From dhcp-v4@bucknell.edu  Mon Nov  1 10:29:41 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA11206
	for <dhc-archive@odin.ietf.org>; Mon, 1 Nov 1999 10:29:41 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA19394;
	Mon, 1 Nov 1999 10:28:09 -0500 (EST)
Date: Mon, 1 Nov 1999 10:28:09 -0500 (EST)
Message-Id: <199911011452.JAA00300@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v4@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: dhcp-v4@bucknell.edu (by way of Ralph Droms <dhcp-v4@bucknell.edu>)
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: I-D ACTION:draft-ietf-dhc-dhcp-dns-11.txt
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

A New Internet-Draft is available from the on-line Internet-Drafts
directories.
This draft is a work item of the Dynamic Host Configuration Working Group
of the IETF.

	Title		: Interaction between DHCP and DNS
	Author(s)	: M. Stapp, Y. Rekhter
	Filename	: draft-ietf-dhc-dhcp-dns-11.txt
	Pages		: 18
	Date		: 28-Oct-99
	
DHCP provides a powerful mechanism for IP host configuration.
However, the configuration capability provided by DHCP does not
include updating DNS, and specifically updating the name to address
and address to name mappings maintained in the DNS.
This document specifies how DHCP clients and servers should use the
Dynamic DNS Updates mechanism in [RFC2136] to update the DNS name to
address and address to name mappings so that the mappings for DHCP
clients will be consistent with the IP addresses that the clients
acquire via DHCP.

A URL for this Internet-Draft is:

http://www.ietf.org/internet-drafts/draft-ietf-dhc-dhcp-dns-11.txt

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-dhc-dhcp-dns-11.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-dhc-dhcp-dns-11.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.
Content-Type: text/plain
Content-ID:	<19991029091856.I-D@ietf.org>

ENCODING mime
FILE /internet-drafts/draft-ietf-dhc-dhcp-dns-11.txt

<ftp://ftp.ietf.org/internet-drafts/draft-ietf-dhc-dhcp-dns-11.txt>



From dhcp-v4@bucknell.edu  Mon Nov  1 10:31:24 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA11649
	for <dhc-archive@odin.ietf.org>; Mon, 1 Nov 1999 10:31:22 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA19422;
	Mon, 1 Nov 1999 10:30:55 -0500 (EST)
Date: Mon, 1 Nov 1999 10:30:55 -0500 (EST)
Message-Id: <63D30D6E10CFD11190A90000F805FE8601CE6FAC@lespaul.process.com>
Errors-To: droms@bucknell.edu
Reply-To: Gonczi@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Steve Gonczi <Gonczi@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: X-Server on DHCP
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.0.1460.8)
Mime-Version: 1.0

Possibly, your IP address that the X client is trying to display to is
hard-wired to
your old address?
I am not familiar with your specific X server, but  it is possible that your
X session
was set up to a hard-wired IP address. Also, you may have an environmental
variable 
set on the UNIX system that points to your old IP address.

The error message just says:
"The X terminal client is trying to display to the default window on X
server
"Myip" and can not".

All else being equal, your  IP address is the most likely suspect.
BTW this is really not a DHCP issue.
> ----------
> From: 	dhcp-v4@bucknell.edu[SMTP:dhcp-v4@bucknell.edu]
> Reply To: 	dhcp-v4@bucknell.edu
> Sent: 	Sunday, October 31, 1999 5:34 PM
> To: 	Multiple recipients of list
> Subject: 	X-Server on DHCP
> 
> 
> message sent to dhcp-v4@bucknell.edu by LAMDV@YAHOO.COM follows.
> ----------------------------------------------------------------
> Hi,
> 
> I was wondering if I can get any tip how to resolve my
> problem.  Currently I am using an X-server software
> (X-Win32) to connect to my school unix server from
> home.  Before switching to an ISP that has dynamic IP,
> I was successfully remoted the unix server without any
> trouble.  However, I keep receiving an error message
> "/usr/openwin/bin/xterm Xt error: Can't open display:
> MYIP:0" eversince I switch my ISP.  I contacted the
> program vendor but so far has not had any luck
> following their instruction.  I was wondering if any
> of you have any suggestion for me since I found your
> email addresses searching for DHCP informations. 
> Thanks in advance for all your help.
> 
> Best regards,
> 
> Daniel
> 
> 
> 
> 
> =====
> -----------------------------
> She...bewild my heart I know not why
> Yet, I love her 'til I die !
> 
> -William Shakespeare-
> __________________________________________________
> Do You Yahoo!?
> Bid and sell for free at http://auctions.yahoo.com
> 


From dhcp-v4@bucknell.edu  Mon Nov  1 13:37:58 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA13010
	for <dhc-archive@odin.ietf.org>; Mon, 1 Nov 1999 13:37:58 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA26266;
	Mon, 1 Nov 1999 12:00:38 -0500 (EST)
Date: Mon, 1 Nov 1999 12:00:38 -0500 (EST)
Message-Id: <195339D90D04D3118B0E0090274EFB320129C5BD@msgcreo4.creo.bc.ca>
Errors-To: droms@bucknell.edu
Reply-To: PBaker@creo.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Peter Baker <PBaker@creo.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: I-D ACTION:draft-ietf-dhc-authentication-12.txt
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

Please remove me from this mailing list.

Thanks,
..Peter Baker


-----Original Message-----
From: dhcp-v4@bucknell.edu [mailto:dhcp-v4@bucknell.edu]
Sent: Sunday, October 31, 1999 2:31 PM
To: Multiple recipients of list
Subject: I-D ACTION:draft-ietf-dhc-authentication-12.txt


A New Internet-Draft is available from the on-line Internet-Drafts
directories.
This draft is a work item of the Dynamic Host Configuration Working Group
of the IETF.

	Title		: Authentication for DHCP Messages
	Author(s)	: R. Droms, W. Arbaugh
	Filename	: draft-ietf-dhc-authentication-12.txt
	Pages		: 16
	Date		: 26-Oct-99
	
The Dynamic Host Configuration Protocol (DHCP) provides a framework
for passing configuration information to hosts on a TCP/IP network.
In some situations, network administrators may wish to constrain the
allocation of addresses to authorized hosts.  Additionally, some
network administrators may wish to provide for authentication of the
source and contents of DHCP messages.  This document defines a new
DHCP option through which authorization tickets can be easily
generated and newly attached hosts with proper authorization can be
automatically configured from an authenticated DHCP server.

A URL for this Internet-Draft is:

http://www.ietf.org/internet-drafts/draft-ietf-dhc-authentication-12.txt

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-dhc-authentication-12.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-dhc-authentication-12.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.
Content-Type: text/plain
Content-ID:	<19991026145042.I-D@ietf.org>

ENCODING mime
FILE /internet-drafts/draft-ietf-dhc-authentication-12.txt

<ftp://ftp.ietf.org/internet-drafts/draft-ietf-dhc-authentication-12.txt>


From dhcp-v4@bucknell.edu  Tue Nov  2 09:28:44 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA13090
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 09:28:44 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA29795;
	Tue, 2 Nov 1999 09:28:06 -0500 (EST)
Date: Tue, 2 Nov 1999 09:28:06 -0500 (EST)
Message-Id: <Version.32.19991101191654.00e0f990@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Authentication draft
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

*Please* be sure to read and review carefully the current DHCP
authentication draft:


http://www.ietf.org/internet-drafts/draft-ietf-dhc-authentication-12.txt

If you have comments, bring them up on the mailing list as soon as
possible.  Bill and I would *very* much like to take the draft to last call
after the DC WG meeting.  At the last couple of meetings, we've received
last minute comments that have caused us to recycle the draft.  We'd like
to avoid that at all costs for this meeting.  Thanks...

- Ralph




From dhcp-v4@bucknell.edu  Tue Nov  2 09:32:43 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA14175
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 09:32:42 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA24741;
	Tue, 2 Nov 1999 09:32:12 -0500 (EST)
Date: Tue, 2 Nov 1999 09:32:12 -0500 (EST)
Message-Id: <Version.32.19991101210404.00df2ad0@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v4@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: dhcp-v4@bucknell.edu (by way of Ralph Droms <dhcp-v4@bucknell.edu>)
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: dhcp impl!!!
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


message sent to dhcp-v4@bucknell.edu by KILLERMAMA@HOTMAIL.COM follows.
-----------------------------------------------------------------------

sir,
     I have installed DHCP ON my LINUX machine.I need to add data
to the Network  Registrar Database through command line 
interface(CLI)nrcmd.i am unable to get the command line interface.if u could 
send
me some material regarding the confoguration of dhcp on linux.

thanking u
sriman



______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com



From dhcp-v6@bucknell.edu  Tue Nov  2 10:12:28 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA25727
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 10:12:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA15078;
	Tue, 2 Nov 1999 10:09:00 -0500 (EST)
Date: Tue, 2 Nov 1999 10:09:00 -0500 (EST)
Message-Id: <199911021435.JAA22603@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Revised DHC WG agenda
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Here is the most recent DHC WG agenda for the upcoming IETF meeting in DC.
Notice that the 0900 session is overfull and I may need to shift some
topics to the 1300 session.

- Ralph


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

  Yves T'Joens (Alcatel)         DHCP reconfigure message

  Bill Arbaugh (Penn)            DHCPv4 authentication

  Carl Smith (Sun)               Name service ordering option

  Kim Kinnear (Cisco)            Failover protocol

  Jerome Privat (BT)             User Class option

  Jerome Privat (BT)             Doublephase configuration

  Mike Patrick (Motorola)        Agent options

  Subir Das (Telcordia)          Roaming among commercial
                                 wireless networks

  Bernard Aboba (Microsoft)      Kerberos authentication

  Bernie Volz (Process Software) Server load balancing

  Bill Sommerfield (Epilogue)    Server selection option

  Bernie Volz (Process Software) LDAP schema

  Barr Hibbs (Pacbell)           DHCP MIB

  Barr Hibbs (Pacbell)           Client option clarifications
 
  Juan Garay (Bell Labs)         Option for NAT/IPSec interaction

Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================
  Mark Stapp (Cisco)             DHCP-DNS interaction

  Andreas Gustafsson             A DNS RR for DHCP information
  (Internet Engines, Inc.)

  Mike Carney (Sun)              DHCP options guidelines

  Mike Carney (Sun)              Connectathon

Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
  Mike Carney (Sun)              DHCPv6
  Jim Bound (Compaq)
  Charlie Perkins (Nokia)




From dhcp-v4@bucknell.edu  Tue Nov  2 10:14:29 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA26332
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 10:14:28 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA00554;
	Tue, 2 Nov 1999 10:13:54 -0500 (EST)
Date: Tue, 2 Nov 1999 10:13:54 -0500 (EST)
Message-Id: <199911021435.JAA22603@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Revised DHC WG agenda
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Here is the most recent DHC WG agenda for the upcoming IETF meeting in DC.
Notice that the 0900 session is overfull and I may need to shift some
topics to the 1300 session.

- Ralph


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

  Yves T'Joens (Alcatel)         DHCP reconfigure message

  Bill Arbaugh (Penn)            DHCPv4 authentication

  Carl Smith (Sun)               Name service ordering option

  Kim Kinnear (Cisco)            Failover protocol

  Jerome Privat (BT)             User Class option

  Jerome Privat (BT)             Doublephase configuration

  Mike Patrick (Motorola)        Agent options

  Subir Das (Telcordia)          Roaming among commercial
                                 wireless networks

  Bernard Aboba (Microsoft)      Kerberos authentication

  Bernie Volz (Process Software) Server load balancing

  Bill Sommerfield (Epilogue)    Server selection option

  Bernie Volz (Process Software) LDAP schema

  Barr Hibbs (Pacbell)           DHCP MIB

  Barr Hibbs (Pacbell)           Client option clarifications
 
  Juan Garay (Bell Labs)         Option for NAT/IPSec interaction

Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================
  Mark Stapp (Cisco)             DHCP-DNS interaction

  Andreas Gustafsson             A DNS RR for DHCP information
  (Internet Engines, Inc.)

  Mike Carney (Sun)              DHCP options guidelines

  Mike Carney (Sun)              Connectathon

Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
  Mike Carney (Sun)              DHCPv6
  Jim Bound (Compaq)
  Charlie Perkins (Nokia)




From dhcp-v4@bucknell.edu  Tue Nov  2 11:51:30 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA02039
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 11:51:29 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA27375;
	Tue, 2 Nov 1999 11:49:46 -0500 (EST)
Date: Tue, 2 Nov 1999 11:49:46 -0500 (EST)
Message-Id: <009E08AE.A517A4E4.134@process.com>
Errors-To: droms@bucknell.edu
Reply-To: volz@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Bernie Volz <volz@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Authentication draft
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4

Per Ralph's request, here are some comments on the 
http://www.ietf.org/internet-drafts/draft-ietf-dhc-authentication-12.txt.
Most of these are more editorial type issues.

1. [Minor] In section 5.1, "This protocol is focused [on] solving ...".

2. In section 5.2, can the sizes of the secret ID (32-bits) and HMAC-MD5
fields (256-bits I believe) be explicited specified? This also means that
the length for the options can be specified, since I believe they are
fixed.

3. In section 5.3, 'counter' is referenced. What is this? I assume it
should be 'Replay Detection'. Also, see counter references in 5.6.1.

4. In Appendix A, the key is generated using the "subnet address". Do you
mean the [proposed] address for the client or the SUBNET address (ie,
client's address ANDed with subnet mask).

5. In section 5.5.1, item 2, it references section 5.2. This should be
5.3.

6. [Minor References to other documents are somewhat inconsistent. Note
for example the references to the DHCP specification in section 5.5.2
and 5.5.3.

7. In sections 5.6.2, .3 there are references to section 4.1 and 4.2.
There are no corresponding sections in the draft.

- Bernie Volz
  Process Software


From dhcp-v4@bucknell.edu  Tue Nov  2 15:46:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA02277
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 15:46:02 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA14649;
	Tue, 2 Nov 1999 15:43:05 -0500 (EST)
Date: Tue, 2 Nov 1999 15:43:05 -0500 (EST)
Message-Id: <Version.32.19991102150717.00e26970@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: Internet-Drafts@ietf.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Internet-Drafts@ietf.org (by way of Ralph Droms <Internet-Drafts@ietf.org>)
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: I-D ACTION:draft-itsumo-drcp-00.txt
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

A New Internet-Draft is available from the on-line Internet-Drafts
directories.


	Title		: Dynamic Registration and Configuration Protocol (DRCP)
	Author(s)	: A. McAuley, S. Das, S. Baba, Y. Shobatake
	Filename	: draft-itsumo-drcp-00.txt
	Pages		: 22
	Date		: 28-Oct-99
	
The Dynamic Host Configuration Protocol (DHCP) provides a framework
for passing configuration information to hosts [DHC].  DHCP was,
however, designed for hosts on a fixed LAN, not for nodes roaming
among commercial wireless networks.  Mobile IP [MIP], when combined
with some recent proposals [e.g., MIPA, MIPD, MIPN, HAW, CEL, TIA],
gives a powerful plug and play capability for roaming hosts, but the
additional functionality may not be needed or may be better provided
using some alternative mechanisms.  This draft proposes an enhanced
version of DHCP for roaming users, called the Dynamic Registration
and Configuration Protocol (DRCP).  DRCP borrows heavily from DHCP
and can revert to using DHCP protocol if only DHCP servers are
present in the network; but adds features critical to roaming users.
Most importantly, DRCP allows rapid configuration using several
modifications, including moving address consistency checking from the
critical path.  Other new features allow:  a) clients to know when to
get a new address independent of the access technology, b) efficient
use of scarce wireless bandwidth, c) nodes to be a relay agent or
server depending on a client's authorization, d) clients to be
routers, e) associating a priority with addresses, and f) identifying
users (e.g., by NAI [NAI]) rather than hardware address.  DRCP also
adds options for QOS negotiation, service activation, and
authentication.
In some sense DRCP is to DHCP what DHCP is to BOOTP.  DRCP directly
supports intra-domain registration and configuration, but must be
combined with other protocols to provide features such as continuous
connectivity (e.g., using Mobile IP or SIP [SIPM]), or inter-domain
authentication, authorization and accounting (e.g., using Radius
[RAD] or Diameter [DIA]).

