksankar)" <ksankar at cisco.com>
Date: Thu, 30 Mar 2006 16:38:40 -0500
Subject: Re: [p2p-sip] Questions on #5 "A P2P SIP Arch - 2 layer approach".

 Krishna Sankar (ksankar) wrote:

>Folks,
>
> Have some quick questions on slide #5 "A P2P SIP Arch - 2 layer
>approach". I assume:
>
> a) This is could be a network topology, definitely not a
>logical topology.
> b) ONs & SNs are all SIP end-points and will be hashed as
>per Chord algorithm.
>  
>
ON and SN are NOT SIP entities.
ONs do not participate in DHT.

> c) So even though there could be a visibility and
>reachability association between specific SNs and ONs, the P2P substrate
>will not reflect this, as the hash could place them in different areas
>in the ring. In fact, if the hash didn't, it will degenerate to a
>routing table (not that that is bad, but that is not what we want to
>achieve at a macro level)
>  
>
ONs are not in the (Chord) ring.

> d) Hopefully this diagram is not overloaded with NAT
>traversal as well. In that case, the SN will represent the ONs which
>hang off of it. In this case the hashing will happen on the ON name but
>the IP address will be that of the associated SN. The SN knows how to
>get to the Ones it represents.
>  
>
The diagram is not overloaded with NAT traversal.

> e) Extending a) further, the Chord ring representation of
>this diagram will be very different.
>  
>
Please see the answers in the above.

Thanks.

Eunsoo

_______________________________________________
p2p-sip mailing list
p2p-sip at cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/p2p-sip


		
---------------------------------
Yahoo! Messenger with Voice. Make PC-to-Phone Calls to the US (and 30+ countries) for 2?/min or less.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.cs.columbia.edu/pipermail/p2p-sip/attachments/20060330/daa9d819/attachment-0001.html

