From daemon@optimus.ietf.org  Sat Jun  1 22:19:11 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA00122
	for <ipoverib-archive@odin.ietf.org>; Sat, 1 Jun 2002 22:19:11 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id WAA00338
	for ipoverib-archive@odin.ietf.org; Sat, 1 Jun 2002 22:19:37 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id WAA00158;
	Sat, 1 Jun 2002 22:10:22 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id SAA22775
	for <ipoverib@optimus.ietf.org>; Sat, 1 Jun 2002 18:06:45 -0400 (EDT)
Received: from pacelinesystems.com (faux-un.pacelinesystems.com [4.17.165.180])
	by ietf.org (8.9.1a/8.9.1a) with SMTP id SAA27999
	for <ipoverib@ietf.org>; Sat, 1 Jun 2002 18:06:16 -0400 (EDT)
Received: (qmail 755 invoked by uid 1007); 1 Jun 2002 22:06:14 -0000
Received: from hal@pacelinesystems.com by lully.pacelinesystems.com
	 by uid 1004 with qmail-scanner-1.10 (uvscan: v4.1.40/v4205. . Clear:0. Processed in 0.160677 secs); 01 Jun 2002 22:06:14 -0000
Received: from unknown (HELO dopey.pacelinesystems.com) (10.0.1.5)
  by 0 with SMTP; 1 Jun 2002 22:06:14 -0000
Received: by dopey.pacelinesystems.com with Internet Mail Service (5.5.1960.3)
	id <MDA0B7ZZ>; Sat, 1 Jun 2002 18:04:23 -0400
Message-ID: <85749B052B61D44097CE7B350C120F2C059DF4@dopey.pacelinesystems.com>
From: "Rosenstock, Hal" <hal@pacelinesystems.com>
To: "'H.K. Jerry Chu'" <Jerry.Chu@eng.sun.com>, ipoverib@ietf.org,
        roy.k.larsen@intel.com
Subject: RE: [Ipoverib] moving forward
Date: Sat, 1 Jun 2002 18:04:22 -0400 
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.1960.3)
Content-Type: text/plain;
	charset="iso-8859-1"
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

Note that IB is big endian bytes and little endian bit fields. I don't know
if you specify an IBA bit fields. See vol 1.0a p.45 Figure 4 (Byte Order
Example) and Figure 5 (Bit Order Example).

-- Hal

-----Original Message-----
From: H.K. Jerry Chu [mailto:Jerry.Chu@eng.sun.com]
Sent: Friday, May 31, 2002 4:23 PM
To: ipoverib@ietf.org; roy.k.larsen@intel.com
Subject: RE: [Ipoverib] moving forward


>I think the encapsulation draft is in good shape.  The only remaining
>clarification I would like to see is explicit memory layout in those
figures
>containing more than one byte.  For example, in figure 3, is the Reserved
>byte MSB or LSB? 

All the figures are in network byte-order (big endian).

>I have been assuming it has been MSB->LSB but that isn't
>true for figure 2.

I'm confused. I thought all the IB spec. follow the networking
convention too.

>Do we also need to make an explicit statement about
>endian-ness for multi-byte values?

No. The convention is network byte order.

Jerry

>
>WRT multicast, I don't think we have much choice other than to proceed as
>you suggest.  For those that want to make forward progress in development
>until conformant CAs and SMs are available, we can temporarily equate
>multicast to broadcast.
>
>Roy
>
>-----Original Message-----
>From: H.K. Jerry Chu [mailto:Jerry.Chu@eng.sun.com]
>Sent: Wednesday, May 29, 2002 9:32 PM
>To: ipoverib@ietf.org
>Subject: [Ipoverib] moving forward
>
>
>For the past month or so IBTA has been working on resolving
>a number of multicast related IB issues (summarized below)
>that affect how IPoIB operates.
>
>1. a "join-to-send" operation that is required for a multicast
>sender that is not already a member of the target multicast
>group.
>
>2. a facility to enable an IB client (e.g. the IPoIB driver) to
>be notified when a new IB multicast group is created, or an
>existing one is deleted.
>
>3. a facility to allow the software emulation of the promiscuous
>multicast mode.


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Thu Jun  6 20:52:29 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA21394
	for <ipoverib-archive@odin.ietf.org>; Thu, 6 Jun 2002 20:52:28 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id UAA11723
	for ipoverib-archive@odin.ietf.org; Thu, 6 Jun 2002 20:52:59 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id UAA11147;
	Thu, 6 Jun 2002 20:43:25 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id UAA11114
	for <ipoverib@optimus.ietf.org>; Thu, 6 Jun 2002 20:43:23 -0400 (EDT)
Received: from e1.ny.us.ibm.com (e1.ny.us.ibm.com [32.97.182.101])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA20818
	for <ipoverib@ietf.org>; Thu, 6 Jun 2002 20:42:52 -0400 (EDT)
Received: from northrelay03.pok.ibm.com (northrelay03.pok.ibm.com [9.56.224.151])
	by e1.ny.us.ibm.com (8.12.2/8.12.2) with ESMTP id g570gqg5228168;
	Thu, 6 Jun 2002 20:42:52 -0400
Received: from gateway.beaverton.ibm.com (gateway.beaverton.ibm.com [138.95.180.1])
	by northrelay03.pok.ibm.com (8.11.1m3/NCO/VER6.1) with ESMTP id g570gjH98274;
	Thu, 6 Jun 2002 20:42:49 -0400
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway.beaverton.ibm.com (8.10.0.Beta10/8.11.6) with ESMTP id g570gie13841;
	Thu, 6 Jun 2002 17:42:45 -0700 (PDT)
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g570gh525077;
	Thu, 6 Jun 2002 17:42:43 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206070042.g570gh525077@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: Jerry.Chu@eng.sun.com
Date: Thu, 6 Jun 2002 17:42:41 -0700 (PDT)
Cc: ipoverib@ietf.org
In-Reply-To: <200206010303.g5133t6U602310@jurassic.eng.sun.com> from "H.K. Jerry Chu" at May 31, 2002 07:05:42 PM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

> 
> A couple of weeks ago I asked IBTA for a summary to the IPoIB
> reflector on the changes likely to come as part of IBA 1.1
> spec later this year that will affect IPoIB. Unfortunately,
> the request was declined because IBTA rules forbid any 1.1
> info to go outside of member companies at this point. But
> based on my knowledge of their progress I felt comfortable to
> move the current drafts forward without risking incompatibility
> with IBA 1.1. If the latter does happen, we can always amend
> the drafts then.
> 
> But yesterday I heard from David Brean of IBTA/MGTWG that the
> item 3
> 
> a facility to allow the software emulation of the promiscuous
> multicast mode
> 
> remains unresolved within IBTA. Vivek also raised some concern
> too. IMO it is probably not worth holding the drafts for a single
> issue. But I decided to give it another try to resolve this issue
> in the next few days. If we can't reach a consensus before next
> Friday, I'd suggest we move the rest of the drafts forward
> leaving this issue behind to be addressed later.

        Are you suggesting that we define the algorithm based on
        points (1) and (2) ignoring (3) or are you suggesting that we
        define IPoIB multicast based on the current IB specification ? 


> 
> Let me restate that the promiscuous multicast mode is not only
> required for the proper operation of IP group mamagement protocol,
> it is also how a sender's multicast pkt gets picked up by an IP
> multicast router. The details are described in
> 
> rfc1112 -
> "Note that the multicast routers receive all IP multicast
> datagrams, and therefore need not be addressed explicitly..."
> 
> rfc2710 -
> "For each interface over which the router is operating the MLD
> protocol, the router must configure that interface to listen to all
> link-layer multicast address that can be generated by IPv6
> multicasts..."
> 
> IBA 1.0 spec. does not provide a promiscuous multicast mode in
> hardware or allow one to be emulated in software. So far there
> have been three proposals to address this issue:
> 
> 1. Require the IPoIB driver to send a copy of every outgoing
> multicast pkt to the all router multicast group. The algorithm
> looks like this: for a pkt destined to a multicast group M
> 
> 	if (M doesn't exist)
> 		send pkt to all_router
> 	else if (M != all_router)
> 		send pkt to all_router;
> 		send pkt to M;
> 	else send pkt to M;
> 		
> 2. Modify IP group management protocol module to do some extra
> work in case the link type is IB. An IP router will explicitly
> join any new multicast group that comes into existence. The
> router needs to figure out when to leave too (details from
> Vivek).
> 
> 3. Ask IBTA to add a facility to allow promiscuous multicast
> mode to be emulated in software.

        And a 4th would be to use only one multicast group at the IB
        layer for all IP multicast packets such as use broadcast_GID
        for all IP multicast. Not a good solution but an option.

> 
> 1. above seems to be the least painful. But it requires every
> node to be capable of sending pkts to all_router multicast
> group.

        The most bothersome issue of this option is the deliberate
        insertion of duplicates into the network. A big no-no in my
        opinion. 

> 
> The biggest problem with 2. IMO is that it requires changes to
> existing IP code in every version of group management code
> including IGMPv1, IGMPv2, MLD..., etc, and the change doesn't
> seem trivial. It may present a big obstacle for many OSes to
> request IB only changes in generic IP code.

        That is not the case. 

        In a generic implementation, the routing daemon (mrouted/gated
        for example) updates the kernel routing and multicast group
        tables. This is done since the IP forwarding code looks into
        this table when making its decisions.

        The multicast group information is always linked to the
        interface it is relevant to (ie. the subnet the multicast
        group is on). All one has to do is to add code that looks at
        the interface type and takes interface specific action(s). In
        the case of IPoIB it would join the corresponding IB MGIDs.

        Such a change is no different than adding an IB specific
        module. It does not effect the IGMPv1/v2/MLD code.


> >From what I've heard 3 should be simple to add. The SA has all
> the MC info already. It simply needs to supply the caller a
> list of all the existing MC groups, and notify the caller of
> any new group coming into existence so that the caller (driver
> of an IP router) can ensure it "join to observe" every
> multicast group.

        I am unclear on the meaning of 'join to observe'. A router may
        have to send packets out on the MGID - so it will have to
        'join' for receive and send.

	
> 
> Please express your opinion if you have one in the next few
> days. If we can't get to a consensus I'd say we shelf it and
> proceed with the rest of stuff. Note that since the
> encapsulation draft has a dependency on the link/multicast
> draft, it can't move forward alone. Also I don't want to wait

	The link-multicast draft defines the :

		- mapping from IP multicast to IB MGID
		- the definition of the IPoIB link

        We need to define the multicast forwarding methodology. In the
        interest of forward movement I suggest we do one of the
        following: 

        a) Add the mapping and link-definition pieces to the
        encapsulation draft. These two items are quite straigtforward
        and simple enough. Define the multicast forwarding later when
        IBTA publishes its specification.

        b) Publish the link-multicast draft now without multicast
        forwarding. It will define only the mapping and the notion of
        IPoIB link. The mutlicast forwarding solution is published
        later when IBTA specification comes out.

        We can live without having IPoIB multicast routing defined
        until september. 

> until September for the IBA 1.1 spec. We have other work to do.

        Does the other work include advanced capabilities (and
        DHCP/MIB)? 

Vivek

> 
> Jerry
> 
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 


-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)
-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Thu Jun  6 23:15:33 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA00320
	for <ipoverib-archive@odin.ietf.org>; Thu, 6 Jun 2002 23:15:32 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id XAA19041
	for ipoverib-archive@odin.ietf.org; Thu, 6 Jun 2002 23:16:04 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id XAA18885;
	Thu, 6 Jun 2002 23:09:13 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id XAA18851
	for <ipoverib@optimus.ietf.org>; Thu, 6 Jun 2002 23:09:12 -0400 (EDT)
Received: from pheriche.sun.com (pheriche.sun.com [192.18.98.34])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA29878
	for <ipoverib@ietf.org>; Thu, 6 Jun 2002 23:08:39 -0400 (EDT)
Received: from jurassic.eng.sun.com ([129.146.17.55])
	by pheriche.sun.com (8.9.3+Sun/8.9.3) with ESMTP id VAA29017;
	Thu, 6 Jun 2002 21:09:11 -0600 (MDT)
Received: from taipei (taipei.Eng.Sun.COM [129.146.85.178])
	by jurassic.eng.sun.com (8.12.4+Sun/8.12.4) with SMTP id g5739AAw804392;
	Thu, 6 Jun 2002 20:09:10 -0700 (PDT)
Message-Id: <200206070309.g5739AAw804392@jurassic.eng.sun.com>
Date: Thu, 6 Jun 2002 20:09:05 -0700 (PDT)
From: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Reply-To: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: kashyapv@us.ibm.com
Cc: ipoverib@ietf.org
MIME-Version: 1.0
Content-Type: TEXT/plain; charset=us-ascii
Content-MD5: 05cDSUc9key41StmtkNjeg==
X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4.6_06 SunOS 5.8 sun4u sparc 
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

>> A couple of weeks ago I asked IBTA for a summary to the IPoIB
>> reflector on the changes likely to come as part of IBA 1.1
>> spec later this year that will affect IPoIB. Unfortunately,
>> the request was declined because IBTA rules forbid any 1.1
>> info to go outside of member companies at this point. But
>> based on my knowledge of their progress I felt comfortable to
>> move the current drafts forward without risking incompatibility
>> with IBA 1.1. If the latter does happen, we can always amend
>> the drafts then.
>> 
>> But yesterday I heard from David Brean of IBTA/MGTWG that the
>> item 3
>> 
>> a facility to allow the software emulation of the promiscuous
>> multicast mode
>> 
>> remains unresolved within IBTA. Vivek also raised some concern
>> too. IMO it is probably not worth holding the drafts for a single
>> issue. But I decided to give it another try to resolve this issue
>> in the next few days. If we can't reach a consensus before next
>> Friday, I'd suggest we move the rest of the drafts forward
>> leaving this issue behind to be addressed later.
>
>        Are you suggesting that we define the algorithm based on
>        points (1) and (2) ignoring (3) or are you suggesting that we
>        define IPoIB multicast based on the current IB specification ? 

Confused. I laid down three known and mutually exclusively proposals
and asked for people's opinion. Personally I prefer (3) because
I heard from some members of IBTA that (3) is not difficult at all,
and it makes promiscuous MC mode a non-issue for IPoIB. If we reach
consensus on (3) then we need to apply persuasion power to IBTA.

>
>
>> 
>> Let me restate that the promiscuous multicast mode is not only
>> required for the proper operation of IP group mamagement protocol,
>> it is also how a sender's multicast pkt gets picked up by an IP
>> multicast router. The details are described in
>> 
>> rfc1112 -
>> "Note that the multicast routers receive all IP multicast
>> datagrams, and therefore need not be addressed explicitly..."
>> 
>> rfc2710 -
>> "For each interface over which the router is operating the MLD
>> protocol, the router must configure that interface to listen to all
>> link-layer multicast address that can be generated by IPv6
>> multicasts..."
>> 
>> IBA 1.0 spec. does not provide a promiscuous multicast mode in
>> hardware or allow one to be emulated in software. So far there
>> have been three proposals to address this issue:
>> 
>> 1. Require the IPoIB driver to send a copy of every outgoing
>> multicast pkt to the all router multicast group. The algorithm
>> looks like this: for a pkt destined to a multicast group M
>> 
>> 	if (M doesn't exist)
>> 		send pkt to all_router
>> 	else if (M != all_router)
>> 		send pkt to all_router;
>> 		send pkt to M;
>> 	else send pkt to M;
>> 		
>> 2. Modify IP group management protocol module to do some extra
>> work in case the link type is IB. An IP router will explicitly
>> join any new multicast group that comes into existence. The
>> router needs to figure out when to leave too (details from
>> Vivek).
>> 
>> 3. Ask IBTA to add a facility to allow promiscuous multicast
>> mode to be emulated in software.
>
>        And a 4th would be to use only one multicast group at the IB
>        layer for all IP multicast packets such as use broadcast_GID
>        for all IP multicast. Not a good solution but an option.