A URL for this Internet-Draft is:

http://www.ietf.org/internet-drafts/draft-itsumo-drcp-00.txt

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-itsumo-drcp-00.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-itsumo-drcp-00.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.
Content-Type: text/plain
Content-ID:	<19991028141850.I-D@ietf.org>

ENCODING mime
FILE /internet-drafts/draft-itsumo-drcp-00.txt

<ftp://ftp.ietf.org/internet-drafts/draft-itsumo-drcp-00.txt>



From dhcp-v4@bucknell.edu  Tue Nov  2 17:39:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA16230
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 17:39:02 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA28375;
	Tue, 2 Nov 1999 17:37:42 -0500 (EST)
Date: Tue, 2 Nov 1999 17:37:42 -0500 (EST)
Message-Id: <199911022228.RAA04342@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Possible ambiguity in definition of DHCP client identifier option.
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


I'm pretty sure it's not what's intended, but because of the way the
parts of the client identifier option are labeled, it's possible for
somebody to interpret the definition of the client-identifier option
in the following way:

  Btw, I just read Ted's latest e-mail, and the impression I get is that the
  server is using the type as part of the client identifier, which I believe
  is in violation of RFC 2132's description of option 61, which puts the type
  separate from what it calls the client identifier.  I want to make sure that
  people understand how I am interpreting client identifier so that we
  understand each other.  Ted, are you including the type as part of the
  client identifier?  If so, why do you think that's valid (given the diagram
  describing option 61 in 2132).

Here's the text from the RFC:

   For correct identification of clients, each client's client-
   identifier MUST be unique among the client-identifiers used on the
   subnet to which the client is attached.  Vendors and system
   administrators are responsible for choosing client-identifiers that
   meet this requirement for uniqueness.

   The code for this option is 61, and its minimum length is 2.

   Code   Len   Type  Client-Identifier
   +-----+-----+-----+-----+-----+---
   |  61 |  n  |  t1 |  i1 |  i2 | ...
   +-----+-----+-----+-----+-----+---

As you can see, it's not clear whether the phrase "client identifier"
in the opening paragraph is referring to the entire contents of the
option, or the contents that are labeled "Client-Identifier" in the
diagram.   I have always interpreted this as the contents of the
option from the first byte of payload to the last, not just the part
following the type code.

Any clues?


From dhcp-v4@bucknell.edu  Tue Nov  2 19:22:09 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA16668
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 19:22:09 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA09923;
	Tue, 2 Nov 1999 19:21:14 -0500 (EST)
Date: Tue, 2 Nov 1999 19:21:14 -0500 (EST)
Message-Id: <63D30D6E10CFD11190A90000F805FE8601CE6FB1@lespaul.process.com>
Errors-To: droms@bucknell.edu
Reply-To: Gonczi@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Steve Gonczi <Gonczi@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.0.1460.8)
Mime-Version: 1.0

Ted's interpretation makes sense. That is how I always read it.

RFC 2131 says on page 9:

"The 'client identifier' is an opaque key, not to be interpreted by the
server;"

RFC 2132 says on page 30:

"Identifiers SHOULD be treated as opaque objects by DHCP servers.
The client identifier MAY consist of type-value pairs similar to the
'htype'/'chaddr' fields defined in [3]. "

It seems clear (to me), that the "client identifier" 
is the whole payload of the option, and NOT just i1..in


From dhcp-v4@bucknell.edu  Tue Nov  2 19:59:29 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA27661
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 19:59:28 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA14750;
	Tue, 2 Nov 1999 19:58:57 -0500 (EST)
Date: Tue, 2 Nov 1999 19:58:57 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE20@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


I agree with Steve that a client identifier should be the entire data
payload of option 61 -- an opaque string whose composition should not be the
concern of the server.

I also think that the current RFC wording is a slight disservice to
implementors because some I've encountered read the text suggesting how to
construct a client identifier from hardware type and MAC address as a
requirement that only octet strings which can be decomposed into a hardware
type and address part are legitimate client identifiers.

--Barr




From dhcp-v4@bucknell.edu  Tue Nov  2 20:37:19 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA06861
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 20:37:18 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA16187;
	Tue, 2 Nov 1999 20:35:31 -0500 (EST)
Date: Tue, 2 Nov 1999 20:35:31 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.941591377.12297.mwc@atlantic.east.sun.com.>
Errors-To: droms@bucknell.edu
Reply-To: Michael.Carney@East.Sun.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

> 
> I agree with Steve that a client identifier should be the entire data
> payload of option 61 -- an opaque string whose composition should not be the
> concern of the server.
> 
> I also think that the current RFC wording is a slight disservice to
> implementors because some I've encountered read the text suggesting how to
> construct a client identifier from hardware type and MAC address as a
> requirement that only octet strings which can be decomposed into a hardware
> type and address part are legitimate client identifiers.
> 
> --Barr
> 
> 

Historically, a client id has always been an "opaque string" to the server. The
text suggesting how one might form one is just that - a suggestion. Folks were
asking for guidance on how one might form a client id, so we added the text. The
draft text seems very clear on this; I wonder if people are just used to seeing
client identifiers  of the ARP type + hardware address so much that they're
making incorrect assumptions about client identifiers.





From dhcp-v4@bucknell.edu  Tue Nov  2 21:22:07 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA19463
	for <dhc-archive@odin.ietf.org>; Tue, 2 Nov 1999 21:22:06 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA32064;
	Tue, 2 Nov 1999 21:20:20 -0500 (EST)
Date: Tue, 2 Nov 1999 21:20:20 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE25@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


Mike--

you always have good historical perspectives on these matters....


> Historically, a client id has always been an "opaque string" to the
> server. The text suggesting how one might form one is just that - a
> suggestion. Folks were asking for guidance on how one might form a client
> id, so we added the text. The draft text seems very clear on this; I
> wonder if people are just used to seeing client identifiers  of the ARP
> type + hardware address so much that they're making incorrect assumptions
> about client identifiers.
> 
--I can tell you for a fact that more than one vendor of clients believes
wholeheartedly that the only possible value for a client identifier is
constructed only from the hardware type and chaddr.  The example text is a
bit unfortunate because it is so easily misinterpreted.

--Barr






From dhcp-v4@bucknell.edu  Wed Nov  3 04:49:32 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id EAA01778
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 04:49:32 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id EAA10432;
	Wed, 3 Nov 1999 04:49:03 -0500 (EST)
Date: Wed, 3 Nov 1999 04:49:03 -0500 (EST)
Message-Id: <F5690081C776D111BDA300805FE6BFC33DA093@NTSG24>
Errors-To: droms@bucknell.edu
Reply-To: ORIGIN@SNPE.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "ORIGIN (Prest.)" <ORIGIN@SNPE.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

Hi,

I would like to extract the list there is on the dhcp manager (adress ip
with the scope, and adress in use, adress reserved).
It would be easier to administrate.

In winnt\system32\dhcp, there is a database file I don't manage to read....?

Can you help me?

Marc Broussard


From dhcp-v4@bucknell.edu  Wed Nov  3 09:48:43 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA10555
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 09:48:43 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA15905;
	Wed, 3 Nov 1999 09:47:39 -0500 (EST)
Date: Wed, 3 Nov 1999 09:47:39 -0500 (EST)
Message-Id: <009E0962.67991AFA.84@process.com>
Errors-To: droms@bucknell.edu
Reply-To: volz@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Bernie Volz <volz@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4

Seems like most people feel the type is included as part of the client
identifier ... 

In the DHCP LDAP Schema draft we've "collapsed" mac-address and
client-identifier into one attribute (dhcpclientidentifier). The idea
was that since the client identifier should be prefixed by 0, if it
was not a hardware address, as specified in RFC2132, we could simply
combine these fields. That way, if you wanted to identify a client
by mac address, you could be using <media-type>:<mac-address>.

So, the question is ... does everyone (client implementation) follow the
rules - specifying t1 as 0 if it is not a mac address.

- Bernie Volz
  Process Software


From dhcp-v4@bucknell.edu  Wed Nov  3 09:56:12 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA13877
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 09:56:12 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA15728;
	Wed, 3 Nov 1999 09:54:36 -0500 (EST)
Date: Wed, 3 Nov 1999 09:54:36 -0500 (EST)
Message-Id: <4.2.0.58.19991103091656.0153e140@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: kkinnear@cisco.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Kim Kinnear <kkinnear@cisco.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 


I too believe that the client-identifier is the *entire* payload
of the client-identifier option, and that it is opaque to the
DHCP server.

I think the draft is reasonably clear on this point, but that
doesn't mean that it couldn't be made even more clear -- which
would probably be a good idea.

Cheers -- Kim



From dhcp-v4@bucknell.edu  Wed Nov  3 10:48:17 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA01202
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 10:48:13 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA01126;
	Wed, 3 Nov 1999 10:47:27 -0500 (EST)
Date: Wed, 3 Nov 1999 10:47:27 -0500 (EST)
Message-Id: <199911031450.JAA00859@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier option. 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> I think the draft is reasonably clear on this point, but that
> doesn't mean that it couldn't be made even more clear -- which
> would probably be a good idea.

Yup.  Basically, the goal of making a change would be to help people
who don't already have a good understanding of the DHCP protocol that
has evolved over time.   This is one of those cases where although
it's obvious to a seasoned DHCP hacker what the text means, it's not
so obvious to a neophyte, and it's to everybody's advantage to make
sure that neophytes have the information they need.   :')

BTW, I don't know if it's possible at this late date, but I'd really
like to see the text "the client MAY choose to explicitly provide the
identifier through the 'client identifier' option" to "the client
SHOULD choose to explicitly provide the identifier through the 'client
identifier' option".   Right now I think client implementors tend to
read this as saying that the client identifier is optional (that's
certainly how I read it) and may not feel any real motivation to use a
client identifier.   To say the least, the way client identifiers are
currently handled is quite a bummer.   If I had my druthers, even
though the ISC DHCP client doesn't currently implement it, I'd change
this to a MUST, with a caveat that servers should allow for clients
that don't supply client identifiers for backward compatibility.

			       _MelloN_



From dhcp-v4@bucknell.edu  Wed Nov  3 10:50:55 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA02331
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 10:50:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA13524;
	Wed, 3 Nov 1999 10:49:57 -0500 (EST)
Date: Wed, 3 Nov 1999 10:49:57 -0500 (EST)
Message-Id: <EA5107D3A9ADD211B7530000C041E4F4011B2AD1@honts318.wal-mart.com>
Errors-To: droms@bucknell.edu
Reply-To: ndlane@wal-mart.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Nathan Lane <ndlane@wal-mart.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0

At least the largest one does as of service pack 2 or later - Windows NT 4.0.
When setting a CID that is different from the MAC address, t1 is 0.
Most embedded systems vendors don't let us change the CID, so they 
do the MAC thing that has been discussed.

-Nathan Lane
Wal-Mart Stores, Inc.
(501) 277-5786

> -----Original Message-----
> From:	Bernie Volz [SMTP:volz@process.com]
> Sent:	Wednesday, November 03, 1999 8:52 AM
> To:	Multiple recipients of list
> Subject:	RE: Possible ambiguity in definition of DHCP client identifier op
> 
> 
> So, the question is ... does everyone (client implementation) follow the
> rules - specifying t1 as 0 if it is not a mac address.
> 
> - Bernie Volz
>   Process Software


From dhcp-v4@bucknell.edu  Wed Nov  3 10:52:44 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA02984
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 10:52:43 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA00222;
	Wed, 3 Nov 1999 10:52:16 -0500 (EST)
Date: Wed, 3 Nov 1999 10:52:16 -0500 (EST)
Message-Id: <199911031504.KAA00906@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier op 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> So, the question is ... does everyone (client implementation) follow the
> rules - specifying t1 as 0 if it is not a mac address.

Absolutely not, nor does the protocol require it.  Further, it's not
specified that cid==1:4a:2b:ac:19:ad means the same thing as
htype==1/hlen=6/chaddr==4a:2b:ac:19:ad:dc:dc:dc:dc:dc:dc:dc:dc:dc:dc
(dc == don't care).

If you want to collapse the lease database lookup key into a single
key value, which I think is reasonable, you really need to have some
kind of additional type that says where the key came from.   I would
suggest zero for htype/hlen/chaddr and 61 for dhcp-client-identifier.
Otherwise I think that your specification for the construction of a
key based on htype/hlen/chaddr is correct.

			       _MelloN_


From dhcp-v4@bucknell.edu  Wed Nov  3 10:55:11 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA03979
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 10:55:09 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA25937;
	Wed, 3 Nov 1999 10:54:22 -0500 (EST)
Date: Wed, 3 Nov 1999 10:54:22 -0500 (EST)
Message-Id: <4.2.0.58.19991103100245.0162f380@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: kkinnear@cisco.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Kim Kinnear <kkinnear@cisco.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 

At 09:49 AM 11/3/99 -0500, Bernie Volz wrote:
>Seems like most people feel the type is included as part of the client
>identifier ... 
>
>In the DHCP LDAP Schema draft we've "collapsed" mac-address and
>client-identifier into one attribute (dhcpclientidentifier). The idea
>was that since the client identifier should be prefixed by 0, if it
>was not a hardware address, as specified in RFC2132, we could simply
>combine these fields. That way, if you wanted to identify a client
>by mac address, you could be using <media-type>:<mac-address>.
>
>So, the question is ... does everyone (client implementation) follow the
>rules - specifying t1 as 0 if it is not a mac address.

         I think the only rule about the content of the
         client-identifier is that there *aren't* any rules about
         content of the client identifier -- isn't that what
         "opaque" means?  The only rule is that they must be
         unique within an adminstrative domain.

         There are certainly suggestions, but no rules about the
         form of the content -- at least as I understand it.  So
         we can't depend on *any* specific content in the
         client-identifier.

         And, in fact, I've seen client-identifiers which contain
         all sorts of stuff, strings of various types, etc. etc.

         Cheers -- Kim


>- Bernie Volz
>   Process Software



From dhcp-v4@bucknell.edu  Wed Nov  3 10:56:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA04733
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 10:56:56 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA11411;
	Wed, 3 Nov 1999 10:56:20 -0500 (EST)
Date: Wed, 3 Nov 1999 10:56:20 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.941643507.21295.mwc@atlantic.east.sun.com.>
Errors-To: droms@bucknell.edu
Reply-To: Michael.Carney@East.Sun.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


Hi Bernie,

> Seems like most people feel the type is included as part of the client
> identifier ... 

No, I don't see this at all. The client identifier is an opaque string,
period. You can't make any assumptions about the contents.