I don't think it is adequate for a multicast draft to advocate
something anti-multicast.

>
>> 
>> 1. above seems to be the least painful. But it requires every
>> node to be capable of sending pkts to all_router multicast
>> group.
>
>        The most bothersome issue of this option is the deliberate
>        insertion of duplicates into the network. A big no-no in my
>        opinion. 

Still don't understand why this is a big no-no. In a subnet of 100
nodes and 2 listeners of MC group A, a multicast to A of 2 nodes
plus a separate send to router R is still a vast improvement over
broadcast over 100 nodes.

But like I said I prefer (3) if it doens't cost much to IBA.

>
>> 
>> The biggest problem with 2. IMO is that it requires changes to
>> existing IP code in every version of group management code
>> including IGMPv1, IGMPv2, MLD..., etc, and the change doesn't
>> seem trivial. It may present a big obstacle for many OSes to
>> request IB only changes in generic IP code.
>
>        That is not the case.

What is not the case? "Requiring changes to existing IP code" or
"the change doesn't seem trivial" or "a big obstacle for OSes"?
The latter two can be very subjective.

>
>        In a generic implementation, the routing daemon (mrouted/gated
>        for example) updates the kernel routing and multicast group
>        tables. This is done since the IP forwarding code looks into
>        this table when making its decisions.
>
>        The multicast group information is always linked to the
>        interface it is relevant to (ie. the subnet the multicast
>        group is on). All one has to do is to add code that looks at
>        the interface type and takes interface specific action(s). In
>        the case of IPoIB it would join the corresponding IB MGIDs.
>
>        Such a change is no different than adding an IB specific
>        module. It does not effect the IGMPv1/v2/MLD code.
>
>
>> >From what I've heard 3 should be simple to add. The SA has all
>> the MC info already. It simply needs to supply the caller a
>> list of all the existing MC groups, and notify the caller of
>> any new group coming into existence so that the caller (driver
>> of an IP router) can ensure it "join to observe" every
>> multicast group.
>
>        I am unclear on the meaning of 'join to observe'. A router may
>        have to send packets out on the MGID - so it will have to
>        'join' for receive and send.

For each active MC group (with > 0 member) the router needs to
join in order to receive. But the regular "join to receive" will
count the router as a real member, which must not be the case (otherwise
the MC group won't ever be deleted because its member count will
never drop to 0.) Thus we'll need a special "join to observe" that
still adds the caller to the list of listeners but doesn't cause the
member count maintained by SA to increment.

>
>	
>> 
>> Please express your opinion if you have one in the next few
>> days. If we can't get to a consensus I'd say we shelf it and
>> proceed with the rest of stuff. Note that since the
>> encapsulation draft has a dependency on the link/multicast
>> draft, it can't move forward alone. Also I don't want to wait
>
>	The link-multicast draft defines the :
>
>		- mapping from IP multicast to IB MGID
>		- the definition of the IPoIB link
>
>        We need to define the multicast forwarding methodology. In the
>        interest of forward movement I suggest we do one of the
>        following: 
>
>        a) Add the mapping and link-definition pieces to the
>        encapsulation draft. These two items are quite straigtforward
>        and simple enough. Define the multicast forwarding later when
>        IBTA publishes its specification.
>
>        b) Publish the link-multicast draft now without multicast
>        forwarding. It will define only the mapping and the notion of
>        IPoIB link. The mutlicast forwarding solution is published
>        later when IBTA specification comes out.
>
>        We can live without having IPoIB multicast routing defined
>        until september. 
>
>> until September for the IBA 1.1 spec. We have other work to do.
>
>        Does the other work include advanced capabilities (and
>        DHCP/MIB)? 

Yep.

Jerry

>
>Vivek
>
>> 
>> Jerry
>> 
>> 
>> _______________________________________________
>> IPoverIB mailing list
>> IPoverIB@ietf.org
>> https://www1.ietf.org/mailman/listinfo/ipoverib
>> 
>
>
>-- 
>Vivek Kashyap
>Linux Technology Center, IBM
>kashyapv@us.ibm.com 
>vivk@us.ibm.com 
>503 578 3422 (o)
>-- 
>Vivek Kashyap
>Linux Technology Center, IBM
>kashyapv@us.ibm.com 
>vivk@us.ibm.com 
>503 578 3422 (o)


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Fri Jun  7 16:33:52 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA11813
	for <ipoverib-archive@odin.ietf.org>; Fri, 7 Jun 2002 16:33:52 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id QAA03680
	for ipoverib-archive@odin.ietf.org; Fri, 7 Jun 2002 16:34:24 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id QAA02898;
	Fri, 7 Jun 2002 16:26:38 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id QAA02857
	for <ipoverib@optimus.ietf.org>; Fri, 7 Jun 2002 16:26:34 -0400 (EDT)
Received: from e1.ny.us.ibm.com (e1.ny.us.ibm.com [32.97.182.101])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA11624
	for <ipoverib@ietf.org>; Fri, 7 Jun 2002 16:26:01 -0400 (EDT)
Received: from westrelay03.boulder.ibm.com (westrelay03.boulder.ibm.com [9.17.194.24])
	by e1.ny.us.ibm.com (8.12.2/8.12.2) with ESMTP id g57KQ1g5107284;
	Fri, 7 Jun 2002 16:26:01 -0400
Received: from gateway1.beaverton.ibm.com (gateway1.beaverton.ibm.com [138.95.180.2])
	by westrelay03.boulder.ibm.com (8.11.1m3/NCO/VER6.1) with ESMTP id g57KQ1I196058;
	Fri, 7 Jun 2002 14:26:01 -0600
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway1.beaverton.ibm.com (8.11.6/8.11.6) with ESMTP id g57KPs917459;
	Fri, 7 Jun 2002 13:25:54 -0700
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g57KPxf10627;
	Fri, 7 Jun 2002 13:25:59 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206072025.g57KPxf10627@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: Jerry.Chu@eng.sun.com
Date: Fri, 7 Jun 2002 13:25:58 -0700 (PDT)
Cc: ipoverib@ietf.org
In-Reply-To: <200206070309.g5739AAw804392@jurassic.eng.sun.com> from "H.K. Jerry Chu" at Jun 06, 2002 07:09:05 PM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit


<snip>
> >> 
> >> But yesterday I heard from David Brean of IBTA/MGTWG that the
> >> item 3
> >> 

<snip>

> >
> >        Are you suggesting that we define the algorithm based on
> >        points (1) and (2) ignoring (3) or are you suggesting that we
> >        define IPoIB multicast based on the current IB specification ? 
> 
> Confused. I laid down three known and mutually exclusively proposals
> and asked for people's opinion. Personally I prefer (3) because
> I heard from some members of IBTA that (3) is not difficult at all,
> and it makes promiscuous MC mode a non-issue for IPoIB. If we reach
> consensus on (3) then we need to apply persuasion power to IBTA.

        I was referring to the 3 changes to IB multicast definition
        you mentioned. So I was asking, since it is not clear from
        your comments, whether you meant 'ignore item 3 but use 1 & 2'
        or did you mean ignore them all and work as per the current
        specification. 


<snip>

> >
> >        And a 4th would be to use only one multicast group at the IB
> >        layer for all IP multicast packets such as use broadcast_GID
> >        for all IP multicast. Not a good solution but an option.
> 
> I don't think it is adequate for a multicast draft to advocate
> something anti-multicast.

	Totally agree that this is not a good option - as I noted
	earlier. However, for the sake of completion, there are 4
	options (good or bad).
> 
> >
> >> 
> >> 1. above seems to be the least painful. But it requires every
> >> node to be capable of sending pkts to all_router multicast
> >> group.
> >
> >        The most bothersome issue of this option is the deliberate
> >        insertion of duplicates into the network. A big no-no in my
> >        opinion. 
> 
> Still don't understand why this is a big no-no. In a subnet of 100
> nodes and 2 listeners of MC group A, a multicast to A of 2 nodes
> plus a separate send to router R is still a vast improvement over
> broadcast over 100 nodes.

        I am not comparing it to the broadcast case. I was wondering
        about the following situation:

        How do one ensure that the router is not a member of any IB
        group ? What mechanism stops it from being a member ? Many
        nodes do double duty - they host applications and also act as
        routers - how do they decide?

        As per this solution a router will receive two packets and if
        the IP subnet has multiple routers then multiple routers get
        two packets. If the corresponding IP group has a scope larger
        than the local subnet - each of these routers will now forward
        multiple packets outward. If one or more of these neighbouring
        subnets are IPoIB subnets we can further multiply the number
        of copies by a factor of two at the least. The result would be
        that IPoIB subnets would end up creating multicast storms
        quite often.

        I have not given it more thought than this but I can imagine
        multicast services receiving multiple queries for the same
        request. I have no idea how that would work - probably just
        add more load and nothing else but that is bad too.

        The chances of such errors are greater when one inserts
        duplicate packets into the network expecting the routers to
        forward only one copy.

> 
> But like I said I prefer (3) if it doens't cost much to IBA.
> 
> >
> >> 
> >> The biggest problem with 2. IMO is that it requires changes to
> >> existing IP code in every version of group management code
> >> including IGMPv1, IGMPv2, MLD..., etc, and the change doesn't
> >> seem trivial. It may present a big obstacle for many OSes to
> >> request IB only changes in generic IP code.
> >
> >        That is not the case.
> 
> What is not the case? "Requiring changes to existing IP code" or
> "the change doesn't seem trivial" or "a big obstacle for OSes"?
> The latter two can be very subjective.

        Take a look at the description below. The change is a simple
        function in the driver with a line or two added (considering
        BSD code) to the 'add multicast group' socket option.

        The suggestion works within the existing infrastructure
        with a minor addition.

        In existing implmementation the routing daemon on receiving an
        IGMP report adds the IP multicast group to the interface list.
        It tells the kernel to add the group by a socket option (in
        BSD/LInux implementations). The function after it adds the IP
        group to the interface can call the interface's device
        specific function - for IPoIB this function will ask the
        driver to join the IB group.

	(IGMP_report)--- n/w -> driver -> IGMP -> mrouted/gated
							|
						    setsockopt(ADD_LGRP)
							|
							:
							:
							add_grp()
							|
						** call device_add_grp() **
						** join relevant IB group **
						** if needed            **

        Only the call marked with '**' is new. Nothing else. Note that
        the same function or similar function will have to exist on
        hosts for IP_ADD_MEMBERSHIP directive when a host application
        joins a particular IP group. In this case the router does the
        same thing on receiving an IGMP_HOST_MEMBERSHIP_REPORT.

        The change is simple and does not require changes to 'every
        version of group mgmt code'. It does not require changes to
        routing daemons, nor does it imply large, complicated changes
        in OSes.

> 
> >
> >        In a generic implementation, the routing daemon (mrouted/gated
> >        for example) updates the kernel routing and multicast group
> >        tables. This is done since the IP forwarding code looks into
> >        this table when making its decisions.
> >
> >        The multicast group information is always linked to the
> >        interface it is relevant to (ie. the subnet the multicast
> >        group is on). All one has to do is to add code that looks at
> >        the interface type and takes interface specific action(s). In
> >        the case of IPoIB it would join the corresponding IB MGIDs.
> >
> >        Such a change is no different than adding an IB specific
> >        module. It does not effect the IGMPv1/v2/MLD code.
> >
> >
> >> >From what I've heard 3 should be simple to add. The SA has all
> >> the MC info already. It simply needs to supply the caller a
> >> list of all the existing MC groups, and notify the caller of
> >> any new group coming into existence so that the caller (driver
> >> of an IP router) can ensure it "join to observe" every
> >> multicast group.
> >
> >        I am unclear on the meaning of 'join to observe'. A router may
> >        have to send packets out on the MGID - so it will have to
> >        'join' for receive and send.
> 
> For each active MC group (with > 0 member) the router needs to
> join in order to receive. But the regular "join to receive" will
> count the router as a real member, which must not be the case (otherwise
> the MC group won't ever be deleted because its member count will
> never drop to 0.) Thus we'll need a special "join to observe" that
> still adds the caller to the list of listeners but doesn't cause the
> member count maintained by SA to increment.

	That is a thought but it has some complications. This would be
	yet another kind of 'join' - not just 
		'send-only-but-not-receive' or 'send-and-receive' 
		but 
		'send-and-receive-but-not-be-counted-a-member'.


	Well it is upto IBTA to define all this but it is getting
        messy. In my view, as far as IPoIB is concerned, the routers
        can leave when the routing daemon indicates that the kernel
        can delete the IP group. It becomes a simple side-effect in
        the kernel. I feel that is a much simpler approach.

        The alternative requires the MADs to be handled; then passed
        to the driver QP or passed to an IPoIB-specific IB agent that
        then (exactly like the scheme with mrouted daemon) asks the
        driver to join/leave the IB groups. It also requires changes
        to IBTA that may or may not make into the 1.1 specification in
        september. 

> 
> >
> >	
> >> 
> >> Please express your opinion if you have one in the next few
> >> days. If we can't get to a consensus I'd say we shelf it and
> >> proceed with the rest of stuff. Note that since the
> >> encapsulation draft has a dependency on the link/multicast
> >> draft, it can't move forward alone. Also I don't want to wait
> >
> >	The link-multicast draft defines the :
> >
> >		- mapping from IP multicast to IB MGID
> >		- the definition of the IPoIB link
> >
> >        We need to define the multicast forwarding methodology. In the
> >        interest of forward movement I suggest we do one of the
> >        following: 
> >
> >        a) Add the mapping and link-definition pieces to the
> >        encapsulation draft. These two items are quite straigtforward
> >        and simple enough. Define the multicast forwarding later when
> >        IBTA publishes its specification.
> >
> >        b) Publish the link-multicast draft now without multicast
> >        forwarding. It will define only the mapping and the notion of
> >        IPoIB link. The mutlicast forwarding solution is published
> >        later when IBTA specification comes out.
> >
> >        We can live without having IPoIB multicast routing defined
> >        until september. 

        Don't you agree that option a) or b) above is the best way to
        move forward?

> >
> >> until September for the IBA 1.1 spec. We have other work to do.
> >
> >        Does the other work include advanced capabilities (and
> >        DHCP/MIB)? 
> 
> Yep.

	OK. In that case I'll update the advanced capabilities draft
	and also plan to give a presentation on it to move the
	discussion forward. 

Vivek

-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Mon Jun 10 10:14:31 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA05173
	for <ipoverib-archive@odin.ietf.org>; Mon, 10 Jun 2002 10:14:31 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id KAA05841
	for ipoverib-archive@odin.ietf.org; Mon, 10 Jun 2002 10:15:05 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id KAA05516;
	Mon, 10 Jun 2002 10:07:38 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id KAA05486
	for <ipoverib@optimus.ietf.org>; Mon, 10 Jun 2002 10:07:37 -0400 (EDT)
Received: from palrel10.hp.com (palrel10.hp.com [156.153.255.245])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA04857
	for <ipoverib@ietf.org>; Mon, 10 Jun 2002 10:07:02 -0400 (EDT)
Received: from psalpha0.cup.hp.com (psalpha0.cup.hp.com [15.13.189.230])
	by palrel10.hp.com (Postfix) with ESMTP
	id 2EBCEC008A1; Mon, 10 Jun 2002 07:07:01 -0700 (PDT)
Received: from mkrause.cup.hp.com ([15.113.161.2])
	by psalpha0.cup.hp.com (8.9.3/8.8.6) with ESMTP id HAA29557;
	Mon, 10 Jun 2002 07:07:00 -0700 (PDT)
Message-Id: <5.1.0.14.2.20020610070103.03740c10@esalpha2.cup.hp.com>
X-Sender: krause@esalpha2.cup.hp.com
X-Mailer: QUALCOMM Windows Eudora Version 5.1
Date: Mon, 10 Jun 2002 07:05:53 -0700
To: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>, ipoverib@ietf.org
From: Michael Krause <krause@cup.hp.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
In-Reply-To: <200206010303.g5133t6U602310@jurassic.eng.sun.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

At 08:05 PM 5/31/2002 -0700, H.K. Jerry Chu wrote:

>IBA 1.0 spec. does not provide a promiscuous multicast mode in
>hardware or allow one to be emulated in software. So far there
>have been three proposals to address this issue:
>
>1. Require the IPoIB driver to send a copy of every outgoing
>multicast pkt to the all router multicast group. The algorithm
>looks like this: for a pkt destined to a multicast group M
>
>         if (M doesn't exist)
>                 send pkt to all_router
>         else if (M != all_router)
>                 send pkt to all_router;
>                 send pkt to M;
>         else send pkt to M;

Strongly oppose this as previously discussed many months back.

>
>2. Modify IP group management protocol module to do some extra
>work in case the link type is IB. An IP router will explicitly
>join any new multicast group that comes into existence. The
>router needs to figure out when to leave too (details from
>Vivek).

This does not require any change to the IP router code - the router joins 
any multicast group through event notification if it and group are in the 
same partition.

>3. Ask IBTA to add a facility to allow promiscuous multicast
>mode to be emulated in software.

Don't see anything that must be added by the IBTA to allow this to work.


>1. above seems to be the least painful. But it requires every
>node to be capable of sending pkts to all_router multicast
>group.
>
>The biggest problem with 2. IMO is that it requires changes to
>existing IP code in every version of group management code
>including IGMPv1, IGMPv2, MLD..., etc, and the change doesn't
>seem trivial. It may present a big obstacle for many OSes to
>request IB only changes in generic IP code.
>
> >From what I've heard 3 should be simple to add. The SA has all
>the MC info already. It simply needs to supply the caller a
>list of all the existing MC groups, and notify the caller of
>any new group coming into existence so that the caller (driver
>of an IP router) can ensure it "join to observe" every
>multicast group.

I believe group event notification is supposed to be defined as part of 
V1.1.  There is no reason to have an observe capability.

Recommend just proceeding with the other work and waiting until V1.1 which 
solves many of the multicast issues required and also solves a number of 
errata from IB as a whole.  This IB version, IMO, what will be deployed as 
volume product in 2003 so aligning with its functionality bodes best for 
interoperability and minimizing development costs.

Mike


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Mon Jun 10 10:14:40 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA05190
	for <ipoverib-archive@odin.ietf.org>; Mon, 10 Jun 2002 10:14:35 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id KAA05855
	for ipoverib-archive@odin.ietf.org; Mon, 10 Jun 2002 10:15:09 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id KAA05472;
	Mon, 10 Jun 2002 10:07:34 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id KAA05441
	for <ipoverib@optimus.ietf.org>; Mon, 10 Jun 2002 10:07:32 -0400 (EDT)
Received: from palrel12.hp.com (palrel12.hp.com [156.153.255.237])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA04853
	for <ipoverib@ietf.org>; Mon, 10 Jun 2002 10:06:57 -0400 (EDT)
Received: from psalpha0.cup.hp.com (psalpha0.cup.hp.com [15.13.189.230])
	by palrel12.hp.com (Postfix) with ESMTP
	id DED02E00A1B; Mon, 10 Jun 2002 07:06:59 -0700 (PDT)
Received: from mkrause.cup.hp.com ([15.113.161.2])
	by psalpha0.cup.hp.com (8.9.3/8.8.6) with ESMTP id HAA29553;
	Mon, 10 Jun 2002 07:06:59 -0700 (PDT)
Message-Id: <5.1.0.14.2.20020610065509.03730240@esalpha2.cup.hp.com>
X-Sender: krause@esalpha2.cup.hp.com
X-Mailer: QUALCOMM Windows Eudora Version 5.1
Date: Mon, 10 Jun 2002 06:59:05 -0700
To: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>, kashyapv@us.ibm.com
From: Michael Krause <krause@cup.hp.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
Cc: ipoverib@ietf.org
In-Reply-To: <200206070309.g5739AAw804392@jurassic.eng.sun.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

At 08:09 PM 6/6/2002 -0700, H.K. Jerry Chu wrote:


>For each active MC group (with > 0 member) the router needs to
>join in order to receive. But the regular "join to receive" will
>count the router as a real member, which must not be the case (otherwise
>the MC group won't ever be deleted because its member count will
>never drop to 0.) Thus we'll need a special "join to observe" that
>still adds the caller to the list of listeners but doesn't cause the
>member count maintained by SA to increment.

Routers send / receive multicast packets thus are real members just like 
any other endnode within an IB subnet especially if they are using the 
subnet as an intermediate hop along the way to other subnets and the 
multicast group spans all subnets even if not locally used by any other 
endnode.  The SA would need to identify the router and determine whether 
there is anyone using this service to determine whether the router should 
be told to go away.

Mike


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@ns.ietf.org  Mon Jun 10 13:41:38 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA15671
	for <ipoverib-archive@odin.ietf.org>; Mon, 10 Jun 2002 13:41:38 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id NAA18363
	for ipoverib-archive@odin.ietf.org; Mon, 10 Jun 2002 13:40:55 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA18137;
	Mon, 10 Jun 2002 13:33:54 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA18103
	for <ipoverib@ns.ietf.org>; Mon, 10 Jun 2002 13:33:52 -0400 (EDT)
Received: from nwkea-mail-1.sun.com (nwkea-mail-1.sun.com [192.18.42.13])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA15281
	for <ipoverib@ietf.org>; Mon, 10 Jun 2002 13:33:17 -0400 (EDT)
Received: from jurassic.eng.sun.com ([129.146.17.55])
	by nwkea-mail-1.sun.com (8.9.3+Sun/8.9.3) with ESMTP id KAA17146;
	Mon, 10 Jun 2002 10:32:57 -0700 (PDT)
Received: from taipei (taipei.Eng.Sun.COM [129.146.85.178])
	by jurassic.eng.sun.com (8.12.4+Sun/8.12.4) with SMTP id g5AHWuwU597011;
	Mon, 10 Jun 2002 10:32:56 -0700 (PDT)
Message-Id: <200206101732.g5AHWuwU597011@jurassic.eng.sun.com>
Date: Mon, 10 Jun 2002 10:32:50 -0700 (PDT)
From: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Reply-To: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: krause@cup.hp.com
Cc: ipoverib@ietf.org
MIME-Version: 1.0
Content-Type: TEXT/plain; charset=us-ascii
Content-MD5: pWkXCLonC3hDpK60IOcQZg==
X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4.6_06 SunOS 5.8 sun4u sparc 
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org


>>For each active MC group (with > 0 member) the router needs to
>>join in order to receive. But the regular "join to receive" will
>>count the router as a real member, which must not be the case (otherwise
>>the MC group won't ever be deleted because its member count will
>>never drop to 0.) Thus we'll need a special "join to observe" that
>>still adds the caller to the list of listeners but doesn't cause the
>>member count maintained by SA to increment.
>
>Routers send / receive multicast packets thus are real members just like 
>any other endnode within an IB subnet especially if they are using the 
>subnet as an intermediate hop along the way to other subnets and the 
>multicast group spans all subnets even if not locally used by any other 
>endnode.  The SA would need to identify the router and determine whether 
>there is anyone using this service to determine whether the router should 
>be told to go away.

Note that we're talking about IP routers here. How would SA at the IB
level knows that a node is performing IP forwarding function? When
an IB MC group is created because an IP MC listener comes into
existence, the IP router needs to be notified in order to join the
MC group to pick up pkts. Later when the only IP MC listener quits,
how would SA knows that there is no more "real listener" on the MC
group if the IP router is counted as a real listener? There seems
to be some confusion that IP multicast routing protocols will somehow
magically work this out. I don't think so.

If IBTA believes it will have all the necessary facility to allow
an emulation of promiscous MC mode that is great. I just don't see
how that works without the above. (And I really hope to be proven
wrong.)

Jerry

>
>Mike
>


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@ns.ietf.org  Mon Jun 10 13:51:26 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA16151
	for <ipoverib-archive@odin.ietf.org>; Mon, 10 Jun 2002 13:51:26 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id NAA18972
	for ipoverib-archive@odin.ietf.org; Mon, 10 Jun 2002 13:52:00 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA18659;
	Mon, 10 Jun 2002 13:44:46 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA18626
	for <ipoverib@ns.ietf.org>; Mon, 10 Jun 2002 13:44:44 -0400 (EDT)
Received: from kathmandu.sun.com (kathmandu.sun.com [192.18.98.36])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA15823
	for <ipoverib@ietf.org>; Mon, 10 Jun 2002 13:44:09 -0400 (EDT)
Received: from jurassic.eng.sun.com ([129.146.17.55])
	by kathmandu.sun.com (8.9.3+Sun/8.9.3) with ESMTP id LAA08304;
	Mon, 10 Jun 2002 11:44:32 -0600 (MDT)
Received: from taipei (taipei.Eng.Sun.COM [129.146.85.178])
	by jurassic.eng.sun.com (8.12.4+Sun/8.12.4) with SMTP id g5AHiWwU599855;
	Mon, 10 Jun 2002 10:44:32 -0700 (PDT)
Message-Id: <200206101744.g5AHiWwU599855@jurassic.eng.sun.com>
Date: Mon, 10 Jun 2002 10:44:26 -0700 (PDT)
From: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Reply-To: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: ipoverib@ietf.org, krause@cup.hp.com
MIME-Version: 1.0
Content-Type: TEXT/plain; charset=us-ascii
Content-MD5: VCFPGef9amFTUmsijSIGaQ==
X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4.6_06 SunOS 5.8 sun4u sparc 
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

>At 08:05 PM 5/31/2002 -0700, H.K. Jerry Chu wrote:
>
>>IBA 1.0 spec. does not provide a promiscuous multicast mode in
>>hardware or allow one to be emulated in software. So far there
>>have been three proposals to address this issue:
>>
>>1. Require the IPoIB driver to send a copy of every outgoing
>>multicast pkt to the all router multicast group. The algorithm
>>looks like this: for a pkt destined to a multicast group M
>>
>>         if (M doesn't exist)
>>                 send pkt to all_router
>>         else if (M != all_router)
>>                 send pkt to all_router;
>>                 send pkt to M;
>>         else send pkt to M;
>
>Strongly oppose this as previously discussed many months back.

I hear you :-).

>
>>
>>2. Modify IP group management protocol module to do some extra
>>work in case the link type is IB. An IP router will explicitly
>>join any new multicast group that comes into existence. The
>>router needs to figure out when to leave too (details from
>>Vivek).
>
>This does not require any change to the IP router code - the router joins 
>any multicast group through event notification if it and group are in the 
>same partition.

2 above is Vivek's proposal that requires changes to IP routing code.
The event notification is 3 below.

>
>>3. Ask IBTA to add a facility to allow promiscuous multicast
>>mode to be emulated in software.
>
>Don't see anything that must be added by the IBTA to allow this to work.

Great!

>
>
>>1. above seems to be the least painful. But it requires every
>>node to be capable of sending pkts to all_router multicast
>>group.
>>
>>The biggest problem with 2. IMO is that it requires changes to
>>existing IP code in every version of group management code
>>including IGMPv1, IGMPv2, MLD..., etc, and the change doesn't
>>seem trivial. It may present a big obstacle for many OSes to
>>request IB only changes in generic IP code.
>>
>> >From what I've heard 3 should be simple to add. The SA has all
>>the MC info already. It simply needs to supply the caller a
>>list of all the existing MC groups, and notify the caller of
>>any new group coming into existence so that the caller (driver
>>of an IP router) can ensure it "join to observe" every
>>multicast group.
>
>I believe group event notification is supposed to be defined as part of 
>V1.1.  There is no reason to have an observe capability.

Ok, I trust that you will ensure all the facilities will be there
in V1.1 for an emulation of promiscuous MC mode to be used by IPoIB.
 
>
>Recommend just proceeding with the other work and waiting until V1.1 which 
>solves many of the multicast issues required and also solves a number of 
>errata from IB as a whole.  This IB version, IMO, what will be deployed as 
>volume product in 2003 so aligning with its functionality bodes best for 
>interoperability and minimizing development costs.

That is what I suggested a couple of weeks ago. I'll remove all the
discussion on promiscuous MC node from the link/multicast draft. We
need to make progress to move these drafts to last calls in oder to
continue other work.