> 
> In the DHCP LDAP Schema draft we've "collapsed" mac-address and
> client-identifier into one attribute (dhcpclientidentifier). The idea
> was that since the client identifier should be prefixed by 0, if it
> was not a hardware address, as specified in RFC2132, we could simply
> combine these fields. That way, if you wanted to identify a client
> by mac address, you could be using <media-type>:<mac-address>.

I don't think collapsing these different attributes into one makes sense. Some
folks use a client-identifier which happens to be an asset tag number. In this
case, the client-identifier has no relationship at all to the mac address.
IMHO, I think you should revisit this decision.

> 
> So, the question is ... does everyone (client implementation) follow the
> rules - specifying t1 as 0 if it is not a mac address.

What rules?

> 
> - Bernie Volz
>   Process Software


Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v4@bucknell.edu  Wed Nov  3 12:44:47 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA17343
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 12:44:46 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA28364;
	Wed, 3 Nov 1999 12:43:24 -0500 (EST)
Date: Wed, 3 Nov 1999 12:43:24 -0500 (EST)
Message-Id: <009E0973.981F98C3.225@process.com>
Errors-To: droms@bucknell.edu
Reply-To: volz@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Bernie Volz <volz@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4

Mike:

>> So, the question is ... does everyone (client implementation) follow the
>> rules - specifying t1 as 0 if it is not a mac address.
>
>What rules?
>
>> - Bernie Volz
>>   Process Software
>
>Mike Carney

>From RFC 2132, see ">>" lines. The text does use MAY, SHOULD, and
should. So, these are definitely not MUSTs.

9.14. Client-identifier

   This option is used by DHCP clients to specify their unique
   identifier.  DHCP servers use this value to index their database of
   address bindings.  This value is expected to be unique for all
   clients in an administrative domain.

   Identifiers SHOULD be treated as opaque objects by DHCP servers.

   The client identifier MAY consist of type-value pairs similar to the
   'htype'/'chaddr' fields defined in [3]. For instance, it MAY consist
   of a hardware type and hardware address. In this case the type field
>>   SHOULD be one of the ARP hardware types defined in STD2 [22].  A
>>   hardware type of 0 (zero) should be used when the value field
>>   contains an identifier other than a hardware address (e.g. a fully
>>   qualified domain name).

   For correct identification of clients, each client's client-
   identifier MUST be unique among the client-identifiers used on the
   subnet to which the client is attached.  Vendors and system
   administrators are responsible for choosing client-identifiers that
   meet this requirement for uniqueness.

   The code for this option is 61, and its minimum length is 2.

   Code   Len   Type  Client-Identifier
   +-----+-----+-----+-----+-----+---
   |  61 |  n  |  t1 |  i1 |  i2 | ...
   +-----+-----+-----+-----+-----+---


From dhcp-v4@bucknell.edu  Wed Nov  3 12:47:52 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA18557
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 12:47:49 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA17919;
	Wed, 3 Nov 1999 12:47:08 -0500 (EST)
Date: Wed, 3 Nov 1999 12:47:08 -0500 (EST)
Message-Id: <47FEF5D2BE22D311AA8600508B10891532BD7B@fs-sd-exch1.coppermountain.com>
Errors-To: droms@bucknell.edu
Reply-To: emichelsen@coppermountain.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Eric Michelsen <emichelsen@coppermountain.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

> -----Original Message-----
> From: Mike Carney - SNT Internet Engineering
> [mailto:Michael.Carney@East.Sun.COM]
> Sent: Wednesday, November 03, 1999 7:57
> To: Multiple recipients of list
> Subject: RE: Possible ambiguity in definition of DHCP client 
> identifier
> op
> 
> 
> 
> Hi Bernie,
> 
> > Seems like most people feel the type is included as part of 
> the client
> > identifier ... 
> 
> No, I don't see this at all. The client identifier is an 
> opaque string,
> period. You can't make any assumptions about the contents.

It *is* an opaque string.  And the string *includes* the "type" byte, which
means there is no distinction between the "type" byte, and any other byte.
The presence of a "type" byte is very confusing, because it suggests some
distinction when in fact there is none.  If the string is indeed "opaque,"
then all distinctions between bytes should be removed.


-------------------------------------------
Eric L. Michelsen, Copper Mountain Networks


From dhcp-v4@bucknell.edu  Wed Nov  3 12:49:46 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA19506
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 12:49:45 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA24224;
	Wed, 3 Nov 1999 12:49:07 -0500 (EST)
Date: Wed, 3 Nov 1999 12:49:07 -0500 (EST)
Message-Id: <EA5107D3A9ADD211B7530000C041E4F4011B2AD4@honts318.wal-mart.com>
Errors-To: droms@bucknell.edu
Reply-To: ndlane@wal-mart.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Nathan Lane <ndlane@wal-mart.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0



> -----Original Message-----
> From:	Mike Carney - SNT Internet Engineering [SMTP:Michael.Carney@East.Sun.COM]
> Sent:	Wednesday, November 03, 1999 9:57 AM
> To:	Multiple recipients of list
> Subject:	RE: Possible ambiguity in definition of DHCP client identifier op
> 
> 
> Hi Bernie,
> 
> > Seems like most people feel the type is included as part of the client
> > identifier ... 
> 
> No, I don't see this at all. The client identifier is an opaque string,
> period. You can't make any assumptions about the contents.
> 
	[Nathan Lane]  *I* can as an implementor of DHCP network services.  Section 9.14 of 2132 specifies "MAY" consist of type value pairs similar to htype/chaddr.  In the case where it is defined as type/value pairs, a data format is suggested.  My view is that the CID should be given to me via an externalized process or ruleset determined by me within the server.  I agree that the actual DHCP server itself should not assign any cosmic significance to any part of the CID save for the uniqueness within the subnet (perhaps we should change the word "subnet" to "administrative domain of the DHCP server" in 2132), but the server should be able to give the CID to my business rule that is abstracted from the server.  Cisco's product does this well with the TCL or C extensions that are possible with the product.

> I don't think collapsing these different attributes into one makes sense. Some
> folks use a client-identifier which happens to be an asset tag number. In this
> case, the client-identifier has no relationship at all to the mac address.
> IMHO, I think you should revisit this decision.
> 
	[Nathan Lane]  An asset tag number or ToS definitions are the path we are going down until user-class is widely implemented.

> > 
> > So, the question is ... does everyone (client implementation) follow the
> > rules - specifying t1 as 0 if it is not a mac address.
> 
> What rules?
> 
	[Nathan Lane]  Again, doesn't section 9.14 specify a suggestion for a rule?  Am I missing something here?

> > 
> > - Bernie Volz
> >   Process Software
> 
> 
> Mike Carney
> SNT Internet Engineering
> Sun Microsystems, Inc.
> Mailstop UBUR02-212
> One Network Drive
> Burlington, MA 01803
> 
	-Nathan Lane
	Wal-Mart Stores, Inc.
	(501) 277-5786



From dhcp-v4@bucknell.edu  Wed Nov  3 12:51:53 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA20509
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 12:51:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA29445;
	Wed, 3 Nov 1999 12:51:03 -0500 (EST)
Date: Wed, 3 Nov 1999 12:51:03 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE28@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


Ted, et al.--

I am in agreement with Kim and you about the "opaque-ness" of the client
identifier, and I think the text could be revised to be more explicit.


> BTW, I don't know if it's possible at this late date, but I'd really
> like to see the text "the client MAY choose to explicitly provide the
> identifier through the 'client identifier' option" to "the client
> SHOULD choose to explicitly provide the identifier through the 'client
> identifier' option".   Right now I think client implementors tend to
> read this as saying that the client identifier is optional (that's
> certainly how I read it) and may not feel any real motivation to use a
> client identifier.   To say the least, the way client identifiers are
> currently handled is quite a bummer.   If I had my druthers, even
> though the ISC DHCP client doesn't currently implement it, I'd change
> this to a MUST, with a caveat that servers should allow for clients
> that don't supply client identifiers for backward compatibility.
> 
--I'd go a lot further than that!  Not only would I change the wording from
MAY to SHOULD, but I would add text strongly suggesting that whatever value
is chosen for the client identifier, it SHOULD NOT be something that can
easily change, such as a MAC address:  in my experience, one or two NICs
must be replaced every week in our client population, not to mention the
near-constant movement of roaming laptop users, whose MAC address changes
every time they change docking stations!

--Barr



From dhcp-v4@bucknell.edu  Wed Nov  3 12:54:26 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA21613
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 12:54:25 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA28832;
	Wed, 3 Nov 1999 12:53:09 -0500 (EST)
Date: Wed, 3 Nov 1999 12:53:09 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE27@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


Bernie, et al.--

we have a genuine difference of opinion about the composition of client
identifier.

I assert that the entire data payload of option 61 MUST be treated as an
opaque string, so the SUGGESTED encoding of <type><some-data> is erroneous:
I have lots of working examples of clients that send our server client
identifiers such as "63IDDZ100679" and "Mike's PC" just as I have examples
such as 01:00:DD:01:00:FC:DE.  Note that both of the former are "missing"
the type octet, while the latter does carry it.  Both types work "just fine"
because our server treats the entire data payload as an opaque string --
only humans read an encoding scheme into the data.

The drawing in RFC 2131 is unfortunate, because it gives the impression that
option 61 deviates from the general option format of <code><length><data>.
Think about what happens if you apply the interpretation of a distinct type
octet literally:

(1) when the type octet is an ARP type (such as Ethernet or Token-Ring) the
client identifier should be identically equal to the chaddr field.  If that
is truly the case, why do we have a client identifier?  It is completely
redundant in that case!  Worse, if the type octet matches an ARP type,
shouldn't that also imply the length of the remaining data?  What happens if
there is a mismatch?  Does that invalidate the following octets?

(2) if the type octet were 0, implying the remainder of the data is an
NVT-ASCII string, then that seems okay, but what value does that extra octet
offer?  The server MUST treat the client identifier as an opaque string, so
the encoding is pointless.

(3) what do we do with the following cases?
-- a character string encoded as UTF-8
-- a binary string that is not a hardware address

My assertion is that it is not only better, but essential to treat the
entire data payload of option 61 as an uninterpretable (i.e., opaque) value.

What this may do to you LDAP schema is another problem entirely.

--Barr


> Seems like most people feel the type is included as part of the client
> identifier ... 
> 
> In the DHCP LDAP Schema draft we've "collapsed" mac-address and
> client-identifier into one attribute (dhcpclientidentifier). The idea
> was that since the client identifier should be prefixed by 0, if it
> was not a hardware address, as specified in RFC2132, we could simply
> combine these fields. That way, if you wanted to identify a client
> by mac address, you could be using <media-type>:<mac-address>.
> 
> So, the question is ... does everyone (client implementation) follow the
> rules - specifying t1 as 0 if it is not a mac address.
> 
> 


From dhcp-v4@bucknell.edu  Wed Nov  3 12:55:56 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA22254
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 12:55:55 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA02469;
	Wed, 3 Nov 1999 12:55:11 -0500 (EST)
Date: Wed, 3 Nov 1999 12:55:11 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE29@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> > So, the question is ... does everyone (client implementation) follow the
> > rules - specifying t1 as 0 if it is not a mac address.
> 
> Absolutely not, nor does the protocol require it.  Further, it's not
> specified that cid==1:4a:2b:ac:19:ad means the same thing as
> htype==1/hlen=6/chaddr==4a:2b:ac:19:ad:dc:dc:dc:dc:dc:dc:dc:dc:dc:dc
> (dc == don't care).
> 
> If you want to collapse the lease database lookup key into a single
> key value, which I think is reasonable, you really need to have some
> kind of additional type that says where the key came from.   I would
> suggest zero for htype/hlen/chaddr and 61 for dhcp-client-identifier.
> Otherwise I think that your specification for the construction of a
> key based on htype/hlen/chaddr is correct.
> 
I agree with the need to identify the source of the key, but would suggest
that there are more than two possible values:  in my case, we have local
modifications that permit lookup of leases by hostname (soon to be changed
to FQDN, I think), so I think there are at least three values for the key
source.

--Barr




From dhcp-v4@bucknell.edu  Wed Nov  3 13:56:52 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA21340
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 13:56:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA07983;
	Wed, 3 Nov 1999 13:55:30 -0500 (EST)
Date: Wed, 3 Nov 1999 13:55:30 -0500 (EST)
Message-Id: <199911031748.MAA01669@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier op 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


That's not a rule - that's a suggestion.   The SHOULD is predicated on
a MAY, so it's only as strong as the MAY.   Personally, I've always
been allergic to the idea that the client-identifier's structure
should be documented at all, but I've been called crazy before... :')

			       _MelloN_


From dhcp-v4@bucknell.edu  Wed Nov  3 13:59:37 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA22707
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 13:59:37 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA01817;
	Wed, 3 Nov 1999 13:59:10 -0500 (EST)
Date: Wed, 3 Nov 1999 13:59:10 -0500 (EST)
Message-Id: <199911031805.NAA01831@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier op 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> I agree with the need to identify the source of the key, but would suggest
> that there are more than two possible values:  in my case, we have local
> modifications that permit lookup of leases by hostname (soon to be changed
> to FQDN, I think), so I think there are at least three values for the key
> source.

This is pretty much what I was thinking when I suggested a code of
zero for chaddr and 61 for dhcp-client-identifier.  If you want to
(incorrectly! :') use something else as a client identifier, you could
stick its option code in the first byte of the key.  I'd suggest an
option code of 255 for a key that's neither the hardware address nor
a DHCP option.

Well, actually, I'd suggest not specifying any values other than 61
and 0, and leaving the other values for people to use, actually,
because if you put what I wrote in the previous paragraph into the
LDAP RFC, somebody's going to start claiming that values other than
chaddr and dhcp-client-identifier are blessed as client identifiers in
an RFC.

Actually^2, I'd make the identifier two bytes, to account for the
potentially upcoming transition to 2-byte option codes in DHCPv4NG.

Another advantage of this key :== identifier + data scheme is that it
lets us implement UUID support cleanly when/if it becomes a standard.

			       _MelloN_


From dhcp-v4@bucknell.edu  Wed Nov  3 14:02:07 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA23831
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 14:02:06 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA15108;
	Wed, 3 Nov 1999 14:01:14 -0500 (EST)
Date: Wed, 3 Nov 1999 14:01:14 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.941652457.11827.mwc@atlantic.east.sun.com.>
Errors-To: droms@bucknell.edu
Reply-To: Michael.Carney@east.sun.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@east.sun.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


>  Identifiers SHOULD be treated as opaque objects by DHCP servers.
>  
>    The client identifier MAY consist of type-value pairs similar to the
>    'htype'/'chaddr' fields defined in [3].

9.14 is really confusing. That text was added later in the draft. The
combinations of SHOULD's MAY's, etc.... Geez.



Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v4@bucknell.edu  Wed Nov  3 14:04:03 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA24677
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 14:04:02 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA07388;
	Wed, 3 Nov 1999 14:03:08 -0500 (EST)
Date: Wed, 3 Nov 1999 14:03:08 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.941652637.20993.mwc@atlantic.east.sun.com.>
Errors-To: droms@bucknell.edu
Reply-To: Michael.Carney@East.Sun.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

Hi Nathan,


> > No, I don't see this at all. The client identifier is an opaque string,
> > period. You can't make any assumptions about the contents.
> > 
> 	[Nathan Lane]  *I* can as an implementor of DHCP network services.  Section
> 9.14 of 2132 specifies "MAY" consist of type value pairs similar to
> htype/chaddr.  In the case where it is defined as type/value pairs, a data
> format is suggested.  My view is that the CID should be given to me via an
> externalized process or ruleset determined by me within the server.  I agree
> that the actual DHCP server itself should not assign any cosmic significance
> to any part of the CID save for the uniqueness within the subnet (perhaps we
> should change the word "subnet" to "administrative domain of the DHCP
> server" in 2132), but the server should be able to give the CID to my
> business rule that is abstracted from the server.  Cisco's product does this
> well with the TCL or C extensions that are possible with the product.