Jerry

>
>Mike
>


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@ns.ietf.org  Mon Jun 10 14:09:23 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA16994
	for <ipoverib-archive@odin.ietf.org>; Mon, 10 Jun 2002 14:09:19 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id OAA20288
	for ipoverib-archive@odin.ietf.org; Mon, 10 Jun 2002 14:09:53 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id OAA19982;
	Mon, 10 Jun 2002 14:02:57 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id OAA19957
	for <ipoverib@ns.ietf.org>; Mon, 10 Jun 2002 14:02:55 -0400 (EDT)
Received: from e31.co.us.ibm.com (e31.co.us.ibm.com [32.97.110.129])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA16624
	for <ipoverib@ietf.org>; Mon, 10 Jun 2002 14:02:20 -0400 (EDT)
Received: from westrelay03.boulder.ibm.com (westrelay03.boulder.ibm.com [9.17.194.24])
	by e31.co.us.ibm.com (8.12.2/8.12.2) with ESMTP id g5AHx0cx029476;
	Mon, 10 Jun 2002 13:59:00 -0400
Received: from gateway.beaverton.ibm.com (gateway.beaverton.ibm.com [138.95.180.1])
	by westrelay03.boulder.ibm.com (8.11.1m3/NCO/VER6.1) with ESMTP id g5AHx0S72702;
	Mon, 10 Jun 2002 11:59:00 -0600
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway.beaverton.ibm.com (8.10.0.Beta10/8.11.6) with ESMTP id g5AHwve04526;
	Mon, 10 Jun 2002 10:58:57 -0700 (PDT)
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g5AHwtp03374;
	Mon, 10 Jun 2002 10:58:55 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206101758.g5AHwtp03374@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: Jerry.Chu@eng.sun.com
Date: Mon, 10 Jun 2002 10:58:53 -0700 (PDT)
Cc: ipoverib@ietf.org, krause@cup.hp.com
In-Reply-To: <200206101744.g5AHiWwU599855@jurassic.eng.sun.com> from "H.K. Jerry Chu" at Jun 10, 2002 09:44:26 AM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

> 
> >At 08:05 PM 5/31/2002 -0700, H.K. Jerry Chu wrote:
> >
> >>IBA 1.0 spec. does not provide a promiscuous multicast mode in
> >>hardware or allow one to be emulated in software. So far there
> >>have been three proposals to address this issue:
> >>
> >>1. Require the IPoIB driver to send a copy of every outgoing
> >>multicast pkt to the all router multicast group. The algorithm
> >>looks like this: for a pkt destined to a multicast group M
> >>
> >>         if (M doesn't exist)
> >>                 send pkt to all_router
> >>         else if (M != all_router)
> >>                 send pkt to all_router;
> >>                 send pkt to M;
> >>         else send pkt to M;
> >
> >Strongly oppose this as previously discussed many months back.
> 
> I hear you :-).
> 
> >
> >>
> >>2. Modify IP group management protocol module to do some extra
> >>work in case the link type is IB. An IP router will explicitly
> >>join any new multicast group that comes into existence. The
> >>router needs to figure out when to leave too (details from
> >>Vivek).
> >
> >This does not require any change to the IP router code - the router joins 
> >any multicast group through event notification if it and group are in the 
> >same partition.
> 
> 2 above is Vivek's proposal that requires changes to IP routing code.
> The event notification is 3 below.

        Jerry, have you seen my previous reply on what the impact of
        the solution is. It is quite simple.


Vivek

> 
> >
> >>3. Ask IBTA to add a facility to allow promiscuous multicast
> >>mode to be emulated in software.
> >
> >Don't see anything that must be added by the IBTA to allow this to work.
> 
> Great!
> 
> >
> >
> >>1. above seems to be the least painful. But it requires every
> >>node to be capable of sending pkts to all_router multicast
> >>group.
> >>
> >>The biggest problem with 2. IMO is that it requires changes to
> >>existing IP code in every version of group management code
> >>including IGMPv1, IGMPv2, MLD..., etc, and the change doesn't
> >>seem trivial. It may present a big obstacle for many OSes to
> >>request IB only changes in generic IP code.
> >>
> >> >From what I've heard 3 should be simple to add. The SA has all
> >>the MC info already. It simply needs to supply the caller a
> >>list of all the existing MC groups, and notify the caller of
> >>any new group coming into existence so that the caller (driver
> >>of an IP router) can ensure it "join to observe" every
> >>multicast group.
> >
> >I believe group event notification is supposed to be defined as part of 
> >V1.1.  There is no reason to have an observe capability.
> 
> Ok, I trust that you will ensure all the facilities will be there
> in V1.1 for an emulation of promiscuous MC mode to be used by IPoIB.
>  
> >
> >Recommend just proceeding with the other work and waiting until V1.1 which 
> >solves many of the multicast issues required and also solves a number of 
> >errata from IB as a whole.  This IB version, IMO, what will be deployed as 
> >volume product in 2003 so aligning with its functionality bodes best for 
> >interoperability and minimizing development costs.
> 
> That is what I suggested a couple of weeks ago. I'll remove all the
> discussion on promiscuous MC node from the link/multicast draft. We
> need to make progress to move these drafts to last calls in oder to
> continue other work.
> 
> Jerry
> 
> >
> >Mike
> >
> 
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 


-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Tue Jun 18 18:16:44 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA06627
	for <ipoverib-archive@odin.ietf.org>; Tue, 18 Jun 2002 18:16:43 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id SAA19580
	for ipoverib-archive@odin.ietf.org; Tue, 18 Jun 2002 18:17:23 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id SAA19379;
	Tue, 18 Jun 2002 18:15:54 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id SAA19347
	for <ipoverib@optimus.ietf.org>; Tue, 18 Jun 2002 18:15:52 -0400 (EDT)
Received: from mail-green.research.att.com (mail-green.research.att.com [135.207.30.103])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA06589
	for <ipoverib@ietf.org>; Tue, 18 Jun 2002 18:15:12 -0400 (EDT)
Received: from alliance.research.att.com (alliance.research.att.com [135.207.26.26])
	by mail-green.research.att.com (Postfix) with ESMTP
	id 763C11E06B; Tue, 18 Jun 2002 18:15:51 -0400 (EDT)
Received: from windsor.research.att.com (windsor.research.att.com [135.207.26.46])
	by alliance.research.att.com (8.8.7/8.8.7) with ESMTP id SAA14911;
	Tue, 18 Jun 2002 18:15:49 -0400 (EDT)
From: Bill Fenner <fenner@research.att.com>
Received: (from fenner@localhost)
	by windsor.research.att.com (8.8.8+Sun/8.8.5) id PAA19716;
	Tue, 18 Jun 2002 15:15:49 -0700 (PDT)
Message-Id: <200206182215.PAA19716@windsor.research.att.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
To: kashyapv@us.ibm.com
Subject: Re: [Ipoverib] remaining multicast issues - one more try
Cc: Jerry.Chu@eng.sun.com, ipoverib@ietf.org
Date: Tue, 18 Jun 2002 15:15:49 -0700
Versions: dmail (solaris) 2.4c/makemail 2.9d
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org


I removed a couple of round trips from the quoting here, in order
to simplify the exchange.  Apologies if I misunderstood something and
acidentally modified or misattributed someone's meaning.

Jerry's original #2 suggestion:

>2. Modify IP group management protocol module to do some extra
>work in case the link type is IB. An IP router will explicitly
>join any new multicast group that comes into existence. The
>router needs to figure out when to leave too (details from
>Vivek).


Vivek replies:

>        In existing implmementation the routing daemon on receiving an
>        IGMP report adds the IP multicast group to the interface list.
>        It tells the kernel to add the group by a socket option (in
>        BSD/LInux implementations). The function after it adds the IP
>        group to the interface can call the interface's device
>        specific function - for IPoIB this function will ask the
>        driver to join the IB group.

How does the router receive the IGMP report?  In IGMPv1 and IGMPv2, the
report is sent to the group being reported, with the assumption that the
router, being promiscuous to multicast, will receive it.  IGMPv3 changes
this to send to a protocol-specific "all-IGMPv3-routers" group, so doesn't
have this problem in an IB environment, but IGMPv3 is not widely deployed.

Existing router implementations of which I'm aware (mrouted/gated/pimd/etc)
don't specifically add a group to an interface in the kernel; instead they
wait for data to arrivate and add a forwarding cache entry for the given
source and group.  The router implementations would have to be modified
to inform the kernel of group memberships on a given interface.

Modifying host implementations to send IGMP to the all-routers group
could work, but that's back to modifying the IP stack for an IB-specific
case.

  Bill

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Tue Jun 18 19:11:44 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA08018
	for <ipoverib-archive@odin.ietf.org>; Tue, 18 Jun 2002 19:11:44 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id TAA23238
	for ipoverib-archive@odin.ietf.org; Tue, 18 Jun 2002 19:12:24 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA23185;
	Tue, 18 Jun 2002 19:11:47 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA23157
	for <ipoverib@optimus.ietf.org>; Tue, 18 Jun 2002 19:11:45 -0400 (EDT)
Received: from e2.ny.us.ibm.com (e2.ny.us.ibm.com [32.97.182.102])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA07978
	for <ipoverib@ietf.org>; Tue, 18 Jun 2002 19:11:04 -0400 (EDT)
Received: from northrelay03.pok.ibm.com (northrelay03.pok.ibm.com [9.56.224.151])
	by e2.ny.us.ibm.com (8.12.2/8.12.2) with ESMTP id g5INBBe8040550;
	Tue, 18 Jun 2002 19:11:11 -0400
Received: from gateway1.beaverton.ibm.com (gateway1.beaverton.ibm.com [138.95.180.2])
	by northrelay03.pok.ibm.com (8.11.1m3/NCO/VER6.1) with ESMTP id g5INB7a126362;
	Tue, 18 Jun 2002 19:11:07 -0400
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway1.beaverton.ibm.com (8.11.6/8.11.6) with ESMTP id g5INAJ931076;
	Tue, 18 Jun 2002 16:10:19 -0700
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g5INB7x10076;
	Tue, 18 Jun 2002 16:11:07 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206182311.g5INB7x10076@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: fenner@research.att.com (Bill Fenner)
Date: Tue, 18 Jun 2002 16:11:06 -0700 (PDT)
Cc: kashyapv@us.ibm.com, Jerry.Chu@eng.sun.com, ipoverib@ietf.org
In-Reply-To: <200206182215.PAA19716@windsor.research.att.com> from "Bill Fenner" at Jun 18, 2002 02:15:49 PM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

> 
> 
> I removed a couple of round trips from the quoting here, in order
> to simplify the exchange.  Apologies if I misunderstood something and
> acidentally modified or misattributed someone's meaning.
> 
> Jerry's original #2 suggestion:
> 
> >2. Modify IP group management protocol module to do some extra
> >work in case the link type is IB. An IP router will explicitly
> >join any new multicast group that comes into existence. The
> >router needs to figure out when to leave too (details from
> >Vivek).
> 
> 
> Vivek replies:
> 
> >        In existing implmementation the routing daemon on receiving an
> >        IGMP report adds the IP multicast group to the interface list.
> >        It tells the kernel to add the group by a socket option (in
> >        BSD/LInux implementations). The function after it adds the IP
> >        group to the interface can call the interface's device
> >        specific function - for IPoIB this function will ask the
> >        driver to join the IB group.
> 
> How does the router receive the IGMP report?  In IGMPv1 and IGMPv2, the
> report is sent to the group being reported, with the assumption that the
> router, being promiscuous to multicast, will receive it.  IGMPv3 changes
> this to send to a protocol-specific "all-IGMPv3-routers" group, so doesn't
> have this problem in an IB environment, but IGMPv3 is not widely deployed.

        For IGMPv1/v2 the following will have to be done in the driver
        in the thread where IP_ADD_MEMBERSHIP is called:

                        check for corresponding IB group
                        if IB group exists 
                                IB_join it
                                IB_send report to group
                        else
                                IB_create IB group
                                IB_send report to all-routers MGID or
                                could send it to all-hosts group

        It is easy to ensure that the report, is sent out the
        all-hosts MGID. After the first such send, subsequent periodic
        reports need only be sent on the group specific MGID.

        One can't avoid this step but one can localise it in the
        driver. 

> 
> Existing router implementations of which I'm aware (mrouted/gated/pimd/etc)
> don't specifically add a group to an interface in the kernel; instead they
> wait for data to arrivate and add a forwarding cache entry for the given
> source and group.  The router implementations would have to be modified
> to inform the kernel of group memberships on a given interface.

        My suggestion is based on my reading of BSD and Linux
        multicast routing howto's:

        For example, in the case of BSD - a setsockopt()
        DVMRP_ADD_LGRP is sent down by mrouted/gated in response to
        IGMP_HOST_MEMBERSHIP_REPORT. The daemon lets the kernel know
        of the group and the interface. In this execution thread, when
        the group entry is created in the kernel, the associated
        interface can be called to join the relevant IB group (or such
        a join can be scheduled). For, the interfaces that don't have
        such a need the call can be a no-op (one can of course code
        such changes in manyways). Implementations that don't support
        IPoIB are unaffected and non-IPoIB drivers are unaffected
        too. 

        This group_list is consulted by the ip forwarding code to
        decide if a datagram needs to be output on the network(s)
        connected to local interfac(es). Only if there is a group
        entry associated with the interface is the packet output.

        mrouted/gated also send down ADD_MRT command to update the
        multicast route entries (in BSD).

        In this model there is no change required in the routing
        daemons. 

> 
> Modifying host implementations to send IGMP to the all-routers group
> could work, but that's back to modifying the IP stack for an IB-specific
> case.

	No that would be done in the IPoIB interface driver.

Vivek

-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Tue Jun 18 19:19:02 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA08186
	for <ipoverib-archive@odin.ietf.org>; Tue, 18 Jun 2002 19:19:02 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id TAA23579
	for ipoverib-archive@odin.ietf.org; Tue, 18 Jun 2002 19:19:42 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA23529;
	Tue, 18 Jun 2002 19:19:13 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA23496
	for <ipoverib@optimus.ietf.org>; Tue, 18 Jun 2002 19:19:11 -0400 (EDT)
Received: from patan.sun.com (patan.Sun.COM [192.18.98.43])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA08175
	for <ipoverib@ietf.org>; Tue, 18 Jun 2002 19:18:30 -0400 (EDT)
Received: from jurassic.eng.sun.com ([129.146.17.55])
	by patan.sun.com (8.9.3+Sun/8.9.3) with ESMTP id RAA24239;
	Tue, 18 Jun 2002 17:21:16 -0600 (MDT)
Received: from taipei (taipei.Eng.Sun.COM [129.146.85.178])
	by jurassic.eng.sun.com (8.12.4+Sun/8.12.4) with SMTP id g5INJ8K4595079;
	Tue, 18 Jun 2002 16:19:08 -0700 (PDT)
Message-Id: <200206182319.g5INJ8K4595079@jurassic.eng.sun.com>
Date: Tue, 18 Jun 2002 16:19:01 -0700 (PDT)
From: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Reply-To: "H.K. Jerry Chu" <Jerry.Chu@eng.sun.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: fenner@research.att.com
Cc: ipoverib@ietf.org
MIME-Version: 1.0
Content-Type: TEXT/plain; charset=us-ascii
Content-MD5: fcICkr85fxrNeoh42/LVKA==
X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4.6_06 SunOS 5.8 sun4u sparc 
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

Thanks for the explanation, Bill. So if I understand correctly,
proposal # 2 won't work unless changes are made to the IP code on
both the host side and the router side (with the exception of IGMPv3,
which has been widely deployed), right?

I proposed a "hack" (#1) initially to require the host side IPoIB driver
to copy every outbound multicast pkt to all-router. There was a small
problem if a router itself is a genuine member of the targe multicast
group but this problem can be worked around easily (yes more hack :-(.
This workaround is not something one would brag about. But at least
it is transparent to IP implementations.

A different proposal #2 below was made later (by Vivek). My problem
with that propsal is that it requires link-specific changes to router
implementation. Now it looks like it requires changes to host side
as well.

I have also tried to get IBTA to add support for a software emulation of
promiscuous mulicast mode (#3). From what I heard IBTA was close to
getting one done. But there seemed to be some misunderstanding leading
them to believing IETF has a solution already (which I don't believe
is the case), and a solution is no longer needed from them.