Sorry - by "You can't" I was speaking from a DHCP server's perspective.
Whether a user / admin choses through administrative/site policy to assign
some special significance to the bits in the client identifier used is fine.
The point I was trying to make was that the clent identifier is opaque from a
protocol perspective.


> > 
> > What rules?
> > 
> 	[Nathan Lane]  Again, doesn't section 9.14 specify a suggestion for a rule? 
> Am I missing something here?

It's a suggestion for forming a client identifier - not a "rule"



Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v4@bucknell.edu  Wed Nov  3 14:05:54 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA25529
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 14:05:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA11105;
	Wed, 3 Nov 1999 14:05:19 -0500 (EST)
Date: Wed, 3 Nov 1999 14:05:19 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE2B@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


Ted, et al.--

as always, you have cogent comments, but "actually" squared?  hehehe....


> > I agree with the need to identify the source of the key, but would
> suggest
> > that there are more than two possible values:  in my case, we have local
> > modifications that permit lookup of leases by hostname (soon to be
> changed
> > to FQDN, I think), so I think there are at least three values for the
> key
> > source.
> 
> This is pretty much what I was thinking when I suggested a code of
> zero for chaddr and 61 for dhcp-client-identifier.  If you want to
> (incorrectly! :') use something else as a client identifier, you could
> stick its option code in the first byte of the key.  I'd suggest an
> option code of 255 for a key that's neither the hardware address nor
> a DHCP option.
> 
> Well, actually, I'd suggest not specifying any values other than 61
> and 0, and leaving the other values for people to use, actually,
> because if you put what I wrote in the previous paragraph into the
> LDAP RFC, somebody's going to start claiming that values other than
> chaddr and dhcp-client-identifier are blessed as client identifiers in
> an RFC.
> 
> Actually^2, I'd make the identifier two bytes, to account for the
> potentially upcoming transition to 2-byte option codes in DHCPv4NG.
> 
> Another advantage of this key :== identifier + data scheme is that itI
> must not have clearly worded my comments....  I am NOT suggesting that we
> somehow try to fit lots of other vlues/encodings/fields into
> client-identifer, but rather that if an LDAP schema proposes that an index
> into a table of leases be coded as to the source of the index value
> (htype/chaddr and client-identifer as the initial sources proposed) then
> there are probably more choices, depending on local implementations.
> 
> This is not the same thing as saying that some other value is to be used
> by the server to identify a lease for assignment purposes -- I'm not
> proposing that at all.  For historical reasons and corporate culture, I
> must provide a way to find leases based on hostname, but all I offer is a
> display of lease information -- I don't substitute option 12, or a
> combination of options 12 and 15, for either the client hardware address
> or client identifier for address assignment, and I'm not proposing any
> changes to RFC 2131.  I'm simply observing that there are other ways in
> which organizations and individuals might like to use to find a lease, so
> it seems appropriate to provide a means to identify more than two sources
> for a key.
> 
> 
> lets us implement UUID support cleanly when/if it becomes a standard.
> 
Actually, the Intel-originated RFC for UUID is current, although it has a
sunset date of about 2003, so maybe option 97 is already a valid source
indicator.

--Barr



From dhcp-v4@bucknell.edu  Wed Nov  3 14:07:44 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA26418
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 14:07:44 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA06002;
	Wed, 3 Nov 1999 14:07:13 -0500 (EST)
Date: Wed, 3 Nov 1999 14:07:13 -0500 (EST)
Message-Id: <199911031847.NAA02069@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier op 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> as always, you have cogent comments, but "actually" squared?  hehehe....

Look, if I don't break the language, somebody else is going to have
to!

> Actually, the Intel-originated RFC for UUID is current, although it has a
> sunset date of about 2003, so maybe option 97 is already a valid source
> indicator.

Yes, but it needs to be made officially part of the protocol (my vote
would be in DHCPv4NG).  Either that, or in DHCPv4NG we should make the
client identifier and UUID the same (I would actually prefer this, but
in some cases a UUID may be difficult to come up with - e.g., in a
kernel that's using DHCP to get an IP address before it's mounted a
root filesystem).

			       _MelloN_


From dhcp-v4@bucknell.edu  Wed Nov  3 15:11:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA29597
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 15:11:37 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA09408;
	Wed, 3 Nov 1999 15:09:52 -0500 (EST)
Date: Wed, 3 Nov 1999 15:09:52 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE2D@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> > Actually, the Intel-originated RFC for UUID is current, although it has
> a
> > sunset date of about 2003, so maybe option 97 is already a valid source
> > indicator.
> 
> Yes, but it needs to be made officially part of the protocol (my vote
> would be in DHCPv4NG).  Either that, or in DHCPv4NG we should make the
> client identifier and UUID the same (I would actually prefer this, but
> in some cases a UUID may be difficult to come up with - e.g., in a
> kernel that's using DHCP to get an IP address before it's mounted a
> root filesystem).
> 
> 
Ted, et al.--

the latter bit, about a kernel using DHCP, is precisely what Intel's PXE
specification is about.

I, too, would like a globally unique identifier, but I wonder if we would
break existing behavior in an unfriendly way to replace the current
requirements of client identifier.

You've convinced me that I must take the time to write a new I-D proposing
some changes to the client identifier, which I should be able to distribute
before the WG meeting.

--Barr



From dhcp-v4@bucknell.edu  Wed Nov  3 15:14:59 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA01251
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 15:14:58 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA27424;
	Wed, 3 Nov 1999 15:14:03 -0500 (EST)
Date: Wed, 3 Nov 1999 15:14:03 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE2E@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> That's not a rule - that's a suggestion.   The SHOULD is predicated on
> a MAY, so it's only as strong as the MAY.   Personally, I've always
> been allergic to the idea that the client-identifier's structure
> should be documented at all, but I've been called crazy before... :')
> 
agreed: by documenting one possible way to construct a client identifier,
we've led lots of people astray....

--Barr



From dhcp-v4@bucknell.edu  Wed Nov  3 22:38:42 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA22585
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 22:38:41 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA05219;
	Wed, 3 Nov 1999 22:37:13 -0500 (EST)
Date: Wed, 3 Nov 1999 22:37:13 -0500 (EST)
Message-Id: <199911040328.TAA17000@scv3.apple.com>
Errors-To: droms@bucknell.edu
Reply-To: cheshire@apple.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Stuart Cheshire <cheshire@apple.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="US-ASCII"
Mime-Version: 1.0
X-Mailer: Claris Emailer 2.0v3, January 22, 1998

Greetings,

The engineer at Apple who used to work on our DHCP client moved to a 
different company recently, and, for now, that responsibility has fallen 
to me. I've subscribed to this mailing list and I hope in future to keep 
Apple more in touch with what's going on with new DHCP work.

For now, I have a simple question. Mac OS does not currently put a vendor 
class identifier option in its DHCP packets. It seems to me that this 
would be a useful thing to do, both for the benefit of network 
administators, and for our benefit, so that when people complain that 
"DHCP doesn't work" and send us packet traces, we can tell what version 
of Open Transport (Apple's networking stack) was being used. From my 
reading RFC 2132 I believe that, like the client identifier option, the 
vendor class is an opaque string. I'm considering adding a vendor class 
identifier of the following form to all DHCP packets we send:

-------------------------------------------------------------
|60|27|M|a|c| |O|S| |O|p|e|n| |T|r|a|n|s|p|o|r|t| |2|.|5|.|3|
-------------------------------------------------------------

(i.e. option code 60, length 27, followed by 27 bytes of ascii text)

Is this acceptable, or have I misinterpreted RFC 2132?

>9.13. Vendor class identifier
>
>   This option is used by DHCP clients to optionally identify the vendor
>   type and configuration of a DHCP client.  The information is a string
>   of n octets, interpreted by servers.  Vendors may choose to define
>   specific vendor class identifiers to convey particular configuration
>   or other identification information about a client.  For example, the
>   identifier may encode the client's hardware configuration.  Servers
>   not equipped to interpret the class-specific information sent by a
>   client MUST ignore it (although it may be reported). Servers that
>   respond SHOULD only use option 43 to return the vendor-specific
>   information to the client.
>
>   The code for this option is 60, and its minimum length is 1.
>
>   Code   Len   Vendor class Identifier
>   +-----+-----+-----+-----+---
>   |  60 |  n  |  i1 |  i2 | ...
>   +-----+-----+-----+-----+---


Stuart Cheshire <cheshire@apple.com>
 * <A HREF="http://ResComp.Stanford.EDU/~cheshire/">Web Page</A>
 * Wizard Without Portfolio, Apple Computer



From dhcp-v4@bucknell.edu  Wed Nov  3 23:13:55 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA07889
	for <dhc-archive@odin.ietf.org>; Wed, 3 Nov 1999 23:13:55 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA21009;
	Wed, 3 Nov 1999 23:13:33 -0500 (EST)
Date: Wed, 3 Nov 1999 23:13:33 -0500 (EST)
Message-Id: <382100491F.B2EAJURGEN@smtp.rad.net.id>
Errors-To: droms@bucknell.edu
Reply-To: jurgen@wwdg.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Jurgen <jurgen@wwdg.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: Becky! ver 1.25.04
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=US-ASCII
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

unsubscribe dhcp-v4




From dhcp-v4@bucknell.edu  Thu Nov  4 02:32:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id CAA15616
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 02:32:56 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id CAA28362;
	Thu, 4 Nov 1999 02:32:34 -0500 (EST)
Date: Thu, 4 Nov 1999 02:32:34 -0500 (EST)
Message-Id: <H00000af053d4b01@MHS>
Errors-To: droms@bucknell.edu
Reply-To: W.Leiter@alcatel.de
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: W.Leiter@alcatel.de
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: WG:
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: multipart/mixed; boundary="openmail-part-0a55e49a-00000001"
Mime-Version: 1.0


--openmail-part-0a55e49a-00000001
Content-Type: text/plain; charset=US-ASCII; name="BDY.TXT"
Content-Disposition: inline; filename="BDY.TXT"
Content-Transfer-Encoding: 7bit

unsubscribe dhcp-v4



--openmail-part-0a55e49a-00000001--



From dhcp-v4@bucknell.edu  Thu Nov  4 05:41:15 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id FAA17455
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 05:41:15 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id FAA29980;
	Thu, 4 Nov 1999 05:40:17 -0500 (EST)
Date: Thu, 4 Nov 1999 05:40:17 -0500 (EST)
Message-Id: <00e501bf26ac$d1737b60$969d9fc2@intranetsolutions.com>
Errors-To: droms@bucknell.edu
Reply-To: paul.roberts@intranet-solutions.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Paul Roberts" <paul.roberts@intranet-solutions.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: Microsoft Outlook Express 5.00.2014.211
Content-Transfer-Encoding: 7bit
Content-Type: text/plain;
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

Stuart,

I am glad you have taken the opportunity to announce your presence on
this list. My company has implemented DHCP at many organisations
throughout the UK and my experience has taught me that the Apple Open
Transport DHCP client has typically been the "poor relation" as far as
DHCP clients are concerned. I must admit, I am not up to date on the
features and functionality of the latest versions of the client and I
was wondering if you could tell me whether the client now has the
ability to send the client hostname in option 12 of the DHCP packet,
or whether this is an enhancement planned for the future?

The reason I ask this is that we do lots of work with DDNS too, and it
has always been difficult integrating Apple clients into a DHCP and
DDNS environment when the Apple clients do not supply a host name to
the DHCP server.

Regards,

Paul Roberts

+---------------------------------------------+
| Paul Roberts, Intranet & Intranet Solutions |
|  Tel: +44 1256 346344 Fax: +44 1256 346355  |
| E-mail: paul.roberts@intranet-solutions.com |
|   WWW: http://www.intranet-solutions.com    |
+---------------------------------------------+



----- Original Message -----
From: Stuart Cheshire <cheshire@apple.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Sent: 04 November 1999 03:39
Subject: DHCP Vendor Class question


> Greetings,
>
> The engineer at Apple who used to work on our DHCP client moved to a
> different company recently, and, for now, that responsibility has
fallen
> to me. I've subscribed to this mailing list and I hope in future to
keep
> Apple more in touch with what's going on with new DHCP work.
>
> For now, I have a simple question. Mac OS does not currently put a
vendor
> class identifier option in its DHCP packets. It seems to me that
this
> would be a useful thing to do, both for the benefit of network
> administators, and for our benefit, so that when people complain
that
> "DHCP doesn't work" and send us packet traces, we can tell what
version
> of Open Transport (Apple's networking stack) was being used. From my
> reading RFC 2132 I believe that, like the client identifier option,
the
> vendor class is an opaque string. I'm considering adding a vendor
class
> identifier of the following form to all DHCP packets we send:
>
> -------------------------------------------------------------
> |60|27|M|a|c| |O|S| |O|p|e|n| |T|r|a|n|s|p|o|r|t| |2|.|5|.|3|
> -------------------------------------------------------------
>
> (i.e. option code 60, length 27, followed by 27 bytes of ascii text)
>
> Is this acceptable, or have I misinterpreted RFC 2132?
>
> >9.13. Vendor class identifier
> >
> >   This option is used by DHCP clients to optionally identify the
vendor
> >   type and configuration of a DHCP client.  The information is a
string
> >   of n octets, interpreted by servers.  Vendors may choose to
define
> >   specific vendor class identifiers to convey particular
configuration
> >   or other identification information about a client.  For
example, the
> >   identifier may encode the client's hardware configuration.
Servers
> >   not equipped to interpret the class-specific information sent by
a
> >   client MUST ignore it (although it may be reported). Servers
that
> >   respond SHOULD only use option 43 to return the vendor-specific
> >   information to the client.
> >
> >   The code for this option is 60, and its minimum length is 1.
> >
> >   Code   Len   Vendor class Identifier
> >   +-----+-----+-----+-----+---
> >   |  60 |  n  |  i1 |  i2 | ...
> >   +-----+-----+-----+-----+---
>
>
> Stuart Cheshire <cheshire@apple.com>
>  * <A HREF="http://ResComp.Stanford.EDU/~cheshire/">Web Page</A>
>  * Wizard Without Portfolio, Apple Computer
>



From dhcp-v4@bucknell.edu  Thu Nov  4 08:56:34 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id IAA23130
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 08:56:34 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA06107;
	Thu, 4 Nov 1999 08:55:56 -0500 (EST)
Date: Thu, 4 Nov 1999 08:55:56 -0500 (EST)
Message-Id: <009E0A28.E4146C42.165@process.com>
Errors-To: droms@bucknell.edu
Reply-To: volz@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Bernie Volz <volz@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4

>In the DHCP LDAP Schema draft we've "collapsed" mac-address and
>client-identifier into one attribute (dhcpclientidentifier). The idea
>was that since the client identifier should be prefixed by 0, if it
>was not a hardware address, as specified in RFC2132, we could simply
>combine these fields. That way, if you wanted to identify a client
>by mac address, you could be using <media-type>:<mac-address>.