My personal opinion (WG co-chair hat off) is that if adding promiscuous
mode support is doable and does not negatively impact IBA architecture,
that solution is much preferred than the above two. And one never knows
if other protocol or any future ones will require promiscuous multicast
mode (albeit it is arguably a bad design). If others on the list agree
we should let the voice be heard by IBTA members.

BTW, which one of the three above do you prefer, Bill?

Jerry

>I removed a couple of round trips from the quoting here, in order
>to simplify the exchange.  Apologies if I misunderstood something and
>acidentally modified or misattributed someone's meaning.
>
>Jerry's original #2 suggestion:
>
>>2. Modify IP group management protocol module to do some extra
>>work in case the link type is IB. An IP router will explicitly
>>join any new multicast group that comes into existence. The
>>router needs to figure out when to leave too (details from
>>Vivek).
>
>
>Vivek replies:
>
>>        In existing implmementation the routing daemon on receiving an
>>        IGMP report adds the IP multicast group to the interface list.
>>        It tells the kernel to add the group by a socket option (in
>>        BSD/LInux implementations). The function after it adds the IP
>>        group to the interface can call the interface's device
>>        specific function - for IPoIB this function will ask the
>>        driver to join the IB group.
>
>How does the router receive the IGMP report?  In IGMPv1 and IGMPv2, the
>report is sent to the group being reported, with the assumption that the
>router, being promiscuous to multicast, will receive it.  IGMPv3 changes
>this to send to a protocol-specific "all-IGMPv3-routers" group, so doesn't
>have this problem in an IB environment, but IGMPv3 is not widely deployed.
>
>Existing router implementations of which I'm aware (mrouted/gated/pimd/etc)
>don't specifically add a group to an interface in the kernel; instead they
>wait for data to arrivate and add a forwarding cache entry for the given
>source and group.  The router implementations would have to be modified
>to inform the kernel of group memberships on a given interface.
>
>Modifying host implementations to send IGMP to the all-routers group
>could work, but that's back to modifying the IP stack for an IB-specific
>case.
>
>  Bill


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Tue Jun 18 19:39:41 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA08519
	for <ipoverib-archive@odin.ietf.org>; Tue, 18 Jun 2002 19:39:40 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id TAA25206
	for ipoverib-archive@odin.ietf.org; Tue, 18 Jun 2002 19:40:21 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA25179;
	Tue, 18 Jun 2002 19:39:24 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA25113
	for <ipoverib@optimus.ietf.org>; Tue, 18 Jun 2002 19:39:21 -0400 (EDT)
Received: from mail-green.research.att.com (H-135-207-30-103.research.att.com [135.207.30.103])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA08503
	for <ipoverib@ietf.org>; Tue, 18 Jun 2002 19:38:39 -0400 (EDT)
Received: from alliance.research.att.com (alliance.research.att.com [135.207.26.26])
	by mail-green.research.att.com (Postfix) with ESMTP
	id E81E01E024; Tue, 18 Jun 2002 19:38:36 -0400 (EDT)
Received: from windsor.research.att.com (windsor.research.att.com [135.207.26.46])
	by alliance.research.att.com (8.8.7/8.8.7) with ESMTP id TAA15595;
	Tue, 18 Jun 2002 19:38:36 -0400 (EDT)
From: Bill Fenner <fenner@research.att.com>
Received: (from fenner@localhost)
	by windsor.research.att.com (8.8.8+Sun/8.8.5) id QAA20605;
	Tue, 18 Jun 2002 16:38:35 -0700 (PDT)
Message-Id: <200206182338.QAA20605@windsor.research.att.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
To: kashyapv@us.ibm.com
Subject: Re: [Ipoverib] remaining multicast issues - one more try
Cc: Jerry.Chu@eng.sun.com, ipoverib@ietf.org
Date: Tue, 18 Jun 2002 16:38:35 -0700
Versions: dmail (solaris) 2.4c/makemail 2.9d
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org


Well, it would be a little difficult for the host driver to do exactly
what you describe:

>                        check for corresponding IB group
>                        if IB group exists 
>                                IB_join it
>                                IB_send report to group
>                        else
>                                IB_create IB group
>                                IB_send report to all-routers MGID or
>                                could send it to all-hosts group

The report is sent from the stack, not from the driver, so the driver
can't easily affect where it's sent.  The driver could send a seperate
message, though.

Let's take FreeBSD as a concrete example.

	Application calls IP_ADD_MEMBERSHIP
	  IP calls in_addmulti()
	    in_addmulti() calls if_addmulti()
	      if_addmulti():
		- increments the reference count if it exists
		- calls the media-family-specific multicast mapping routine
		  (if_resolvemulti() ) in order to get a media address
		  for this multicast address.
		  [presumably, in an IB network this contacts the
		   multicast server and finds out if there is an IB
		   group formed yet, and requests its creation if not.
		   Note that there's no kernel infrastructure in place for
		   this to take time, the mapping is expected to be
		   algorithmic.  Perhaps this routine could initiate
		   the exchange and return a placeholder, and packets
		   could be queued at if_output() until the protocol
		   exchange completes.]
		- calls the driver's IOCTL with SIOCADDMULTI to notify
                  that the list of multicast addresses changed.
            in_addmulti() calls igmp_joingroup() which may send one or more
              IGMP messages.

The only driver-specific portions are the last two pieces of if_addmulti() -
in particular, the sending of IGMP messages to a different group would
have to be performed by intercepting them in the normal if_output() routine,
e.g. examining every packet that is sent for a time after joining the
new group to see if it's the IGMP report and redirecting it.  That seems
a little hokey.

>        For example, in the case of BSD - a setsockopt()
>        DVMRP_ADD_LGRP is sent down by mrouted/gated in response to
>        IGMP_HOST_MEMBERSHIP_REPORT.

In my experience, this was last true in 1995 or 1996 when the kernel
was in charge of multicast routing.  The more recent architecture is
as I described earlier, i.e. the kernel just has a forwarding cache and
does not learn of group memberships on interfaces.

If the driver sends a special message when it creates a new IB group,
a system (like a multicast router) that is in multicast promiscuous mode
can join each of these groups.  Garbage-collecting might be difficult,
once you have only routers as members; a group might still be transiting
the IB subnet.

  Bill

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Tue Jun 18 19:47:22 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA08732
	for <ipoverib-archive@odin.ietf.org>; Tue, 18 Jun 2002 19:47:22 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id TAA25565
	for ipoverib-archive@odin.ietf.org; Tue, 18 Jun 2002 19:48:01 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA25540;
	Tue, 18 Jun 2002 19:46:21 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA25485
	for <ipoverib@optimus.ietf.org>; Tue, 18 Jun 2002 19:46:19 -0400 (EDT)
Received: from e1.ny.us.ibm.com (e1.ny.us.ibm.com [32.97.182.101])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA08648
	for <ipoverib@ietf.org>; Tue, 18 Jun 2002 19:45:38 -0400 (EDT)
Received: from westrelay03.boulder.ibm.com (westrelay03.boulder.ibm.com [9.17.194.24])
	by e1.ny.us.ibm.com (8.12.2/8.12.2) with ESMTP id g5INjXg5229190;
	Tue, 18 Jun 2002 19:45:33 -0400
Received: from gateway1.beaverton.ibm.com (gateway1.beaverton.ibm.com [138.95.180.2])
	by westrelay03.boulder.ibm.com (8.11.1m3/NCO/VER6.2) with ESMTP id g5INjWW15450;
	Tue, 18 Jun 2002 17:45:32 -0600
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway1.beaverton.ibm.com (8.11.6/8.11.6) with ESMTP id g5INig931624;
	Tue, 18 Jun 2002 16:44:42 -0700
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g5INjUP14669;
	Tue, 18 Jun 2002 16:45:30 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206182345.g5INjUP14669@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: Jerry.Chu@eng.sun.com
Date: Tue, 18 Jun 2002 16:45:29 -0700 (PDT)
Cc: fenner@research.att.com, ipoverib@ietf.org
In-Reply-To: <200206182319.g5INJ8K4595079@jurassic.eng.sun.com> from "H.K. Jerry Chu" at Jun 18, 2002 03:19:01 PM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

> 
> Thanks for the explanation, Bill. So if I understand correctly,
> proposal # 2 won't work unless changes are made to the IP code on
> both the host side and the router side (with the exception of IGMPv3,
> which has been widely deployed), right?

        Read my previous mail - I have put forth this example earlier
        too. Let us discuss this in detail before knocking it off.

> 
> I proposed a "hack" (#1) initially to require the host side IPoIB driver
> to copy every outbound multicast pkt to all-router. There was a small
> problem if a router itself is a genuine member of the targe multicast
> group but this problem can be worked around easily (yes more hack :-(.
> This workaround is not something one would brag about. But at least
> it is transparent to IP implementations.

        As noted earlier in the thread - this has the possibility of
        inserting duplicates in the network. To implement the 'hack'
        wouldn't the router have to change too and the host too ?

	As was noted on June 7th in this thread:

            "How do one ensure that the router is not a member of any
            IB group ? What mechanism stops it from being a member ?
            Many nodes do double duty - they host applications and
            also act as routers - how do they decide?

            As per this solution a router will receive two packets and
            if the IP subnet has multiple routers then multiple
            routers get two packets. If the corresponding IP group has
            a scope larger than the local subnet - each of these
            routers will now forward multiple packets outward. If one
            or more of these neighbouring subnets are IPoIB subnets we
            can further multiply the number of copies by a factor of
            two at the least. The result would be that IPoIB
            subnetswould end up creating multicast storms quite
            often.  "

		

> 
> A different proposal #2 below was made later (by Vivek). My problem
> with that propsal is that it requires link-specific changes to router
> implementation. Now it looks like it requires changes to host side
> as well.

        Not later :) - I proposed it in my earliest drafts too since
        the IBTA spec was very limited in this regard.


        See my previous mail. I had written in this thread on June 7th
        regarding this proposal :

                "
		Take a look at the description below. The change is a
                simple function in the driver with a line or two added
                (considering BSD code) to the 'add multicast group'
                socket option.

                The suggestion works within the existing
                infrastructure with a minor addition.

                In existing implmementation the routing daemon on
                receiving an IGMP report adds the IP multicast group
                to the interface list. It tells the kernel to add the
                group by a socket option (in BSD/LInux
                implementations). The function after it adds the IP
                group to the interface can call the interface's device
                specific function - for IPoIB this function will ask
                the driver to join the IB group.

		(IGMP_report)--- n/w -> driver -> IGMP -> mrouted/gated
                                                        |
                                                    setsockopt(ADD_LGRP)
                                                        |
                                                        :
                                                        :
                                                        add_grp()
                                                        |
                                                ** call device_add_grp() **
                                                ** join relevant IB group **
                                                ** if needed            **

                Only the call marked with '**' is new. Nothing else.
                Note that the same function or similar function will
                have to exist on hosts for IP_ADD_MEMBERSHIP directive
                when a host application joins a particular IP group.
                In this case the router does the same thing on
                receiving an IGMP_HOST_MEMBERSHIP_REPORT.

                The change is simple and does not require changes to
                'every version of group mgmt code'. It does not
                require changes to routing daemons, nor does it imply
                large, complicated changes in OSes. "



For the rest, I am suggesting solutions based on the current state of the draft including the two enhancements regarding group create/delete traps you have mentioned. Promiscuous mode is not included in the IBTA specification.

Vivek

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Wed Jun 19 12:50:11 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA26534
	for <ipoverib-archive@odin.ietf.org>; Wed, 19 Jun 2002 12:50:11 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id MAA11535
	for ipoverib-archive@odin.ietf.org; Wed, 19 Jun 2002 12:50:50 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id MAA11486;
	Wed, 19 Jun 2002 12:49:50 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id MAA11446
	for <ipoverib@optimus.ietf.org>; Wed, 19 Jun 2002 12:49:48 -0400 (EDT)
Received: from pan.ch.intel.com (chfdns01.ch.intel.com [143.182.246.24])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA26510
	for <ipoverib@ietf.org>; Wed, 19 Jun 2002 12:49:07 -0400 (EDT)
Received: from azsmsxvs043.ch.intel.com (azsmsxvs043.ch.intel.com [10.2.248.13])
	by pan.ch.intel.com (8.11.6/8.11.6/d: solo.mc,v 1.42 2002/05/23 22:21:11 root Exp $) with SMTP id g5JGnkZ15781
	for <ipoverib@ietf.org>; Wed, 19 Jun 2002 16:49:46 GMT
Received: from fmsmsx28.fm.intel.com ([132.233.42.28])
 by azsmsxvs043.ch.intel.com (NAVGW 2.5.2.11) with SMTP id M2002061909494414882
 ; Wed, 19 Jun 2002 09:49:44 -0700
Received: by fmsmsx28.fm.intel.com with Internet Mail Service (5.5.2653.19)
	id <M8C6LPA4>; Wed, 19 Jun 2002 09:49:43 -0700
Message-ID: <D9223EB959A5D511A98F00508B68C20C06BC5051@orsmsx108.jf.intel.com>
From: "Larsen, Roy K" <roy.k.larsen@intel.com>
To: "'Vivek Kashyap'" <kashyapv@us.ibm.com>, Jerry.Chu@eng.sun.com
Cc: fenner@research.att.com, ipoverib@ietf.org
Subject: RE: [Ipoverib] remaining multicast issues - one more try
Date: Wed, 19 Jun 2002 09:49:34 -0700
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.2653.19)
Content-Type: text/plain;
	charset="ISO-8859-1"
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

Vivek,

Is the last paragraph of your mail the answer to Bill's question on how the
router will even get the IGMP report to begin with?  I.e. register for group
create/delete traps and adding the filter when they are received.  Is so,
doing something base on a received IGMP report is redundant.  If not, how
does the router receive the report sent to a multicast group it is not a
member of in IGMPv1 and IGMPv2 environments?

Roy

-----Original Message-----
From: Vivek Kashyap [mailto:kashyapv@us.ibm.com]
Sent: Tuesday, June 18, 2002 4:45 PM
To: Jerry.Chu@eng.sun.com
Cc: fenner@research.att.com; ipoverib@ietf.org
Subject: Re: [Ipoverib] remaining multicast issues - one more try


> 
> Thanks for the explanation, Bill. So if I understand correctly,
> proposal # 2 won't work unless changes are made to the IP code on
> both the host side and the router side (with the exception of IGMPv3,
> which has been widely deployed), right?

        Read my previous mail - I have put forth this example earlier
        too. Let us discuss this in detail before knocking it off.

> 
> I proposed a "hack" (#1) initially to require the host side IPoIB driver
> to copy every outbound multicast pkt to all-router. There was a small
> problem if a router itself is a genuine member of the targe multicast
> group but this problem can be worked around easily (yes more hack :-(.
> This workaround is not something one would brag about. But at least
> it is transparent to IP implementations.

        As noted earlier in the thread - this has the possibility of
        inserting duplicates in the network. To implement the 'hack'
        wouldn't the router have to change too and the host too ?

	As was noted on June 7th in this thread:

            "How do one ensure that the router is not a member of any
            IB group ? What mechanism stops it from being a member ?
            Many nodes do double duty - they host applications and
            also act as routers - how do they decide?