Actually, Andy corrected me on this item. The text in the schema draft
was intended to mean that for a client identifier, when it is stored you
prefix it by 00:, just as for a MAC address, you prefix it by the
hardware type field. The client identifier is treated as opaque (not
assumed to have a type field); hence one is added when stored, and
removed when processed. We'll clarify this in the next revision.

- Bernie Volz
  Process Software


From dhcp-v4@bucknell.edu  Thu Nov  4 11:14:15 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA11554
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 11:14:14 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA18961;
	Thu, 4 Nov 1999 11:12:41 -0500 (EST)
Date: Thu, 4 Nov 1999 11:12:41 -0500 (EST)
Message-Id: <199911041455.JAA00604@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier op 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> Actually, Andy corrected me on this item. The text in the schema draft
> was intended to mean that for a client identifier, when it is stored you
> prefix it by 00:, just as for a MAC address, you prefix it by the
> hardware type field. The client identifier is treated as opaque (not
> assumed to have a type field); hence one is added when stored, and
> removed when processed. We'll clarify this in the next revision.

Hm.   Okay, but this has the same problem: you're confusing ARP types
and non-ARP types in the same identifier.   Why not have an identifier
that says where the key came from, followed by the contents of the
key, which for a client-identifier option would be the opaque data,
and for the chaddr would be the ARP type, followed by bytes zero
through hlen of the DHCP chaddr field?   This would allow us to
unambiguously represent UUID (or any nonstandard key that the server
might decide to use) when that becomes necessary.

			       _MelloN_


From dhcp-v4@bucknell.edu  Thu Nov  4 11:19:08 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA13765
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 11:19:07 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA25511;
	Thu, 4 Nov 1999 11:17:59 -0500 (EST)
Date: Thu, 4 Nov 1999 11:17:59 -0500 (EST)
Message-Id: <19991104160125.19296.qmail@hotmail.com>
Errors-To: droms@bucknell.edu
Reply-To: pauwen@hotmail.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Georg Pauwen" <pauwen@hotmail.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; format=flowed
Mime-Version: 1.0

unsubscribe dhcp-v4

______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com


From dhcp-v4@bucknell.edu  Thu Nov  4 12:20:40 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA08323
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 12:20:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA31487;
	Thu, 4 Nov 1999 12:19:03 -0500 (EST)
Date: Thu, 4 Nov 1999 12:19:03 -0500 (EST)
Message-Id: <63D30D6E10CFD11190A90000F805FE8601CE6FB3@lespaul.process.com>
Errors-To: droms@bucknell.edu
Reply-To: Gonczi@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Steve Gonczi <Gonczi@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.0.1460.8)
Mime-Version: 1.0

If I am reading this right, it means that we always prefix a client
identifier
with 00: when stored as an LDAP attribute.

This has nothing to do with prior dicussion of the construction rules of the
payload of 
the DHCP client ID option. 

A client may very well choose to construct the payload in a similar fashion.
We add the 
prefix 00: _when_stored_as_an_LDAP_attribute.

sG
-----Original Message-----
From: Bernie Volz [mailto:volz@process.com]
Sent: Thursday, November 04, 1999 8:57 AM
To: Multiple recipients of list
Subject: RE: Possible ambiguity in definition of DHCP client identifier
op


>In the DHCP LDAP Schema draft we've "collapsed" mac-address and
>client-identifier into one attribute (dhcpclientidentifier). The idea
>was that since the client identifier should be prefixed by 0, if it
>was not a hardware address, as specified in RFC2132, we could simply
>combine these fields. That way, if you wanted to identify a client
>by mac address, you could be using <media-type>:<mac-address>.

Actually, Andy corrected me on this item. The text in the schema draft
was intended to mean that for a client identifier, when it is stored you
prefix it by 00:, just as for a MAC address, you prefix it by the
hardware type field. The client identifier is treated as opaque (not
assumed to have a type field); hence one is added when stored, and
removed when processed. We'll clarify this in the next revision.

- Bernie Volz
  Process Software


From dhcp-v4@bucknell.edu  Thu Nov  4 12:21:51 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA08928
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 12:21:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA04751;
	Thu, 4 Nov 1999 12:21:09 -0500 (EST)
Date: Thu, 4 Nov 1999 12:21:09 -0500 (EST)
Message-Id: <63D30D6E10CFD11190A90000F805FE8601CE6FB4@lespaul.process.com>
Errors-To: droms@bucknell.edu
Reply-To: Gonczi@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Steve Gonczi <Gonczi@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.0.1460.8)
Mime-Version: 1.0

The LDAP representation is _Always_ prefixed by an appropriate
classification byte, indicating the origin.
This is part of the LDAP schema rules, hence it is done consistently by the
storing agent.

Arguably, it would be neater to use a separate argument to classify the ID,
or, at the very least use a 2 byte prefix. 

sG

-----Original Message-----
From: Ted Lemon [mailto:mellon@isc.org]
Sent: Thursday, November 04, 1999 11:16 AM
To: Multiple recipients of list
Subject: Re: Possible ambiguity in definition of DHCP client identifier
op



> Actually, Andy corrected me on this item. The text in the schema draft
> was intended to mean that for a client identifier, when it is stored you
> prefix it by 00:, just as for a MAC address, you prefix it by the
> hardware type field. The client identifier is treated as opaque (not
> assumed to have a type field); hence one is added when stored, and
> removed when processed. We'll clarify this in the next revision.

Hm.   Okay, but this has the same problem: you're confusing ARP types
and non-ARP types in the same identifier.   Why not have an identifier
that says where the key came from, followed by the contents of the
key, which for a client-identifier option would be the opaque data,
and for the chaddr would be the ARP type, followed by bytes zero
through hlen of the DHCP chaddr field?   This would allow us to
unambiguously represent UUID (or any nonstandard key that the server
might decide to use) when that becomes necessary.

			       _MelloN_


From dhcp-v4@bucknell.edu  Thu Nov  4 12:22:26 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA09253
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 12:22:26 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA05345;
	Thu, 4 Nov 1999 12:21:46 -0500 (EST)
Date: Thu, 4 Nov 1999 12:21:46 -0500 (EST)
Message-Id: <199911041632.LAA00850@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier op 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> Arguably, it would be neater to use a separate argument to classify the ID,
> or, at the very least use a 2 byte prefix. 

Works for me, as long as we retain a requirement that there be exactly
one key.

			       _MelloN_


From dhcp-v4@bucknell.edu  Thu Nov  4 13:10:33 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA27671
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 13:10:33 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA19503;
	Thu, 4 Nov 1999 13:09:25 -0500 (EST)
Date: Thu, 4 Nov 1999 13:09:25 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE3C@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


I agree with Ted's suggestion regarding identifying the source of the key...

--Barr


> Why not have an identifier that says where the key came from, followed by
> the contents of the key, which for a client-identifier option would be the
> opaque data, and for the chaddr would be the ARP type, followed by bytes
> zero through hlen of the DHCP chaddr field?   This would allow us to
> unambiguously represent UUID (or any nonstandard key that the server might
> decide to use) when that becomes necessary.
> 
> 


From dhcp-v4@bucknell.edu  Thu Nov  4 13:12:00 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA28307
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 13:12:00 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA07207;
	Thu, 4 Nov 1999 13:11:23 -0500 (EST)
Date: Thu, 4 Nov 1999 13:11:23 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE3D@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> > Arguably, it would be neater to use a separate argument to classify the
> ID,
> > or, at the very least use a 2 byte prefix. 
> 
> Works for me, as long as we retain a requirement that there be exactly
> one key.
> 
..works for me as well...  with the additional suggestion that the
classification argument ought to be obvious, such as 61 for client
identifier or 97 for UUID...

--Barr



From dhcp-v4@bucknell.edu  Thu Nov  4 13:12:52 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA28736
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 13:12:52 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA10343;
	Thu, 4 Nov 1999 13:12:00 -0500 (EST)
Date: Thu, 4 Nov 1999 13:12:00 -0500 (EST)
Message-Id: <61C1908C75C7D211B03B00C04F6B74DF1B8B02@APU>
Errors-To: droms@bucknell.edu
Reply-To: CMatthes@tmrs.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Matthes, Charles" <CMatthes@tmrs.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0

unsubscribe dhcp-v4



From dhcp-v4@bucknell.edu  Thu Nov  4 13:13:52 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA29181
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 13:13:51 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA22560;
	Thu, 4 Nov 1999 13:12:42 -0500 (EST)
Date: Thu, 4 Nov 1999 13:12:42 -0500 (EST)
Message-Id: <199911041745.MAA01111@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Possible ambiguity in definition of DHCP client identifier op 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> If I am reading this right, it means that we always prefix a client
> identifier
> with 00: when stored as an LDAP attribute.

I think you and Bernie are saying the same thing, and I think it's not
the right thing to do.  What you're saying is that if you use the
client identifier as the key when you store it in LDAP, the first byte
of this key is always zero.  If you use the MAC address, the first
byte is the htype field of the DHCP packet and the subsequent bytes
are the significant bytes of chaddr.

What is missing from this is the possibility of having a key that is
derived from anything other than the client hardware address or the
client identifier.  And yet we already have identified the UUID as a
possible alternative option to use as a key, and some people use the
hostname option as the key as well.  It would be nice to be able to
represent these cleanly in an LDAP entry.

So I am proposing that we change the algorithm so that the first byte
(or two bytes, preferably) are a type code, and subsequent bytes are
the payload.  So if the client identifier is used as a key, we'd use a
type code of 61 (btw, Barr and Nathan, you have this as 63 in your
draft), and the rest of the key would be the payload in the client
identifier option.  If we use the hardware address, we'd use a type
code of 0, followed by the htype and hlen bytes of chaddr.  If we use
the UUID, the type code would be the DHCP option code for UUID (I
can't remember what it is) and the rest of the key would be the
payload of the UUID option.  Likewise with the hostname option:

chaddr:		00 00 htype chaddr [0..hlen]
client-id:	00 3D option payload
uuid:		?? ?? option payload
hostname:	00 12 option payload

By doing this, we maximize flexibility, at a fairly low cost.

			       _MelloN_


From dhcp-v4@bucknell.edu  Thu Nov  4 13:15:13 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA29849
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 13:15:13 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA31799;
	Thu, 4 Nov 1999 13:13:26 -0500 (EST)
Date: Thu, 4 Nov 1999 13:13:26 -0500 (EST)
Message-Id: <199911041802.NAA01217@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


I would suggest being less verbose and more structured, but basically
I think you've got the right idea.   Instead of what you've done, I'd
suggest:

> -----------------------------------------------------
> |60|??|a|p|p|l|e|/|O|T|-|2|.|5|.|3|/|i|M|a|c|/|3|0|0|
> -----------------------------------------------------

In other words, start with "apple" (the vendor), then "OT-2.5.3" (the
relevant software version), then "iMac" (the type of computer), then
"300" (the model number, which I completely made up).   Try to be
terse so as not to waste bytes - there aren't that many to go around.
If Apple has a registry of unique model names and numbers, I'd suggest
using that and referencing it.   If getting the model number within
the OT stack is hard, maybe skip it, but it's a big win if you can
send it.

I would also suggest that you implement the user class option soon, if
you can - it's available in Win98SE, and I suspect people are going to
start making heavy use of it soon, so if you can provide the same
functionality, you'll be able to present the same feature set
Microsoft is presenting.

			       _MelloN_


From dhcp-v4@bucknell.edu  Thu Nov  4 13:15:21 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA29920
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 13:15:20 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA08190;
	Thu, 4 Nov 1999 13:14:16 -0500 (EST)
Date: Thu, 4 Nov 1999 13:14:16 -0500 (EST)
Message-Id: <199911041808.KAA14556@schilling.ucdavis.edu>
Errors-To: droms@bucknell.edu
Reply-To: madorough@ucdavis.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Monika (Maria) Dorough" <madorough@ucdavis.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0.2 

Unsubscribe me please.


From dhcp-v4@bucknell.edu  Thu Nov  4 13:53:21 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA13866
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 13:53:20 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA14315;
	Thu, 4 Nov 1999 13:52:15 -0500 (EST)
Date: Thu, 4 Nov 1999 13:52:15 -0500 (EST)
Message-Id: <63D30D6E10CFD11190A90000F805FE8601E32247@lespaul.process.com>
Errors-To: droms@bucknell.edu
Reply-To: Bennett@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Andy Bennett <Bennett@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.0.1460.8)
Mime-Version: 1.0

Ted is right and I think his proposed representation is an improvement -
I'll change it in the next revision to the schema draft.

Andy

> -----Original Message-----
> From: Ted Lemon [mailto:mellon@isc.org]
> Sent: Thursday, November 04, 1999 1:13 PM
> To: Multiple recipients of list
> Subject: Re: Possible ambiguity in definition of DHCP client 
> identifier
> op
> 
> 
> 
> > If I am reading this right, it means that we always prefix a client
> > identifier
> > with 00: when stored as an LDAP attribute.
> 
> I think you and Bernie are saying the same thing, and I think it's not
> the right thing to do.  What you're saying is that if you use the
> client identifier as the key when you store it in LDAP, the first byte
> of this key is always zero.  If you use the MAC address, the first
> byte is the htype field of the DHCP packet and the subsequent bytes
> are the significant bytes of chaddr.
> 
> What is missing from this is the possibility of having a key that is
> derived from anything other than the client hardware address or the
> client identifier.  And yet we already have identified the UUID as a
> possible alternative option to use as a key, and some people use the
> hostname option as the key as well.  It would be nice to be able to
> represent these cleanly in an LDAP entry.
> 
> So I am proposing that we change the algorithm so that the first byte
> (or two bytes, preferably) are a type code, and subsequent bytes are
> the payload.  So if the client identifier is used as a key, we'd use a
> type code of 61 (btw, Barr and Nathan, you have this as 63 in your
> draft), and the rest of the key would be the payload in the client
> identifier option.  If we use the hardware address, we'd use a type
> code of 0, followed by the htype and hlen bytes of chaddr.  If we use
> the UUID, the type code would be the DHCP option code for UUID (I
> can't remember what it is) and the rest of the key would be the
> payload of the UUID option.  Likewise with the hostname option:
> 
> chaddr:		00 00 htype chaddr [0..hlen]
> client-id:	00 3D option payload
> uuid:		?? ?? option payload
> hostname:	00 12 option payload
> 
> By doing this, we maximize flexibility, at a fairly low cost.
> 
> 			       _MelloN_
> 


From dhcp-v4@bucknell.edu  Thu Nov  4 14:44:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA08092
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 14:44:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA03306;
	Thu, 4 Nov 1999 14:43:38 -0500 (EST)
Date: Thu, 4 Nov 1999 14:43:38 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE3F@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> > If I am reading this right, it means that we always prefix a client
> > identifier
> > with 00: when stored as an LDAP attribute.
> 
> I think you and Bernie are saying the same thing, and I think it's not
> the right thing to do.  What you're saying is that if you use the
> client identifier as the key when you store it in LDAP, the first byte
> of this key is always zero.  If you use the MAC address, the first
> byte is the htype field of the DHCP packet and the subsequent bytes
> are the significant bytes of chaddr.
> 
> What is missing from this is the possibility of having a key that is
> derived from anything other than the client hardware address or the
> client identifier.  And yet we already have identified the UUID as a
> possible alternative option to use as a key, and some people use the
> hostname option as the key as well.  It would be nice to be able to
> represent these cleanly in an LDAP entry.
> 
> So I am proposing that we change the algorithm so that the first byte
> (or two bytes, preferably) are a type code, and subsequent bytes are
> the payload.  So if the client identifier is used as a key, we'd use a
> type code of 61 (btw, Barr and Nathan, you have this as 63 in your
> draft), and the rest of the key would be the payload in the client
> identifier option.  If we use the hardware address, we'd use a type
> code of 0, followed by the htype and hlen bytes of chaddr.  If we use
> the UUID, the type code would be the DHCP option code for UUID (I
> can't remember what it is) and the rest of the key would be the
> payload of the UUID option.  Likewise with the hostname option:
> 
> chaddr:		00 00 htype chaddr [0..hlen]
> client-id:	00 3D option payload
> uuid:		?? ?? option payload
> hostname:	00 12 option payload
> 
> By doing this, we maximize flexibility, at a fairly low cost.
> 
> 
Agreeing with Ted, I like using fairly obvious values for the key type, such
as 61 for client identifier, 97 for client machine identifier (UUID), 12 for
hostname, 60 for vendor class identifier, etc., with code zero reserved for
a construct based on chaddr, htype, and hlen.

This is not a sneaky attempt to declare something other than client
identifier or chaddr to be the index for specifying a lease by the server or
client, just observing that there are other useful ways to find information.

--Barr




From dhcp-v4@bucknell.edu  Thu Nov  4 15:31:48 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA28543
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 15:31:47 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA15158;
	Thu, 4 Nov 1999 15:30:16 -0500 (EST)
Date: Thu, 4 Nov 1999 15:30:16 -0500 (EST)
Message-Id: <3821EAC4.98D6C800@connectnet.com>
Errors-To: droms@bucknell.edu
Reply-To: conx@connectnet.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: C J Considine <conx@connectnet.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.7 [en] (WinNT; U)
Content-Transfer-Encoding: 7bit

Doesn't everything relevant have a 12 digit UPC?  Most computer 
resellers won't touch anything without one.

(UPC Council 513-435-3870.)

Ted Lemon wrote:
> 
> I would suggest being less verbose and more structured, but basically
> I think you've got the right idea.   Instead of what you've done, I'd
> suggest:
> 
> > -----------------------------------------------------
> > |60|??|a|p|p|l|e|/|O|T|-|2|.|5|.|3|/|i|M|a|c|/|3|0|0|
> > -----------------------------------------------------
> 
> In other words, start with "apple" (the vendor), then "OT-2.5.3" (the
> relevant software version), then "iMac" (the type of computer), then
> "300" (the model number, which I completely made up).   Try to be
> terse so as not to waste bytes - there aren't that many to go around.
> If Apple has a registry of unique model names and numbers, I'd suggest
> using that and referencing it.   If getting the model number within
> the OT stack is hard, maybe skip it, but it's a big win if you can
> send it.
> 
> I would also suggest that you implement the user class option soon, if
> you can - it's available in Win98SE, and I suspect people are going to
> start making heavy use of it soon, so if you can provide the same
> functionality, you'll be able to present the same feature set
> Microsoft is presenting.
> 
>                                _MelloN_


From dhcp-v4@bucknell.edu  Thu Nov  4 16:47:16 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA00575
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 16:47:15 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA09647;
	Thu, 4 Nov 1999 16:45:31 -0500 (EST)
Date: Thu, 4 Nov 1999 16:45:31 -0500 (EST)
Message-Id: <199911042036.PAA01847@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> Doesn't everything relevant have a 12 digit UPC?  Most computer 
> resellers won't touch anything without one.

Probably, but that doesn't help unless we have a UPC to model+model
number table, which is probably more work to get than just having the
information right there where we need it.

			       _MelloN_


From dhcp-v4@bucknell.edu  Thu Nov  4 16:51:51 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA02925
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 16:51:49 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA01481;
	Thu, 4 Nov 1999 16:50:52 -0500 (EST)
Date: Thu, 4 Nov 1999 16:50:52 -0500 (EST)
Message-Id: <EA5107D3A9ADD211B7530000C041E4F4011B2ADA@honts318.wal-mart.com>
Errors-To: droms@bucknell.edu
Reply-To: ndlane@wal-mart.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Nathan Lane <ndlane@wal-mart.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0



> > type code of 61 (btw, Barr and Nathan, you have this as 63 in your
> > draft), and the rest of the key would be the payload in the client
> 
	[Nathan Lane]  Thanks for pointing that out.

> > identifier option.  If we use the hardware address, we'd use a type
> > code of 0, followed by the htype and hlen bytes of chaddr.  If we use
> > the UUID, the type code would be the DHCP option code for UUID (I
> > can't remember what it is) and the rest of the key would be the
> > payload of the UUID option.  Likewise with the hostname option:
> > 
> > chaddr:		00 00 htype chaddr [0..hlen]
> > client-id:	00 3D option payload
> > uuid:		?? ?? option payload
> > hostname:	00 12 option payload
> > 
> > By doing this, we maximize flexibility, at a fairly low cost.
> > 
> > 
> Agreeing with Ted, I like using fairly obvious values for the key type, such
> as 61 for client identifier, 97 for client machine identifier (UUID), 12 for
> hostname, 60 for vendor class identifier, etc., with code zero reserved for
> a construct based on chaddr, htype, and hlen.
> 
> 
	I like this as well, giving everyone the ultimate in flexibility at a very small cost in added complexity.

	-Nathan Lane



From dhcp-v4@bucknell.edu  Thu Nov  4 18:08:39 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA04929
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 18:08:39 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA21680;
	Thu, 4 Nov 1999 18:07:16 -0500 (EST)
Date: Thu, 4 Nov 1999 18:07:16 -0500 (EST)
Message-Id: <EA5107D3A9ADD211B7530000C041E4F4011B2ADB@honts318.wal-mart.com>
Errors-To: droms@bucknell.edu
Reply-To: ndlane@wal-mart.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Nathan Lane <ndlane@wal-mart.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0

Actually, using the UPC tends to become very difficult.  The actual standards are VERY overloaded and there is definite talk of major change occuring in this arena very soon as the limitations of the current system become glaringly apparent, especially for someone like us.  Then you'd have to decide which UPC to use, which standard (UCC-12, UCC-18, etc.), which SKU it actually represented.  Let's not go there.

-Nathan Lane
Senior Network Engineer
Wal-Mart Stores, Inc.
(501) 277-5786


> -----Original Message-----
> From:	C J Considine [SMTP:conx@connectnet.com]
> Sent:	Thursday, November 04, 1999 2:32 PM
> To:	Multiple recipients of list
> Subject:	Re: DHCP Vendor Class question
> 
> Doesn't everything relevant have a 12 digit UPC?  Most computer 
> resellers won't touch anything without one.
> 
> (UPC Council 513-435-3870.)
> 
> 


From dhcp-v4@bucknell.edu  Thu Nov  4 18:13:52 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA07055
	for <dhc-archive@odin.ietf.org>; Thu, 4 Nov 1999 18:13:52 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA06773;
	Thu, 4 Nov 1999 18:13:14 -0500 (EST)
Date: Thu, 4 Nov 1999 18:13:14 -0500 (EST)
Message-Id: <EA5107D3A9ADD211B7530000C041E4F4011B2ADC@honts318.wal-mart.com>
Errors-To: droms@bucknell.edu
Reply-To: ndlane@wal-mart.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Nathan Lane <ndlane@wal-mart.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0



> -----Original Message-----
> From:	Ted Lemon [SMTP:mellon@isc.org]
> Sent:	Thursday, November 04, 1999 12:14 PM
> To:	Multiple recipients of list
> Subject:	Re: DHCP Vendor Class question 
> 
> 
> I would suggest being less verbose and more structured, but basically
> I think you've got the right idea.   Instead of what you've done, I'd
> suggest:
> 
> > -----------------------------------------------------
> > |60|??|a|p|p|l|e|/|O|T|-|2|.|5|.|3|/|i|M|a|c|/|3|0|0|
> > -----------------------------------------------------
> 
> In other words, start with "apple" (the vendor), then "OT-2.5.3" (the
> relevant software version), then "iMac" (the type of computer), then
> "300" (the model number, which I completely made up).   Try to be
> terse so as not to waste bytes - there aren't that many to go around.
> If Apple has a registry of unique model names and numbers, I'd suggest
> using that and referencing it.   If getting the model number within
> the OT stack is hard, maybe skip it, but it's a big win if you can
> send it.
> 
	[Nathan Lane]  I LIKE THIS!  This is exactly the use of VID that gives me a great deal of information, works within my existing framework and lets me do very intelligent client selection... 

	However, let's make sure that if the Apple client does send this vendor identifier that it expects, uses and processes the possible return of option 43 data in a manner consistent with all of our interpretations of 2131 and 2132.  (I believe most of us feel that option 43 can be returned to the client, if asked for, on a per VID basis or simply ignored if the server is unprepared to handle it.  The client, though, should most definitely be able to function at the network layer in lieu of receipt of option 43.)
>  
> I would also suggest that you implement the user class option soon, if
> you can - it's available in Win98SE, and I suspect people are going to
> start making heavy use of it soon, so if you can provide the same
> functionality, you'll be able to present the same feature set
> Microsoft is presenting.
> 
	[Now, I'd also really, really like to see user class option but the draft isn't far enough to tell my vendors "here, go read it and implement this for me before I'll let you put your product on my network".    I have seen the user class implemented, but it's not "enforcable" yet.

	-Nathan Lane
>  


From dhcp-v4@bucknell.edu  Fri Nov  5 02:30:29 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id CAA25733
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 02:30:29 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id CAA17106;
	Fri, 5 Nov 1999 02:29:46 -0500 (EST)
Date: Fri, 5 Nov 1999 02:29:46 -0500 (EST)
Message-Id: <19991105070333Z8102-195+3017@stovokor.epilogue.com>
Errors-To: droms@bucknell.edu
Reply-To: wes@epilogue.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Bill Sommerfeld <wes@epilogue.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4

> Doesn't everything relevant have a 12 digit UPC?  Most computer 
> resellers won't touch anything without one.
> 
> (UPC Council 513-435-3870.)

nope.  not everything you could think of as a "vendor" is actually
*selling* something with a UPC code.. 

I.e., a free OS might want to use DHCP for autoconfiguration during
installation.

					- Bill


From dhcp-v4@bucknell.edu  Fri Nov  5 07:29:42 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id HAA22806
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 07:29:41 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id HAA21291;
	Fri, 5 Nov 1999 07:29:23 -0500 (EST)
Date: Fri, 5 Nov 1999 07:29:23 -0500 (EST)
Message-Id: <199911051228.HAA24744@uplink.net>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Agneda for WG meeting
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Here is the latest agenda.  Note that I've moved the LDAP schema discussion
to the 1300 session.

I'll post an annotated version with a list of associated I-Ds later today.

For those of you on the agenda - I'll have a PowerPoint-equipped laptop
available for use with the in-room projection equipment.  You're welcome to
use it for your presentation; bring along a floppy for sneaker-netting...

- Ralph

=====


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

  Yves T'Joens (Alcatel)         DHCP reconfigure message

  Bill Arbaugh (Penn)            DHCPv4 authentication

  Carl Smith (Sun)               Name service ordering option

  Kim Kinnear (Cisco)            Failover protocol

  Jerome Privat (BT)             User Class option

  Jerome Privat (BT)             Doublephase configuration

  Mike Patrick (Motorola)        Agent options

  Subir Das (Telcordia)          Roaming among commercial
                                 wireless networks

  Bernard Aboba (Microsoft)      Kerberos authentication

  Bernie Volz (Process Software) Server load balancing

  Bill Sommerfield (Epilogue)    Server selection option

  Barr Hibbs (Pacbell)           DHCP MIB

  Barr Hibbs (Pacbell)           Client option clarifications
  Juan Garay (Bell Labs)         Option for NAT/IPSec interaction

Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================
  Bernie Volz (Process Software) LDAP schema

  Mark Stapp (Cisco)             DHCP-DNS interaction

  Andreas Gustafsson             A DNS RR for DHCP information
  (Internet Engines, Inc.)

  Mike Carney (Sun)              DHCP options guidelines

  Mike Carney (Sun)              Connectathon

Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
  Mike Carney (Sun)              DHCPv6
  Jim Bound (Compaq)
  Charlie Perkins (Nokia)





From dhcp-v6@bucknell.edu  Fri Nov  5 08:10:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id IAA06239
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 08:10:57 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA07365;
	Fri, 5 Nov 1999 08:08:59 -0500 (EST)
Date: Fri, 5 Nov 1999 08:08:59 -0500 (EST)
Message-Id: <199911051228.HAA24744@uplink.net>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Agneda for WG meeting
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Here is the latest agenda.  Note that I've moved the LDAP schema discussion
to the 1300 session.

I'll post an annotated version with a list of associated I-Ds later today.

For those of you on the agenda - I'll have a PowerPoint-equipped laptop
available for use with the in-room projection equipment.  You're welcome to
use it for your presentation; bring along a floppy for sneaker-netting...

- Ralph

=====


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

  Yves T'Joens (Alcatel)         DHCP reconfigure message

  Bill Arbaugh (Penn)            DHCPv4 authentication

  Carl Smith (Sun)               Name service ordering option

  Kim Kinnear (Cisco)            Failover protocol

  Jerome Privat (BT)             User Class option

  Jerome Privat (BT)             Doublephase configuration

  Mike Patrick (Motorola)        Agent options

  Subir Das (Telcordia)          Roaming among commercial
                                 wireless networks

  Bernard Aboba (Microsoft)      Kerberos authentication

  Bernie Volz (Process Software) Server load balancing

  Bill Sommerfield (Epilogue)    Server selection option

  Barr Hibbs (Pacbell)           DHCP MIB

  Barr Hibbs (Pacbell)           Client option clarifications
  Juan Garay (Bell Labs)         Option for NAT/IPSec interaction

Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================
  Bernie Volz (Process Software) LDAP schema

  Mark Stapp (Cisco)             DHCP-DNS interaction

  Andreas Gustafsson             A DNS RR for DHCP information
  (Internet Engines, Inc.)

  Mike Carney (Sun)              DHCP options guidelines

  Mike Carney (Sun)              Connectathon

Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
  Mike Carney (Sun)              DHCPv6
  Jim Bound (Compaq)
  Charlie Perkins (Nokia)





From dhcp-v4@bucknell.edu  Fri Nov  5 11:29:07 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA14886
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 11:29:05 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA23210;
	Fri, 5 Nov 1999 10:30:21 -0500 (EST)
Date: Fri, 5 Nov 1999 10:30:21 -0500 (EST)
Message-Id: <01e001bf2799$13a938c0$dbb2f4d1@computer>
Errors-To: droms@bucknell.edu
Reply-To: jcoignard@digitalairways.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Jerome Coignard" <jcoignard@digitalairways.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: unsuscribe
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: Microsoft Outlook Express 5.00.2615.200
Content-Transfer-Encoding: 7bit
Content-Type: text/plain;
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

please unsuscribe me



From dhcp-v6@bucknell.edu  Fri Nov  5 14:07:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA25554
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 14:07:57 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA08335;
	Fri, 5 Nov 1999 14:06:43 -0500 (EST)
Date: Fri, 5 Nov 1999 14:06:43 -0500 (EST)
Message-Id: <199911051822.NAA12462@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Agenda (I-D versions corrected)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

I missed a few I-D revs in the previous agenda.  Here's the agenda with the
correct I-D rev numbers.