            As per this solution a router will receive two packets and
            if the IP subnet has multiple routers then multiple
            routers get two packets. If the corresponding IP group has
            a scope larger than the local subnet - each of these
            routers will now forward multiple packets outward. If one
            or more of these neighbouring subnets are IPoIB subnets we
            can further multiply the number of copies by a factor of
            two at the least. The result would be that IPoIB
            subnetswould end up creating multicast storms quite
            often.  "

		

> 
> A different proposal #2 below was made later (by Vivek). My problem
> with that propsal is that it requires link-specific changes to router
> implementation. Now it looks like it requires changes to host side
> as well.

        Not later :) - I proposed it in my earliest drafts too since
        the IBTA spec was very limited in this regard.


        See my previous mail. I had written in this thread on June 7th
        regarding this proposal :

                "
		Take a look at the description below. The change is a
                simple function in the driver with a line or two added
                (considering BSD code) to the 'add multicast group'
                socket option.

                The suggestion works within the existing
                infrastructure with a minor addition.

                In existing implmementation the routing daemon on
                receiving an IGMP report adds the IP multicast group
                to the interface list. It tells the kernel to add the
                group by a socket option (in BSD/LInux
                implementations). The function after it adds the IP
                group to the interface can call the interface's device
                specific function - for IPoIB this function will ask
                the driver to join the IB group.

		(IGMP_report)--- n/w -> driver -> IGMP -> mrouted/gated
                                                        |
                                                    setsockopt(ADD_LGRP)
                                                        |
                                                        :
                                                        :
                                                        add_grp()
                                                        |
                                                ** call device_add_grp() **
                                                ** join relevant IB group **
                                                ** if needed            **

                Only the call marked with '**' is new. Nothing else.
                Note that the same function or similar function will
                have to exist on hosts for IP_ADD_MEMBERSHIP directive
                when a host application joins a particular IP group.
                In this case the router does the same thing on
                receiving an IGMP_HOST_MEMBERSHIP_REPORT.

                The change is simple and does not require changes to
                'every version of group mgmt code'. It does not
                require changes to routing daemons, nor does it imply
                large, complicated changes in OSes. "



For the rest, I am suggesting solutions based on the current state of the
draft including the two enhancements regarding group create/delete traps you
have mentioned. Promiscuous mode is not included in the IBTA specification.

Vivek

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Wed Jun 19 13:16:30 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA27638
	for <ipoverib-archive@odin.ietf.org>; Wed, 19 Jun 2002 13:16:30 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id NAA14071
	for ipoverib-archive@odin.ietf.org; Wed, 19 Jun 2002 13:17:10 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA13990;
	Wed, 19 Jun 2002 13:16:26 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA13955
	for <ipoverib@optimus.ietf.org>; Wed, 19 Jun 2002 13:16:22 -0400 (EDT)
Received: from e2.ny.us.ibm.com (e2.ny.us.ibm.com [32.97.182.102])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA27600
	for <ipoverib@ietf.org>; Wed, 19 Jun 2002 13:15:41 -0400 (EDT)
Received: from northrelay03.pok.ibm.com (northrelay03.pok.ibm.com [9.56.224.151])
	by e2.ny.us.ibm.com (8.12.2/8.12.2) with ESMTP id g5JHFYe8127088;
	Wed, 19 Jun 2002 13:15:37 -0400
Received: from gateway.beaverton.ibm.com (gateway.beaverton.ibm.com [138.95.180.1])
	by northrelay03.pok.ibm.com (8.11.1m3/NCO/VER6.1) with ESMTP id g5JHFUZ101774;
	Wed, 19 Jun 2002 13:15:31 -0400
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway.beaverton.ibm.com (8.10.0.Beta10/8.11.6) with ESMTP id g5JHFRe12341;
	Wed, 19 Jun 2002 10:15:27 -0700 (PDT)
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g5JHFOk07494;
	Wed, 19 Jun 2002 10:15:24 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206191715.g5JHFOk07494@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: roy.k.larsen@intel.com (Larsen, Roy K)
Date: Wed, 19 Jun 2002 10:15:23 -0700 (PDT)
Cc: kashyapv@us.ibm.com ('Vivek Kashyap'), Jerry.Chu@eng.sun.com,
        fenner@research.att.com, ipoverib@ietf.org
In-Reply-To: <D9223EB959A5D511A98F00508B68C20C06BC5051@orsmsx108.jf.intel.com> from "Larsen, Roy K" at Jun 19, 2002 08:49:34 AM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

> 
> Vivek,
> 
> Is the last paragraph of your mail the answer to Bill's question on how the
> router will even get the IGMP report to begin with?  I.e. register for group
> create/delete traps and adding the filter when they are received.  Is so,
> doing something base on a received IGMP report is redundant.  If not, how
> does the router receive the report sent to a multicast group it is not a
> member of in IGMPv1 and IGMPv2 environments?

No, the response to his question was in another mail. See the latest
mails from me and Bill for that discussion. 


Vivek

> 
> Roy
> 
> -----Original Message-----
> From: Vivek Kashyap [mailto:kashyapv@us.ibm.com]
> Sent: Tuesday, June 18, 2002 4:45 PM
> To: Jerry.Chu@eng.sun.com
> Cc: fenner@research.att.com; ipoverib@ietf.org
> Subject: Re: [Ipoverib] remaining multicast issues - one more try
> 
> 
> > 
> > Thanks for the explanation, Bill. So if I understand correctly,
> > proposal # 2 won't work unless changes are made to the IP code on
> > both the host side and the router side (with the exception of IGMPv3,
> > which has been widely deployed), right?
> 
>         Read my previous mail - I have put forth this example earlier
>         too. Let us discuss this in detail before knocking it off.
> 
> > 
> > I proposed a "hack" (#1) initially to require the host side IPoIB driver
> > to copy every outbound multicast pkt to all-router. There was a small
> > problem if a router itself is a genuine member of the targe multicast
> > group but this problem can be worked around easily (yes more hack :-(.
> > This workaround is not something one would brag about. But at least
> > it is transparent to IP implementations.
> 
>         As noted earlier in the thread - this has the possibility of
>         inserting duplicates in the network. To implement the 'hack'
>         wouldn't the router have to change too and the host too ?
> 
>   As was noted on June 7th in this thread:
> 
>             "How do one ensure that the router is not a member of any
>             IB group ? What mechanism stops it from being a member ?
>             Many nodes do double duty - they host applications and
>             also act as routers - how do they decide?
> 
>             As per this solution a router will receive two packets and
>             if the IP subnet has multiple routers then multiple
>             routers get two packets. If the corresponding IP group has
>             a scope larger than the local subnet - each of these
>             routers will now forward multiple packets outward. If one
>             or more of these neighbouring subnets are IPoIB subnets we
>             can further multiply the number of copies by a factor of
>             two at the least. The result would be that IPoIB
>             subnetswould end up creating multicast storms quite
>             often.  "
> 
>     
> 
> > 
> > A different proposal #2 below was made later (by Vivek). My problem
> > with that propsal is that it requires link-specific changes to router
> > implementation. Now it looks like it requires changes to host side
> > as well.
> 
>         Not later :) - I proposed it in my earliest drafts too since
>         the IBTA spec was very limited in this regard.
> 
> 
>         See my previous mail. I had written in this thread on June 7th
>         regarding this proposal :
> 
>                 "
>     Take a look at the description below. The change is a
>                 simple function in the driver with a line or two added
>                 (considering BSD code) to the 'add multicast group'
>                 socket option.
> 
>                 The suggestion works within the existing
>                 infrastructure with a minor addition.
> 
>                 In existing implmementation the routing daemon on
>                 receiving an IGMP report adds the IP multicast group
>                 to the interface list. It tells the kernel to add the
>                 group by a socket option (in BSD/LInux
>                 implementations). The function after it adds the IP
>                 group to the interface can call the interface's device
>                 specific function - for IPoIB this function will ask
>                 the driver to join the IB group.
> 
>     (IGMP_report)--- n/w -> driver -> IGMP -> mrouted/gated
>                                                         |
>                                                     setsockopt(ADD_LGRP)
>                                                         |
>                                                         :
>                                                         :
>                                                         add_grp()
>                                                         |
>                                                 ** call device_add_grp() **
>                                                 ** join relevant IB group **
>                                                 ** if needed            **
> 
>                 Only the call marked with '**' is new. Nothing else.
>                 Note that the same function or similar function will
>                 have to exist on hosts for IP_ADD_MEMBERSHIP directive
>                 when a host application joins a particular IP group.
>                 In this case the router does the same thing on
>                 receiving an IGMP_HOST_MEMBERSHIP_REPORT.
> 
>                 The change is simple and does not require changes to
>                 'every version of group mgmt code'. It does not
>                 require changes to routing daemons, nor does it imply
>                 large, complicated changes in OSes. "
> 
> 
> 
> For the rest, I am suggesting solutions based on the current state of the
> draft including the two enhancements regarding group create/delete traps you
> have mentioned. Promiscuous mode is not included in the IBTA specification.
> 
> Vivek
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 


-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Wed Jun 19 13:38:19 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA28469
	for <ipoverib-archive@odin.ietf.org>; Wed, 19 Jun 2002 13:38:18 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id NAA16369
	for ipoverib-archive@odin.ietf.org; Wed, 19 Jun 2002 13:38:58 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA16294;
	Wed, 19 Jun 2002 13:38:11 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id NAA16265
	for <ipoverib@optimus.ietf.org>; Wed, 19 Jun 2002 13:38:09 -0400 (EDT)
Received: from mail1.hd.intel.com (hdfdns01.hd.intel.com [192.52.58.10])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA28454
	for <ipoverib@ietf.org>; Wed, 19 Jun 2002 13:37:28 -0400 (EDT)
Received: from fmsmsxvs040.fm.intel.com (fmsmsxvs040.fm.intel.com [132.233.42.124])
	by mail1.hd.intel.com (8.11.6/8.11.6/d: solo.mc,v 1.42 2002/05/23 22:21:11 root Exp $) with SMTP id g5JHbZp13657
	for <ipoverib@ietf.org>; Wed, 19 Jun 2002 17:37:35 GMT
Received: from fmsmsx26.fm.intel.com ([132.233.42.26])
 by fmsmsxvs040.fm.intel.com (NAVGW 2.5.2.11) with SMTP id M2002061910365210440
 ; Wed, 19 Jun 2002 10:36:52 -0700
Received: by fmsmsx26.fm.intel.com with Internet Mail Service (5.5.2653.19)
	id <N1HSL19P>; Wed, 19 Jun 2002 10:37:33 -0700
Message-ID: <D9223EB959A5D511A98F00508B68C20C06BC5053@orsmsx108.jf.intel.com>
From: "Larsen, Roy K" <roy.k.larsen@intel.com>
To: "'Vivek Kashyap'" <kashyapv@us.ibm.com>
Cc: Jerry.Chu@eng.sun.com, fenner@research.att.com, ipoverib@ietf.org
Subject: RE: [Ipoverib] remaining multicast issues - one more try
Date: Wed, 19 Jun 2002 10:37:26 -0700
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.2653.19)
Content-Type: text/plain;
	charset="iso-8859-1"
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

I found it in the achieves.  Looks like it hasn't made it through our email
system yet.  Bill's response (which I did receive) now has some context and
I think he's covering the same concerns I have.

Roy 

-----Original Message-----
From: Vivek Kashyap [mailto:kashyapv@us.ibm.com]
Sent: Wednesday, June 19, 2002 10:15 AM
To: roy.k.larsen@intel.com
Cc: kashyapv@us.ibm.com; Jerry.Chu@eng.sun.com; fenner@research.att.com;
ipoverib@ietf.org
Subject: Re: [Ipoverib] remaining multicast issues - one more try


> 
> Vivek,
> 
> Is the last paragraph of your mail the answer to Bill's question on how
the
> router will even get the IGMP report to begin with?  I.e. register for
group
> create/delete traps and adding the filter when they are received.  Is so,
> doing something base on a received IGMP report is redundant.  If not, how
> does the router receive the report sent to a multicast group it is not a
> member of in IGMPv1 and IGMPv2 environments?

No, the response to his question was in another mail. See the latest
mails from me and Bill for that discussion. 


Vivek

> 
> Roy
> 
> -----Original Message-----
> From: Vivek Kashyap [mailto:kashyapv@us.ibm.com]
> Sent: Tuesday, June 18, 2002 4:45 PM
> To: Jerry.Chu@eng.sun.com
> Cc: fenner@research.att.com; ipoverib@ietf.org
> Subject: Re: [Ipoverib] remaining multicast issues - one more try
> 
> 
> > 
> > Thanks for the explanation, Bill. So if I understand correctly,
> > proposal # 2 won't work unless changes are made to the IP code on
> > both the host side and the router side (with the exception of IGMPv3,
> > which has been widely deployed), right?
> 
>         Read my previous mail - I have put forth this example earlier
>         too. Let us discuss this in detail before knocking it off.
> 
> > 
> > I proposed a "hack" (#1) initially to require the host side IPoIB driver
> > to copy every outbound multicast pkt to all-router. There was a small
> > problem if a router itself is a genuine member of the targe multicast
> > group but this problem can be worked around easily (yes more hack :-(.
> > This workaround is not something one would brag about. But at least
> > it is transparent to IP implementations.
> 
>         As noted earlier in the thread - this has the possibility of
>         inserting duplicates in the network. To implement the 'hack'
>         wouldn't the router have to change too and the host too ?
> 
>   As was noted on June 7th in this thread:
> 
>             "How do one ensure that the router is not a member of any
>             IB group ? What mechanism stops it from being a member ?
>             Many nodes do double duty - they host applications and
>             also act as routers - how do they decide?
> 
>             As per this solution a router will receive two packets and
>             if the IP subnet has multiple routers then multiple
>             routers get two packets. If the corresponding IP group has
>             a scope larger than the local subnet - each of these
>             routers will now forward multiple packets outward. If one
>             or more of these neighbouring subnets are IPoIB subnets we
>             can further multiply the number of copies by a factor of
>             two at the least. The result would be that IPoIB
>             subnetswould end up creating multicast storms quite
>             often.  "
> 
>     
> 
> > 
> > A different proposal #2 below was made later (by Vivek). My problem
> > with that propsal is that it requires link-specific changes to router
> > implementation. Now it looks like it requires changes to host side
> > as well.
> 
>         Not later :) - I proposed it in my earliest drafts too since
>         the IBTA spec was very limited in this regard.
> 
> 
>         See my previous mail. I had written in this thread on June 7th
>         regarding this proposal :
> 
>                 "
>     Take a look at the description below. The change is a
>                 simple function in the driver with a line or two added
>                 (considering BSD code) to the 'add multicast group'
>                 socket option.
> 
>                 The suggestion works within the existing
>                 infrastructure with a minor addition.
> 
>                 In existing implmementation the routing daemon on
>                 receiving an IGMP report adds the IP multicast group
>                 to the interface list. It tells the kernel to add the
>                 group by a socket option (in BSD/LInux
>                 implementations). The function after it adds the IP
>                 group to the interface can call the interface's device
>                 specific function - for IPoIB this function will ask
>                 the driver to join the IB group.
> 
>     (IGMP_report)--- n/w -> driver -> IGMP -> mrouted/gated
>                                                         |
>                                                     setsockopt(ADD_LGRP)
>                                                         |
>                                                         :
>                                                         :
>                                                         add_grp()
>                                                         |
>                                                 ** call device_add_grp()
**
>                                                 ** join relevant IB group
**
>                                                 ** if needed            **
> 
>                 Only the call marked with '**' is new. Nothing else.
>                 Note that the same function or similar function will
>                 have to exist on hosts for IP_ADD_MEMBERSHIP directive
>                 when a host application joins a particular IP group.
>                 In this case the router does the same thing on
>                 receiving an IGMP_HOST_MEMBERSHIP_REPORT.
> 
>                 The change is simple and does not require changes to
>                 'every version of group mgmt code'. It does not
>                 require changes to routing daemons, nor does it imply
>                 large, complicated changes in OSes. "
> 
> 
> 
> For the rest, I am suggesting solutions based on the current state of the
> draft including the two enhancements regarding group create/delete traps
you
> have mentioned. Promiscuous mode is not included in the IBTA
specification.
> 
> Vivek
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 