Sorry for the extra e-mail...

- Ralph

=====


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

Dynamic host configuration : DHCP reconfigure extension Yves T'Joens (Alcatel)
draft-schrijvp-dhcpv4-reconfigure-00.txt

Authentication for DHCP Messages                        Bill Arbaugh (Penn)
draft-ietf-dhc-authentication-12.txt

The Name Service Search Option for DHCP                 Carl Smith (Sun)
draft-ietf-dhc-nsso-01.txt

DHCP Failover Protocol                                  Kim Kinnear (Cisco)
draft-ietf-dhc-failover-05.txt

The User Class Option for DHCP                          Jerome Privat (BT)
draft-ietf-dhc-userclass-04.txt

Double Phase DHCP Configuration                         Jerome Privat (BT)
draft-privat-dhc-doublephase-01.txt

DHCP Relay Agent Information Option                     Mike Patrick
(Motorola)
draft-ietf-dhc-agent-options-07.txt

Dynamic Registration and Configuration Protocol (DRCP)  Subir Das (Telcordia)
draft-itsumo-drcp-00.txt

DHCP Authentication Via Kerberos V                      Bernard Aboba
(Microsoft)
draft-hornstein-dhc-kerbauth-01.txt

DHC load balancing algorithm                            Bernie Volz
(Process Software)
draft-ietf-dhc-loadb-00.txt

The Server Selection Option for DHCP                    Bill Sommerfield
(Epilogue)
draft-ietf-dhc-sso-03.txt

Dynamic Host Configuration Protocol (DHCP) Server MIB   Barr Hibbs (Pacbell)
draft-ietf-dhc-server-mib-04.txt

Interpreting Client Options for DHCP                    Barr Hibbs (Pacbell)
draft-ietf-dhc-client-options-00.txt

Option for NAT/IPSec interaction                        Jose Brustoloni
(Bell Labs)


Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================

DHCP Schema for LDAP                                    Bernie Volz
(Process Software)
draft-ietf-dhc-schema-01.txt

A DNS RR for DHCP information                           Andreas Gustafsson
(Internet Engines
draft-ietf-dnsind-dhcp-rr-00.txt

Interaction between DHCP and DNS                        Mark Stapp (Cisco)
draft-ietf-dhc-dhcp-dns-11.txt

New Option Review Guidelines and                        Mike Carney (Sun)
  Additional Option Namespace
draft-ietf-dhc-option-review-and-namespace-01.txt

Connectathon                                            Mike Carney (Sun)


Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
Dynamic Host Configuration Protocol for IPv6 (DHCPv6)   Mike Carney (Sun)
draft-ietf-dhc-dhcpv6-14.txt                            Jim  Bound (Compaq)
Extensions for the DHCP for IPv6                        Charlie Perkins
(Nokia)    
draft-ietf-dhc-v6exts-11.txt




From dhcp-v6@bucknell.edu  Fri Nov  5 14:08:00 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA25584
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 14:08:00 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA26623;
	Fri, 5 Nov 1999 14:04:37 -0500 (EST)
Date: Fri, 5 Nov 1999 14:04:37 -0500 (EST)
Message-Id: <Version.32.19991105124954.00e0f840@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Agenda (reformatted) and on-line
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

I've included a reformatted (but otherwise almost unchanged) version
of the WG agenda for our meetings in DC.  The agenda is also available
at http://www.dhcp.org/9911-meeting.html which includes links to the 
Internet Drafts associated with each agenda item.

- Ralph

=====


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

Dynamic host configuration : DHCP reconfigure extension Yves T'Joens (Alcatel)
draft-schrijvp-dhcpv4-reconfigure-00.txt

Authentication for DHCP Messages                        Bill Arbaugh (Penn)
draft-ietf-dhc-authentication-11.txt

The Name Service Search Option for DHCP                 Carl Smith (Sun)
draft-ietf-dhc-nsso-01.txt

DHCP Failover Protocol                                  Kim Kinnear (Cisco)
draft-ietf-dhc-failover-05.txt

The User Class Option for DHCP                          Jerome Privat (BT)
draft-ietf-dhc-userclass-04.txt

Double Phase DHCP Configuration                         Jerome Privat (BT)
draft-privat-dhc-doublephase-01.txt

DHCP Relay Agent Information Option                     Mike Patrick
(Motorola)
draft-ietf-dhc-agent-options-07.txt

Dynamic Registration and Configuration Protocol (DRCP)  Subir Das (Telcordia)
draft-itsumo-drcp-00.txt

DHCP Authentication Via Kerberos V                      Bernard Aboba
(Microsoft)
draft-hornstein-dhc-kerbauth-00.txt

DHC load balancing algorithm                            Bernie Volz
(Process Software)
draft-ietf-dhc-loadb-00.txt

The Server Selection Option for DHCP                    Bill Sommerfield
(Epilogue)
draft-ietf-dhc-sso-03.txt

Dynamic Host Configuration Protocol (DHCP) Server MIB   Barr Hibbs (Pacbell)
draft-ietf-dhc-server-mib-04.txt

Interpreting Client Options for DHCP                    Barr Hibbs (Pacbell)
draft-ietf-dhc-client-options-00.txt

Option for NAT/IPSec interaction                        Jose Brustoloni
(Bell Labs)


Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================

DHCP Schema for LDAP                                    Bernie Volz
(Process Software)
draft-ietf-dhc-schema-00.txt

A DNS RR for DHCP information                           Andreas Gustafsson
(Internet Engines)
draft-ietf-dnsind-dhcp-rr-00.txt

Interaction between DHCP and DNS                        Mark Stapp (Cisco)
draft-ietf-dhc-dhcp-dns-10.txt

New Option Review Guidelines and                        Mike Carney (Sun)
  Additional Option Namespace
draft-ietf-dhc-option-review-and-namespace-00.txt

Connectathon                                            Mike Carney (Sun)


Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
Dynamic Host Configuration Protocol for IPv6 (DHCPv6)   Mike Carney (Sun)
draft-ietf-dhc-dhcpv6-14.txt                            Jim  Bound (Compaq)
Extensions for the DHCP for IPv6                        Charlie Perkins
(Nokia)    
draft-ietf-dhc-v6exts-11.txt





From dhcp-v4@bucknell.edu  Fri Nov  5 14:11:41 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA27101
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 14:11:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA13444;
	Fri, 5 Nov 1999 14:10:31 -0500 (EST)
Date: Fri, 5 Nov 1999 14:10:31 -0500 (EST)
Message-Id: <4.2.0.58.19991105094637.00a131c0@mail.join.com>
Errors-To: droms@bucknell.edu
Reply-To: robs@join.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Rob Stevens <robs@join.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"; format=flowed
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 

A
>Ted Lemon wrote:
> >
> > I would suggest being less verbose and more structured, but basically
> > I think you've got the right idea.   Instead of what you've done, I'd
> > suggest:
> >
> > > -----------------------------------------------------
> > > |60|??|a|p|p|l|e|/|O|T|-|2|.|5|.|3|/|i|M|a|c|/|3|0|0|
> > > -----------------------------------------------------

I am not very enthusiastic about constructing vendor-class IDs
with infra-structure like this, for the following reason.

Each unique vendor class can, in principle, define 256
entirely new options (which get encapsulated in super-option
#43). If, to take this specific case, Apple were to define
vendor-specific options then with each variant of the
vendor class they would have to publish some document
desccribing what the vendor options were for that specific
vendor class. The more sub-structure you put into the
vendor class, the more the burden of that paperwork.

Of course, If Apple really desired that each new release
of the OS defined completely new vendor options then
this would be OK. Or if Apple never defined any options
unique to them.

But generally it seems to me that the options that Apple
or any other vendor might define are likely to be identical
across a wide range of hardware and software.

Since there has lately been a flurry of discussion about interpreting
client identifiers, this might be a good time to think about
whether vendor class should also have a well-defined
canonical sub-structure. E.g. that the first sub-field (however
defined) would be what defined the syntax and semantics
of vendor specific options, and the whole field was used
as the key which servers used to determine their values.

Rob Stevens



From dhcp-v4@bucknell.edu  Fri Nov  5 14:19:20 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA00269
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 14:19:19 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA17610;
	Fri, 5 Nov 1999 14:18:49 -0500 (EST)
Date: Fri, 5 Nov 1999 14:18:49 -0500 (EST)
Message-Id: <Version.32.19991105124954.00e0f840@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Agenda (reformatted) and on-line
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

I've included a reformatted (but otherwise almost unchanged) version
of the WG agenda for our meetings in DC.  The agenda is also available
at http://www.dhcp.org/9911-meeting.html which includes links to the 
Internet Drafts associated with each agenda item.

- Ralph

=====


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

Dynamic host configuration : DHCP reconfigure extension Yves T'Joens (Alcatel)
draft-schrijvp-dhcpv4-reconfigure-00.txt

Authentication for DHCP Messages                        Bill Arbaugh (Penn)
draft-ietf-dhc-authentication-11.txt

The Name Service Search Option for DHCP                 Carl Smith (Sun)
draft-ietf-dhc-nsso-01.txt

DHCP Failover Protocol                                  Kim Kinnear (Cisco)
draft-ietf-dhc-failover-05.txt

The User Class Option for DHCP                          Jerome Privat (BT)
draft-ietf-dhc-userclass-04.txt

Double Phase DHCP Configuration                         Jerome Privat (BT)
draft-privat-dhc-doublephase-01.txt

DHCP Relay Agent Information Option                     Mike Patrick
(Motorola)
draft-ietf-dhc-agent-options-07.txt

Dynamic Registration and Configuration Protocol (DRCP)  Subir Das (Telcordia)
draft-itsumo-drcp-00.txt

DHCP Authentication Via Kerberos V                      Bernard Aboba
(Microsoft)
draft-hornstein-dhc-kerbauth-00.txt

DHC load balancing algorithm                            Bernie Volz
(Process Software)
draft-ietf-dhc-loadb-00.txt

The Server Selection Option for DHCP                    Bill Sommerfield
(Epilogue)
draft-ietf-dhc-sso-03.txt

Dynamic Host Configuration Protocol (DHCP) Server MIB   Barr Hibbs (Pacbell)
draft-ietf-dhc-server-mib-04.txt

Interpreting Client Options for DHCP                    Barr Hibbs (Pacbell)
draft-ietf-dhc-client-options-00.txt

Option for NAT/IPSec interaction                        Jose Brustoloni
(Bell Labs)


Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================

DHCP Schema for LDAP                                    Bernie Volz
(Process Software)
draft-ietf-dhc-schema-00.txt

A DNS RR for DHCP information                           Andreas Gustafsson
(Internet Engines)
draft-ietf-dnsind-dhcp-rr-00.txt

Interaction between DHCP and DNS                        Mark Stapp (Cisco)
draft-ietf-dhc-dhcp-dns-10.txt

New Option Review Guidelines and                        Mike Carney (Sun)
  Additional Option Namespace
draft-ietf-dhc-option-review-and-namespace-00.txt

Connectathon                                            Mike Carney (Sun)


Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
Dynamic Host Configuration Protocol for IPv6 (DHCPv6)   Mike Carney (Sun)
draft-ietf-dhc-dhcpv6-14.txt                            Jim  Bound (Compaq)
Extensions for the DHCP for IPv6                        Charlie Perkins
(Nokia)    
draft-ietf-dhc-v6exts-11.txt





From dhcp-v4@bucknell.edu  Fri Nov  5 14:24:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA02466
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 14:24:33 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA19879;
	Fri, 5 Nov 1999 14:23:11 -0500 (EST)
Date: Fri, 5 Nov 1999 14:23:11 -0500 (EST)
Message-Id: <199911051822.NAA12462@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Agenda (I-D versions corrected)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

I missed a few I-D revs in the previous agenda.  Here's the agenda with the
correct I-D rev numbers.

Sorry for the extra e-mail...

- Ralph

=====


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

Dynamic host configuration : DHCP reconfigure extension Yves T'Joens (Alcatel)
draft-schrijvp-dhcpv4-reconfigure-00.txt

Authentication for DHCP Messages                        Bill Arbaugh (Penn)
draft-ietf-dhc-authentication-12.txt

The Name Service Search Option for DHCP                 Carl Smith (Sun)
draft-ietf-dhc-nsso-01.txt

DHCP Failover Protocol                                  Kim Kinnear (Cisco)
draft-ietf-dhc-failover-05.txt

The User Class Option for DHCP                          Jerome Privat (BT)
draft-ietf-dhc-userclass-04.txt

Double Phase DHCP Configuration                         Jerome Privat (BT)
draft-privat-dhc-doublephase-01.txt

DHCP Relay Agent Information Option                     Mike Patrick
(Motorola)
draft-ietf-dhc-agent-options-07.txt

Dynamic Registration and Configuration Protocol (DRCP)  Subir Das (Telcordia)
draft-itsumo-drcp-00.txt

DHCP Authentication Via Kerberos V                      Bernard Aboba
(Microsoft)
draft-hornstein-dhc-kerbauth-01.txt

DHC load balancing algorithm                            Bernie Volz
(Process Software)
draft-ietf-dhc-loadb-00.txt

The Server Selection Option for DHCP                    Bill Sommerfield
(Epilogue)
draft-ietf-dhc-sso-03.txt

Dynamic Host Configuration Protocol (DHCP) Server MIB   Barr Hibbs (Pacbell)
draft-ietf-dhc-server-mib-04.txt

Interpreting Client Options for DHCP                    Barr Hibbs (Pacbell)
draft-ietf-dhc-client-options-00.txt

Option for NAT/IPSec interaction                        Jose Brustoloni
(Bell Labs)


Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================

DHCP Schema for LDAP                                    Bernie Volz
(Process Software)
draft-ietf-dhc-schema-01.txt

A DNS RR for DHCP information                           Andreas Gustafsson
(Internet Engines
draft-ietf-dnsind-dhcp-rr-00.txt

Interaction between DHCP and DNS                        Mark Stapp (Cisco)
draft-ietf-dhc-dhcp-dns-11.txt

New Option Review Guidelines and                        Mike Carney (Sun)
  Additional Option Namespace
draft-ietf-dhc-option-review-and-namespace-01.txt

Connectathon                                            Mike Carney (Sun)


Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
Dynamic Host Configuration Protocol for IPv6 (DHCPv6)   Mike Carney (Sun)
draft-ietf-dhc-dhcpv6-14.txt                            Jim  Bound (Compaq)
Extensions for the DHCP for IPv6                        Charlie Perkins
(Nokia)    
draft-ietf-dhc-v6exts-11.txt




From dhcp-v4@bucknell.edu  Fri Nov  5 15:50:36 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA04362
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 15:50:35 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA06296;
	Fri, 5 Nov 1999 15:49:41 -0500 (EST)
Date: Fri, 5 Nov 1999 15:49:41 -0500 (EST)
Message-Id: <199911051921.OAA06389@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> Each unique vendor class can, in principle, define 256
> entirely new options (which get encapsulated in super-option
> #43). If, to take this specific case, Apple were to define
> vendor-specific options then with each variant of the
> vendor class they would have to publish some document
> desccribing what the vendor options were for that specific
> vendor class. The more sub-structure you put into the
> vendor class, the more the burden of that paperwork.