-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Thu Jun 20 14:17:22 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA09301
	for <ipoverib-archive@odin.ietf.org>; Thu, 20 Jun 2002 14:17:22 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id OAA10722
	for ipoverib-archive@odin.ietf.org; Thu, 20 Jun 2002 14:18:03 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id OAA10686;
	Thu, 20 Jun 2002 14:17:04 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id OAA10634
	for <ipoverib@optimus.ietf.org>; Thu, 20 Jun 2002 14:17:01 -0400 (EDT)
Received: from e21.nc.us.ibm.com (e21.nc.us.ibm.com [32.97.136.227])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA09283
	for <ipoverib@ietf.org>; Thu, 20 Jun 2002 14:16:19 -0400 (EDT)
Received: from westrelay03.boulder.ibm.com (westrelay03.boulder.ibm.com [9.17.194.24])
	by e21.nc.us.ibm.com (8.12.2/8.12.2) with ESMTP id g5KIGjRY109666;
	Thu, 20 Jun 2002 14:16:46 -0400
Received: from gateway.beaverton.ibm.com (gateway.beaverton.ibm.com [138.95.180.1])
	by westrelay03.boulder.ibm.com (8.11.1m3/NCO/VER6.2) with ESMTP id g5KIGe068390;
	Thu, 20 Jun 2002 12:16:40 -0600
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway.beaverton.ibm.com (8.10.0.Beta10/8.11.6) with ESMTP id g5KIGae13752;
	Thu, 20 Jun 2002 11:16:36 -0700 (PDT)
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g5KIGTa03077;
	Thu, 20 Jun 2002 11:16:30 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206201816.g5KIGTa03077@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: fenner@research.att.com (Bill Fenner)
Date: Thu, 20 Jun 2002 11:16:28 -0700 (PDT)
Cc: kashyapv@us.ibm.com, Jerry.Chu@eng.sun.com, ipoverib@ietf.org
In-Reply-To: <200206182338.QAA20605@windsor.research.att.com> from "Bill Fenner" at Jun 18, 2002 03:38:35 PM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

Bill,

        These proposals were made based on IB version 1.0 which
        provides no support of any kind for IB group create/delete
        notifications. So a way has to be found to let the router know
        of the IB groups being used or cause the routers to receive
        the packets in some way. One tack is to send notifications in
        a link level (driver based) protocol, the other was to use
        existing triggers (IGMP messages) and the third was to suggest
        the use of duplicate multicast packets (a copy to routers
        always in addition to the packet to the group).

        Given the lack of underlying support some ungainliness comes
        into the suggested solutions - so the aim has to be to reduce
        such ungainliness (in the protocol and individual
        implementations) but it can't be removed except by adding the
        necessary infrastructure to IBTA.

        With that in mind the IBTA is working on upgrading IB
        multicast support - if correctly done that should remove the
        need for any of the above solutions or leave only minor gaps
        to be filled in by IPoIB WG.

        The next version of IB specification is slated to come out in
        september. Jerry and I have removed the discussion of
        multicast forwarding and group managment from the
        link-and-multicast-02.txt draft (was section 10.0 in the
        -01.txt version). This issue can be taken up, if needed, after
        the next version of IB specification comes out.

        Driver based notifications (a separate low level protocol)
        would be another way to go if the IB spec doesn't deliver the
        necessary changes. Such a solution doesn't have the handicap
        of requiring IP stack changes since it can be localised to the
        drivers. But I believe the upgrades planned in IBTA will do
        the trick.

Vivek



-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Thu Jun 20 14:27:23 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA09488
	for <ipoverib-archive@odin.ietf.org>; Thu, 20 Jun 2002 14:27:23 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id OAA11202
	for ipoverib-archive@odin.ietf.org; Thu, 20 Jun 2002 14:28:04 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id OAA11032;
	Thu, 20 Jun 2002 14:24:34 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id OAA10999
	for <ipoverib@optimus.ietf.org>; Thu, 20 Jun 2002 14:24:32 -0400 (EDT)
Received: from e21.nc.us.ibm.com (e21.nc.us.ibm.com [32.97.136.227])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA09422
	for <ipoverib@ietf.org>; Thu, 20 Jun 2002 14:23:50 -0400 (EDT)
Received: from westrelay03.boulder.ibm.com (westrelay03.boulder.ibm.com [9.17.194.24])
	by e21.nc.us.ibm.com (8.12.2/8.12.2) with ESMTP id g5KIOURY132768
	for <ipoverib@ietf.org>; Thu, 20 Jun 2002 14:24:31 -0400
Received: from gateway.beaverton.ibm.com (gateway.beaverton.ibm.com [138.95.180.1])
	by westrelay03.boulder.ibm.com (8.11.1m3/NCO/VER6.2) with ESMTP id g5KIOU096286
	for <ipoverib@ietf.org>; Thu, 20 Jun 2002 12:24:30 -0600
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway.beaverton.ibm.com (8.10.0.Beta10/8.11.6) with ESMTP id g5KIOQe13962
	for <ipoverib@ietf.org>; Thu, 20 Jun 2002 11:24:26 -0700 (PDT)
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g5KIGTa03077;
	Thu, 20 Jun 2002 11:16:30 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206201816.g5KIGTa03077@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] remaining multicast issues - one more try
To: fenner@research.att.com (Bill Fenner)
Date: Thu, 20 Jun 2002 11:16:28 -0700 (PDT)
Cc: kashyapv@us.ibm.com, Jerry.Chu@eng.sun.com, ipoverib@ietf.org
In-Reply-To: <200206182338.QAA20605@windsor.research.att.com> from "Bill Fenner" at Jun 18, 2002 03:38:35 PM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

Bill,

        These proposals were made based on IB version 1.0 which
        provides no support of any kind for IB group create/delete
        notifications. So a way has to be found to let the router know
        of the IB groups being used or cause the routers to receive
        the packets in some way. One tack is to send notifications in
        a link level (driver based) protocol, the other was to use
        existing triggers (IGMP messages) and the third was to suggest
        the use of duplicate multicast packets (a copy to routers
        always in addition to the packet to the group).

        Given the lack of underlying support some ungainliness comes
        into the suggested solutions - so the aim has to be to reduce
        such ungainliness (in the protocol and individual
        implementations) but it can't be removed except by adding the
        necessary infrastructure to IBTA.

        With that in mind the IBTA is working on upgrading IB
        multicast support - if correctly done that should remove the
        need for any of the above solutions or leave only minor gaps
        to be filled in by IPoIB WG.

        The next version of IB specification is slated to come out in
        september. Jerry and I have removed the discussion of
        multicast forwarding and group managment from the
        link-and-multicast-02.txt draft (was section 10.0 in the
        -01.txt version). This issue can be taken up, if needed, after
        the next version of IB specification comes out.

        Driver based notifications (a separate low level protocol)
        would be another way to go if the IB spec doesn't deliver the
        necessary changes. Such a solution doesn't have the handicap
        of requiring IP stack changes since it can be localised to the
        drivers. But I believe the upgrades planned in IBTA will do
        the trick.

Vivek



-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Fri Jun 21 07:37:02 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id HAA08101
	for <ipoverib-archive@odin.ietf.org>; Fri, 21 Jun 2002 07:37:02 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id HAA06363
	for ipoverib-archive@odin.ietf.org; Fri, 21 Jun 2002 07:37:44 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id HAA06263;
	Fri, 21 Jun 2002 07:36:57 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id HAA06233
	for <ipoverib@optimus.ietf.org>; Fri, 21 Jun 2002 07:36:55 -0400 (EDT)
Received: from CNRI.Reston.VA.US (localhost [127.0.0.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id HAA07879;
	Fri, 21 Jun 2002 07:36:08 -0400 (EDT)
Message-Id: <200206211136.HAA07879@ietf.org>
Mime-Version: 1.0
Content-Type: Multipart/Mixed; Boundary="NextPart"
To: IETF-Announce: ;
Cc: ipoverib@ietf.org
From: Internet-Drafts@ietf.org
Reply-to: Internet-Drafts@ietf.org
Date: Fri, 21 Jun 2002 07:36:07 -0400
Subject: [Ipoverib] I-D ACTION:draft-ietf-ipoib-link-multicast-02.txt
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org

--NextPart

A New Internet-Draft is available from the on-line Internet-Drafts directories.
This draft is a work item of the IP over InfiniBand Working Group of the IETF.

	Title		: IP link and multicast over InfiniBand networks
	Author(s)	: J. Chu, V. Kashyap
	Filename	: draft-ietf-ipoib-link-multicast-02.txt
	Pages		: 12
	Date		: 20-Jun-02
	
This document specifies a method for setting up IP subnets and
multicast services over InfiniBand(TM) networks. Discussions in this
document are applicable to both IPv4 and IPv6, unless explicitly
specified. A separate document will cover unicast and encapsulation
of IP datagrams over InfiniBand networks.

A URL for this Internet-Draft is:
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-link-multicast-02.txt

To remove yourself from the IETF Announcement list, send a message to 
ietf-announce-request with the word unsubscribe in the body of the message.

Internet-Drafts are also available by anonymous FTP. Login with the username
"anonymous" and a password of your e-mail address. After logging in,
type "cd internet-drafts" and then
	"get draft-ietf-ipoib-link-multicast-02.txt".

A list of Internet-Drafts directories can be found in
http://www.ietf.org/shadow.html 
or ftp://ftp.ietf.org/ietf/1shadow-sites.txt


Internet-Drafts can also be obtained by e-mail.

Send a message to:
	mailserv@ietf.org.
In the body type:
	"FILE /internet-drafts/draft-ietf-ipoib-link-multicast-02.txt".
	
NOTE:	The mail server at ietf.org can return the document in
	MIME-encoded form by using the "mpack" utility.  To use this
	feature, insert the command "ENCODING mime" before the "FILE"
	command.  To decode the response(s), you will need "munpack" or
	a MIME-compliant mail reader.  Different MIME-compliant mail readers
	exhibit different behavior, especially when dealing with
	"multipart" MIME messages (i.e. documents which have been split
	up into multiple messages), so check your local documentation on
	how to manipulate these messages.
		
		
Below is the data which will enable a MIME compliant mail reader
implementation to automatically retrieve the ASCII version of the
Internet-Draft.

--NextPart
Content-Type: Multipart/Alternative; Boundary="OtherAccess"

--OtherAccess
Content-Type: Message/External-body;
	access-type="mail-server";
	server="mailserv@ietf.org"

Content-Type: text/plain
Content-ID:	<20020620141500.I-D@ietf.org>

ENCODING mime
FILE /internet-drafts/draft-ietf-ipoib-link-multicast-02.txt

--OtherAccess
Content-Type: Message/External-body;
	name="draft-ietf-ipoib-link-multicast-02.txt";
	site="ftp.ietf.org";
	access-type="anon-ftp";
	directory="internet-drafts"

Content-Type: text/plain
Content-ID:	<20020620141500.I-D@ietf.org>

--OtherAccess--

--NextPart--



_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@ns.ietf.org  Thu Jun 27 19:13:46 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA21921
	for <ipoverib-archive@odin.ietf.org>; Thu, 27 Jun 2002 19:13:46 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id TAA03179
	for ipoverib-archive@odin.ietf.org; Thu, 27 Jun 2002 19:14:29 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA03136;
	Thu, 27 Jun 2002 19:13:27 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id SAA01961
	for <ipoverib@ns.ietf.org>; Thu, 27 Jun 2002 18:57:59 -0400 (EDT)
Received: from nwkea-mail-1.sun.com (nwkea-mail-1.sun.com [192.18.42.13])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA21425
	for <ipoverib@ietf.org>; Thu, 27 Jun 2002 18:57:15 -0400 (EDT)
Received: from jurassic.eng.sun.com ([129.146.17.55])
	by nwkea-mail-1.sun.com (8.9.3+Sun/8.9.3) with ESMTP id PAA25206
	for <ipoverib@ietf.org>; Thu, 27 Jun 2002 15:57:28 -0700 (PDT)
Received: from sun.com (remote02.Taiwan.Sun.COM [129.158.184.212])
	by jurassic.eng.sun.com (8.12.4+Sun/8.12.4) with ESMTP id g5RMvPML511744
	for <ipoverib@ietf.org>; Thu, 27 Jun 2002 15:57:26 -0700 (PDT)
Message-ID: <3D1B9828.40403@sun.com>
Date: Thu, 27 Jun 2002 15:56:40 -0700
From: Jerry Chu <jerry.chu@sun.com>
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:0.9.4) Gecko/20011128 Netscape6/6.2.1
X-Accept-Language: en-us
MIME-Version: 1.0
To: ipoverib <ipoverib@ietf.org>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Subject: [Ipoverib] Yokohama IETF
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

With the Yokohama meeting fast approaching, it is
time to prepare for the agenda. We need to first
decide if there are enough topics to discuss. What I
need from you are -

1. how many of you are planning to attend the meeting
2. how many have slides to present (including the topic)

Please let me know asap. Even if you can't make it to
Yokohama in person, you can still make slides and give
them to me. I will present the slides on your behalf.

On a separate topic, the three basic drafts for IPoIB communication
listed below

draft-ietf-ipoib-link-multicast-02.txt
draft-ietf-ipoib-ip-over-infiniband-01.txt
draft-ietf-ipoib-dhcp-over-infiniband-01.txt

have gone through several rounds of review and seem
to be in pretty good shape for a WG last call, pending
the resolution of the promisucous multicast mode
support. Bill and I would like to see if there is
enough interest to extend the current WG chart to
cover a number of more advanced features such as

1. running IP over non-UD IB transports
2. demultiplexing IP packets into multiple QPs based
on, e.g. layer 4 fields

This is the time to bring up any remaining topics
you may have for the WG to cover. Note that in order
to get the permission from the ADs to add any work
group item to the current charter, the item must be well-defined and
scoped, with enough interest from the WG to support the work.

Your friendly co-chairs
Jerry and Bill







_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@ns.ietf.org  Thu Jun 27 19:35:42 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA22423
	for <ipoverib-archive@odin.ietf.org>; Thu, 27 Jun 2002 19:35:42 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id TAA04105
	for ipoverib-archive@odin.ietf.org; Thu, 27 Jun 2002 19:36:25 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA04090;
	Thu, 27 Jun 2002 19:35:41 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA04060
	for <ipoverib@ns.ietf.org>; Thu, 27 Jun 2002 19:35:39 -0400 (EDT)
Received: from mxout3.netvision.net.il (mxout3.netvision.net.il [194.90.9.24])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA22414
	for <ipoverib@ietf.org>; Thu, 27 Jun 2002 19:34:55 -0400 (EDT)
Received: from vaio505he ([139.92.208.95]) by mxout3.netvision.net.il
 (iPlanet Messaging Server 5.2 HotFix 0.6 (built Jun 11 2002))
 with ESMTPA id <0GYE0055R1J97G@mxout3.netvision.net.il> for ipoverib@ietf.org;
 Fri, 28 Jun 2002 02:35:36 +0300 (IDT)
Date: Fri, 28 Jun 2002 02:36:19 +0200
From: Yaron Haviv <yaronh@voltaire.com>
Subject: RE: [Ipoverib] Yokohama IETF
In-reply-to: <3D1B9828.40403@sun.com>
To: "'Jerry Chu'" <jerry.chu@sun.com>, "'ipoverib'" <ipoverib@ietf.org>
Reply-to: yaronh@voltaire.com
Message-id: <000e01c21e3b$d2ffd300$5fd05c8b@vaio505he>
MIME-version: 1.0
X-MIMEOLE: Produced By Microsoft MimeOLE V5.50.4807.1700
X-Mailer: Microsoft Outlook CWS, Build 9.0.2416 (9.0.2910.0)
Content-type: text/plain; charset=us-ascii
Content-transfer-encoding: 7BIT
Importance: Normal
X-Priority: 3 (Normal)
X-MSMail-priority: Normal
Content-Transfer-Encoding: 7BIT
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7BIT

Jerry,

Regarding the extra topics:
I think the IPoIB over RC is important to improve both performance and
robustness

The demuxing to few QP's might be useful too for some applications

Another thing that IB ARP can help in is detecting SDP (Socket Direct)
clients e.g. through a capability bit
I'm not sure if it requires changes in the WG charter

Yaron

-----Original Message-----
From: ipoverib-admin@ietf.org [mailto:ipoverib-admin@ietf.org]On Behalf Of
Jerry Chu
Sent: Friday, June 28, 2002 12:57 AM
To: ipoverib
Subject: [Ipoverib] Yokohama IETF

With the Yokohama meeting fast approaching, it is
time to prepare for the agenda. We need to first
decide if there are enough topics to discuss. What I
need from you are -

1. how many of you are planning to attend the meeting
2. how many have slides to present (including the topic)

Please let me know asap. Even if you can't make it to
Yokohama in person, you can still make slides and give
them to me. I will present the slides on your behalf.

On a separate topic, the three basic drafts for IPoIB communication
listed below

draft-ietf-ipoib-link-multicast-02.txt
draft-ietf-ipoib-ip-over-infiniband-01.txt
draft-ietf-ipoib-dhcp-over-infiniband-01.txt

have gone through several rounds of review and seem
to be in pretty good shape for a WG last call, pending
the resolution of the promisucous multicast mode
support. Bill and I would like to see if there is
enough interest to extend the current WG chart to
cover a number of more advanced features such as

1. running IP over non-UD IB transports
2. demultiplexing IP packets into multiple QPs based
on, e.g. layer 4 fields

This is the time to bring up any remaining topics
you may have for the WG to cover. Note that in order
to get the permission from the ADs to add any work
group item to the current charter, the item must be well-defined and
scoped, with enough interest from the WG to support the work.

Your friendly co-chairs
Jerry and Bill







_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@ns.ietf.org  Thu Jun 27 19:45:38 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA22838
	for <ipoverib-archive@odin.ietf.org>; Thu, 27 Jun 2002 19:45:38 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id TAA04505
	for ipoverib-archive@odin.ietf.org; Thu, 27 Jun 2002 19:46:21 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA04496;
	Thu, 27 Jun 2002 19:45:44 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id TAA04460
	for <ipoverib@ns.ietf.org>; Thu, 27 Jun 2002 19:45:42 -0400 (EDT)
Received: from e2.ny.us.ibm.com (e2.ny.us.ibm.com [32.97.182.102])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA22830
	for <ipoverib@ietf.org>; Thu, 27 Jun 2002 19:44:58 -0400 (EDT)
Received: from northrelay03.pok.ibm.com (northrelay03.pok.ibm.com [9.56.224.151])
	by e2.ny.us.ibm.com (8.12.2/8.12.2) with ESMTP id g5RNjCe8106760;
	Thu, 27 Jun 2002 19:45:12 -0400
Received: from gateway.beaverton.ibm.com (gateway.beaverton.ibm.com [138.95.180.1])
	by northrelay03.pok.ibm.com (8.11.1m3/NCO/VER6.1) with ESMTP id g5RNj8Z112342;
	Thu, 27 Jun 2002 19:45:09 -0400
Received: from eng4.beaverton.ibm.com (eng4.beaverton.ibm.com [138.95.7.64])
	by gateway.beaverton.ibm.com (8.10.0.Beta10/8.11.6) with ESMTP id g5RNj7e14855;
	Thu, 27 Jun 2002 16:45:07 -0700 (PDT)
Received: (from viv@localhost)
	by eng4.beaverton.ibm.com (8.10.0.Beta10/8.8.5/token.aware-1.2) id g5RNj4S26428;
	Thu, 27 Jun 2002 16:45:04 -0700 (PDT)
From: Vivek Kashyap <kashyapv@us.ibm.com>
Message-Id: <200206272345.g5RNj4S26428@eng4.beaverton.ibm.com>
Subject: Re: [Ipoverib] Yokohama IETF
To: yaronh@voltaire.com
Date: Thu, 27 Jun 2002 16:45:03 -0700 (PDT)
Cc: jerry.chu@sun.com ('Jerry Chu'), ipoverib@ietf.org ('ipoverib')
In-Reply-To: <000e01c21e3b$d2ffd300$5fd05c8b@vaio505he> from "Yaron Haviv" at Jun 28, 2002 01:36:19 AM PST
X-Mailer: ELM [version 2.5 PL3]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Transfer-Encoding: 7bit
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7bit

Jerry, Yaron,

        As you know these ideas have been proposed by me in the draft:

	draft-kashyap-ipoib-advanced-capabilities-00.txt.

	I had presented a set of slides at IETF-53. See slides 16 to 24 at:

	http://www.ietf.org/proceedings/02mar/slides/ipoib-1/index.html

Vivek

PS. I won't be attending IETF-54.

> 
> Jerry,
> 
> Regarding the extra topics:
> I think the IPoIB over RC is important to improve both performance and
> robustness
> 
> The demuxing to few QP's might be useful too for some applications
> 
> Another thing that IB ARP can help in is detecting SDP (Socket Direct)
> clients e.g. through a capability bit
> I'm not sure if it requires changes in the WG charter
> 
> Yaron
> 
> -----Original Message-----
> From: ipoverib-admin@ietf.org [mailto:ipoverib-admin@ietf.org]On Behalf Of
> Jerry Chu
> Sent: Friday, June 28, 2002 12:57 AM
> To: ipoverib
> Subject: [Ipoverib] Yokohama IETF
> 
> With the Yokohama meeting fast approaching, it is
> time to prepare for the agenda. We need to first
> decide if there are enough topics to discuss. What I
> need from you are -
> 
> 1. how many of you are planning to attend the meeting
> 2. how many have slides to present (including the topic)
> 
> Please let me know asap. Even if you can't make it to
> Yokohama in person, you can still make slides and give
> them to me. I will present the slides on your behalf.
> 
> On a separate topic, the three basic drafts for IPoIB communication
> listed below
> 
> draft-ietf-ipoib-link-multicast-02.txt
> draft-ietf-ipoib-ip-over-infiniband-01.txt
> draft-ietf-ipoib-dhcp-over-infiniband-01.txt
> 
> have gone through several rounds of review and seem
> to be in pretty good shape for a WG last call, pending
> the resolution of the promisucous multicast mode
> support. Bill and I would like to see if there is
> enough interest to extend the current WG chart to
> cover a number of more advanced features such as
> 
> 1. running IP over non-UD IB transports
> 2. demultiplexing IP packets into multiple QPs based
> on, e.g. layer 4 fields
> 
> This is the time to bring up any remaining topics
> you may have for the WG to cover. Note that in order
> to get the permission from the ADs to add any work
> group item to the current charter, the item must be well-defined and
> scoped, with enough interest from the WG to support the work.
> 
> Your friendly co-chairs
> Jerry and Bill
> 
> 
> 
> 
> 
> 
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 
> 
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
> 


-- 
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com 
vivk@us.ibm.com 
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



From daemon@optimus.ietf.org  Fri Jun 28 15:09:22 2002
Received: from optimus.ietf.org (ietf.org [132.151.1.19] (may be forged))
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA15227
	for <ipoverib-archive@odin.ietf.org>; Fri, 28 Jun 2002 15:09:22 -0400 (EDT)
Received: (from daemon@localhost)
	by optimus.ietf.org (8.9.1a/8.9.1) id PAA23878
	for ipoverib-archive@odin.ietf.org; Fri, 28 Jun 2002 15:10:07 -0400 (EDT)
Received: from optimus.ietf.org (localhost [127.0.0.1])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id PAA23485;
	Fri, 28 Jun 2002 15:07:02 -0400 (EDT)
Received: from ietf.org (odin [132.151.1.176])
	by optimus.ietf.org (8.9.1a/8.9.1) with ESMTP id PAA23446
	for <ipoverib@optimus.ietf.org>; Fri, 28 Jun 2002 15:07:00 -0400 (EDT)
Received: from mxout1.netvision.net.il (mxout1.netvision.net.il [194.90.9.20])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA15059
	for <ipoverib@ietf.org>; Fri, 28 Jun 2002 15:06:14 -0400 (EDT)
Received: from vaio505he ([139.92.208.142]) by mxout1.netvision.net.il
 (iPlanet Messaging Server 5.2 HotFix 0.6 (built Jun 11 2002))
 with ESMTPA id <0GYF00IGQJJOWD@mxout1.netvision.net.il> for ipoverib@ietf.org;
 Fri, 28 Jun 2002 22:02:14 +0300 (IDT)
Date: Fri, 28 Jun 2002 22:03:36 +0200
From: Yaron Haviv <yaronh@voltaire.com>
Subject: RE: [Ipoverib] Yokohama IETF
In-reply-to: <200206272345.g5RNj4S26428@eng4.beaverton.ibm.com>
To: "'Vivek Kashyap'" <kashyapv@us.ibm.com>
Cc: "'Jerry Chu'" <jerry.chu@sun.com>, "'ipoverib'" <ipoverib@ietf.org>
Reply-to: yaronh@voltaire.com
Message-id: <000301c21ede$e44782e0$8ed05c8b@vaio505he>
MIME-version: 1.0
X-MIMEOLE: Produced By Microsoft MimeOLE V5.50.4807.1700
X-Mailer: Microsoft Outlook CWS, Build 9.0.2416 (9.0.2910.0)
Content-type: text/plain; charset=us-ascii
Content-transfer-encoding: 7BIT
Importance: Normal
X-Priority: 3 (Normal)
X-MSMail-priority: Normal
Content-Transfer-Encoding: 7BIT
Sender: ipoverib-admin@ietf.org
Errors-To: ipoverib-admin@ietf.org
X-Mailman-Version: 1.0
Precedence: bulk
List-Id: IP over InfiniBand WG Discussion List <ipoverib.ietf.org>
X-BeenThere: ipoverib@ietf.org
Content-Transfer-Encoding: 7BIT

Vivek,

Maybe we can add to the IPoIB RC a feature we once discussed
For TCP/UDP checksum offloading no need to specify the exact method now
Just mention it is a possible future goal

Yaron

-----Original Message-----
From: ipoverib-admin@ietf.org [mailto:ipoverib-admin@ietf.org]On Behalf Of
Vivek Kashyap
Sent: Friday, June 28, 2002 1:45 AM
To: yaronh@voltaire.com
Cc: 'Jerry Chu'; 'ipoverib'
Subject: Re: [Ipoverib] Yokohama IETF

Jerry, Yaron,

        As you know these ideas have been proposed by me in the draft:

        draft-kashyap-ipoib-advanced-capabilities-00.txt.

        I had presented a set of slides at IETF-53. See slides 16 to 24 at:

        http://www.ietf.org/proceedings/02mar/slides/ipoib-1/index.html

Vivek

PS. I won't be attending IETF-54.

>
> Jerry,
>
> Regarding the extra topics:
> I think the IPoIB over RC is important to improve both performance and
> robustness
>
> The demuxing to few QP's might be useful too for some applications
>
> Another thing that IB ARP can help in is detecting SDP (Socket Direct)
> clients e.g. through a capability bit
> I'm not sure if it requires changes in the WG charter
>
> Yaron
>
> -----Original Message-----
> From: ipoverib-admin@ietf.org [mailto:ipoverib-admin@ietf.org]On Behalf Of
> Jerry Chu
> Sent: Friday, June 28, 2002 12:57 AM
> To: ipoverib
> Subject: [Ipoverib] Yokohama IETF
>
> With the Yokohama meeting fast approaching, it is
> time to prepare for the agenda. We need to first
> decide if there are enough topics to discuss. What I
> need from you are -
>
> 1. how many of you are planning to attend the meeting
> 2. how many have slides to present (including the topic)
>
> Please let me know asap. Even if you can't make it to
> Yokohama in person, you can still make slides and give
> them to me. I will present the slides on your behalf.
>
> On a separate topic, the three basic drafts for IPoIB communication
> listed below
>
> draft-ietf-ipoib-link-multicast-02.txt
> draft-ietf-ipoib-ip-over-infiniband-01.txt
> draft-ietf-ipoib-dhcp-over-infiniband-01.txt
>
> have gone through several rounds of review and seem
> to be in pretty good shape for a WG last call, pending
> the resolution of the promisucous multicast mode
> support. Bill and I would like to see if there is
> enough interest to extend the current WG chart to
> cover a number of more advanced features such as
>
> 1. running IP over non-UD IB transports
> 2. demultiplexing IP packets into multiple QPs based
> on, e.g. layer 4 fields
>
> This is the time to bring up any remaining topics
> you may have for the WG to cover. Note that in order
> to get the permission from the ADs to add any work
> group item to the current charter, the item must be well-defined and
> scoped, with enough interest from the WG to support the work.
>
> Your friendly co-chairs
> Jerry and Bill
>
>
>
>
>
>
>
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
>
>
> _______________________________________________
> IPoverIB mailing list
> IPoverIB@ietf.org
> https://www1.ietf.org/mailman/listinfo/ipoverib
>


--
Vivek Kashyap
Linux Technology Center, IBM
kashyapv@us.ibm.com
vivk@us.ibm.com
503 578 3422 (o)

_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib


_______________________________________________
IPoverIB mailing list
IPoverIB@ietf.org
https://www1.ietf.org/mailman/listinfo/ipoverib