Rob, what I described is the way that sun currently does vendor
classes, roughly (they use dots instead of slashes, and don't send a
transport version).   A DHCP server that's going to support vendor
classing needs to be able to say "the following vendor option code
space works for the following set of possible values of vendor class."
I do not suggest that Apple implement a completely different set of
options for each version of OT, or for each model of computer
supported, but being able to tell the DHCP server what model of
computer you are is a good thing, IMHO, and I really like the way Sun
did it.   I'm just suggesting that Apple should follow in Sun's
footsteps here.

I don't think that vendor class identifiers can or should have any
sort of well-defined substructure.   I am just suggesting that one
particular vendor, in choosing a substructure that *they* will use, do
it a certain way.

			       _MelloN_


From dhcp-v4@bucknell.edu  Fri Nov  5 15:55:20 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA06332
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 15:55:20 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA12656;
	Fri, 5 Nov 1999 15:54:38 -0500 (EST)
Date: Fri, 5 Nov 1999 15:54:38 -0500 (EST)
Message-Id: <D5E932F578EBD111AC3F00A0C96B1E6F04E953F6@orsmsx31.jf.intel.com>
Errors-To: droms@bucknell.edu
Reply-To: mike.henry@intel.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Henry, Mike" <mike.henry@intel.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

Rob,

I agree completely.  The use of option 60 as suggested by Ted L. (and others
at different times) removes the distinction between "class" and "attribute".
It also begs the question of why option 60 should be considered "opaque".

Another way of looking at this is that including variant information in the
option class effectively makes this variant information ad hoc subtags. The
spec clearly states that option 43 is used to pass (variant) information to
clients and *servers*.  I am certainly not arguing that DHCP servers
actually do this (make use of option 43 info from clients), but it is what
the spec says.

 - Mike Henry

> -----Original Message-----
> From: Rob Stevens [mailto:robs@join.com]
> Sent: Friday, November 05, 1999 11:12 AM
> To: Multiple recipients of list
> Subject: Re: DHCP Vendor Class question
> 
> 
> A
> >Ted Lemon wrote:
> > >
> > > I would suggest being less verbose and more structured, 
> but basically
> > > I think you've got the right idea.   Instead of what 
> you've done, I'd
> > > suggest:
> > >
> > > > -----------------------------------------------------
> > > > |60|??|a|p|p|l|e|/|O|T|-|2|.|5|.|3|/|i|M|a|c|/|3|0|0|
> > > > -----------------------------------------------------
> 
> I am not very enthusiastic about constructing vendor-class IDs
> with infra-structure like this, for the following reason.
> 
> Each unique vendor class can, in principle, define 256
> entirely new options (which get encapsulated in super-option
> #43). If, to take this specific case, Apple were to define
> vendor-specific options then with each variant of the
> vendor class they would have to publish some document
> desccribing what the vendor options were for that specific
> vendor class. The more sub-structure you put into the
> vendor class, the more the burden of that paperwork.
> 
> Of course, If Apple really desired that each new release
> of the OS defined completely new vendor options then
> this would be OK. Or if Apple never defined any options
> unique to them.
> 
> But generally it seems to me that the options that Apple
> or any other vendor might define are likely to be identical
> across a wide range of hardware and software.
> 
> Since there has lately been a flurry of discussion about interpreting
> client identifiers, this might be a good time to think about
> whether vendor class should also have a well-defined
> canonical sub-structure. E.g. that the first sub-field (however
> defined) would be what defined the syntax and semantics
> of vendor specific options, and the whole field was used
> as the key which servers used to determine their values.
> 
> Rob Stevens
> 


From dhcp-v4@bucknell.edu  Fri Nov  5 17:18:49 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA04700
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 17:18:48 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA16122;
	Fri, 5 Nov 1999 17:16:32 -0500 (EST)
Date: Fri, 5 Nov 1999 17:16:32 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE4D@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> I agree completely.  The use of option 60 as suggested by Ted L. (and
> others at different times) removes the distinction between "class" and
> "attribute".  It also begs the question of why option 60 should be
> considered "opaque".
> 
> Another way of looking at this is that including variant information in
> the
> option class effectively makes this variant information ad hoc subtags.
> The spec clearly states that option 43 is used to pass (variant)
> information to clients and *servers*.  I am certainly not arguing that
> DHCP servers actually do this (make use of option 43 info from clients),
> but it is what the spec says.
> 
>  - Mike Henry
> 
> 
..while I agree that vendor class identifiers SHOULD contain sufficient
useful information to help us poor humans to easily differentiate them (and
maybe identify the product to which it applies) I also think that it should
be treated as an opaque string, without a defined structure.

If we don't, I'm afraid we'll simply have a repeat of the confusion
surrounding client identifiers.

That said, the "opaqueness" of a string is relative:  I believe the server
should not interpret the string, but I also believe it is much more
human-friendly for the string to be display characters.  That, however, is
merely my preference for ease in configuring servers to respond.

As for the use of option 43, I agree that the RFC currently says that the
server should use it as you describe, but I'm not sure that is the intended
behavior.  Hopefully we'll have a few moments to discuss this in the Working
Group meeting next week.

--Barr



From dhcp-v4@bucknell.edu  Fri Nov  5 17:23:00 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA05608
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 17:23:00 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA25787;
	Fri, 5 Nov 1999 17:21:17 -0500 (EST)
Date: Fri, 5 Nov 1999 17:21:17 -0500 (EST)
Message-Id: <199911052106.QAA25884@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v4@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: dhcp-v4@bucknell.edu (by way of Ralph Droms <dhcp-v4@bucknell.edu>)
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: extracting list
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


message sent to dhcp-v4@bucknell.edu by ORIGIN@SNPE.COM follows.
----------------------------------------------------------------
Hi,

I would like to extract the list there is on the dhcp manager (adress ip
with the scope, and adress in use, adress reserved).
It would be easier to administrate.

In winnt\system32\dhcp, there is a database file I don't manage to
read....?

Can you help me?

Marc Broussard



From dhcp-v4@bucknell.edu  Fri Nov  5 18:22:07 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA22314
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 18:22:07 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA13941;
	Fri, 5 Nov 1999 18:20:25 -0500 (EST)
Date: Fri, 5 Nov 1999 18:20:25 -0500 (EST)
Message-Id: <199911052226.RAA07202@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> I believe the server should not interpret the string

Um.   What exactly is the server supposed to do, then?   It has to use
the vendor identifier to look up the information it needs to send back
vendor-specific options.   Isn't that interpreting the string?

The question is, is the string a key into a table of possible
vendor-specific option spaces, and is there just one such key per
vendor, or can the vendor stuff meaningful information into it in the
hopes that the server will allow the information to be used?

What the standard says now is that the contents are unspecified, not
that they can't contain information, or that the tag must be a key
into a table, with one possible value per vendor.

I think that the vendor should put into the vendor-identifier option
whatever information will help the DHCP server administrator to send
the right options to the client, and having the string vary depending
on the hardware type makes a great deal of sense to me.  This seems to
be how Sun interprets it, and I think Mike Carney might even have
written the specification.  Mike, can you comment on this?

			       _MelloN_



From dhcp-v4@bucknell.edu  Fri Nov  5 18:24:12 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA22769
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 18:24:11 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA11814;
	Fri, 5 Nov 1999 18:22:32 -0500 (EST)
Date: Fri, 5 Nov 1999 18:22:32 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE50@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> > I believe the server should not interpret the string
> 
> Um.   What exactly is the server supposed to do, then?   It has to use
> the vendor identifier to look up the information it needs to send back
> vendor-specific options.   Isn't that interpreting the string?
> 
What I was trying to say was that I don't believe that the vendor class
identifier should be interpreted any more by the server than is the client
identifier.

> The question is, is the string a key into a table of possible
> vendor-specific option spaces, and is there just one such key per
> vendor, or can the vendor stuff meaningful information into it in the
> hopes that the server will allow the information to be used?
> 
While having the server interpret the identifier and base its response on
the contents has some appeal, how do we specify a general method that can be
easily implemented in the universe of clients and servers?  This MIGHT imply
that vendors should somehow supply "plug-ins" so that servers can act on the
identifier, or it may be that vendors would have to supply written
specifications so that administrators have to deal with vendor-specific
processing.  Check the Intel PXE spec for an example of how this could
happen.

> What the standard says now is that the contents are unspecified, not
> that they can't contain information, or that the tag must be a key
> into a table, with one possible value per vendor.
> 
Agreed.  I overlaid my expectations that there would not be a standardized
encoding of vendor class information, and that user-programmable response to
the contents of options 60 and 43 would not be common, onto the comment.

> I think that the vendor should put into the vendor-identifier option
> whatever information will help the DHCP server administrator to send
> the right options to the client, and having the string vary depending
> on the hardware type makes a great deal of sense to me.  This seems to be
> how Sun interprets it, and I think Mike Carney might even have
> written the specification.  Mike, can you comment on this?
> 
If I gave the impression that I didn't expect the server to somehow use the
information in options 60 and 43 to form responses, I apologize.  What I had
intended to say is that (1) mandated structures for class identifiers
probably can't cover all cases of interest, and (2) suggested structures
often are subject to the vagaries of interpretation, so that I don't believe
an RFC ought to specify the contents or format of the class identifier.

You seem to be leaning in that direction by stating that [one of] the
purpose[s] of class identifiers is to provide the server administrator a way
to deliver the right options to a client.  Hey!  I agree with you, I just
don't think it's a good idea to specify a format, given that we can't
possibly foresee all the ways in which this might be used.

What I think we need in a client class identifier (both vendor and user
classes) is simply a way to supply servers (and clients) with a way to
identify which of the possible universes of option sets, or variations
within a [standard] set is [to be] used in replies.  If Mike Henry is
correct about option 43, then option 60 might be very simple indeed:  just a
string like "ISC," "H-P," or "Sun,"  while the versioning information could
be encoded in option 43.  I don't know that there is one best way, but I
think it deserves a wide airing to reach consensus.

I'm unhappy that I seemed to so poorly communicate my ideas, since we now
find ourselves vigorously discussing how much we agree with each other!

See you in DC!

--Barr



From dhcp-v4@bucknell.edu  Fri Nov  5 19:04:46 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA07119
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 19:04:46 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA31584;
	Fri, 5 Nov 1999 19:02:42 -0500 (EST)
Date: Fri, 5 Nov 1999 19:02:42 -0500 (EST)
Message-Id: <D5E932F578EBD111AC3F00A0C96B1E6F04E953F9@orsmsx31.jf.intel.com>
Errors-To: droms@bucknell.edu
Reply-To: mike.henry@intel.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Henry, Mike" <mike.henry@intel.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

Normally I hang on every work that Ted speaks as gospel, but I have to
disagree on this point.  Exactly what defines a "vendor option code space"
embedded in a "vendor option"?  I doubt you mean the DHCP server has some
method of parsing the "vendor option code space" (in any general way) out of
the option code that also contains ad hoc fields of variant information with
no generally defined boundary between the two.

I really don't think you can have it both ways.  If the object is opaque you
cannot embed information. If you embed information it implies the server is
going to use it, or at least is potentially capable of using it. Unless of
course what you mean (and possibly what Barr Hibbs means) is that the option
is primarily for a human to look at. (Also, at the very least, the variant
information should live in defined fields if a computer is going to use the
information.)

I would suggest that Option 60 be used to indicate the user *class* and that
Option 43 be used to indicate *variant* information about that class. This
makes it possible for either a human or a computer to read the information.
This leaves Option 60 as an opaque identifier and the variant information in
Option 43. Use of Option 43 has the further advantage of providing the
structure of defined fields (that's structure, not content), which makes the
content parseable in general. Also, I think this is consistent with what the
spec currently says.

 - Mike Henry

Ted Lemon wrote> 
> Rob, what I described is the way that sun currently does vendor
> classes, roughly (they use dots instead of slashes, and don't send a
> transport version).   A DHCP server that's going to support vendor
> classing needs to be able to say "the following vendor option code
> space works for the following set of possible values of vendor class."
> I do not suggest that Apple implement a completely different set of
> options for each version of OT, or for each model of computer
> supported, but being able to tell the DHCP server what model of
> computer you are is a good thing, IMHO, and I really like the way Sun
> did it.   I'm just suggesting that Apple should follow in Sun's
> footsteps here.
> 
> I don't think that vendor class identifiers can or should have any
> sort of well-defined substructure.   I am just suggesting that one
> particular vendor, in choosing a substructure that *they* will use, do
> it a certain way.
> 



From dhcp-v4@bucknell.edu  Fri Nov  5 19:06:12 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA07590
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 19:06:10 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA30021;
	Fri, 5 Nov 1999 19:04:37 -0500 (EST)
Date: Fri, 5 Nov 1999 19:04:37 -0500 (EST)
Message-Id: <63D30D6E10CFD11190A90000F805FE8601CE6FB9@lespaul.process.com>
Errors-To: droms@bucknell.edu
Reply-To: Gonczi@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Steve Gonczi <Gonczi@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.0.1460.8)
Mime-Version: 1.0

> I believe the server should not interpret the string

The question is, is the string a key into a table of possible
vendor-specific option spaces, and is there just one such key per
vendor, or can the vendor stuff meaningful information into it in the
hopes that the server will allow the information to be used?

I expect one vendor to have many different keys. One per every device that
has some interesting requirements. It is pure optimism to hope that vendors
will follow some intelligent convention we can rely on to pull useful 
encoded information from a key.

sG


From dhcp-v4@bucknell.edu  Fri Nov  5 20:36:01 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA09205
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 20:36:00 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA01496;
	Fri, 5 Nov 1999 20:27:31 -0500 (EST)
Date: Fri, 5 Nov 1999 20:27:31 -0500 (EST)
Message-Id: <D5E932F578EBD111AC3F00A0C96B1E6F04E953FC@orsmsx31.jf.intel.com>
Errors-To: droms@bucknell.edu
Reply-To: mike.henry@intel.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Henry, Mike" <mike.henry@intel.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Possible ambiguity in definition of DHCP client identifier op
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

A long time ago (1997) we submitted a request to IANA for two new options
and a proposal to add a new "type" to option 61.  The new type was to be of
the "UUID" variety.  IANA was confused by the request for an option 61 type
definition, and as we were not making any progress on getting clarification
on the use or meaning of the "type" field within option 61, we proposed, and
were issued, a new UUID Client Identifier option (Option 97). So, as a
result of this interaction with IANA we were issued DHCP Option numbers 93,
94, and 97.

Subsequently we submitted a draft that defined the use of these new options.
This caused two problems: first, the DHCP working group did not like the
options because they felt that the same functionality could be gained
through the combined use of options 60 and 61, and second, it brought to a
head the general discomfort caused by issuing option numbers before defining
or authorizing their use.

We agreed with the second concern but did not agree with the first (i.e. we
felt there were significant problems with using options 60 and 61 as
suggested by the working group), but nevertheless we agreed to phase out use
of options 93, 94, and 97 over a period of six years.  The definition of
these options and the agreement for their phase out is documented in this
ID:
http://search.ietf.org/internet-drafts/draft-dittert-host-sys-char-03.txt

Sometime after this I wrote an ID defining a "UUID" type for Option 61 (this
draft has expired).  This ID was greeted with mixed reviews, and I have to
admit I did not quite understand the reservations.  These reservations are
clearer now that I have followed this particular mail thread.  The  upshot
of this thread seems to be that, in spite of pretty clear text to the
contrary (see Option 61 text below), a significant number of people appear
to be suggesting that Option 61 does not have "types", and that the "type"
field is (or perhaps is not) part of the identifier proper, and even if the
option text says there are types, well.... there aren't. And, even if there
are, no one uses them, so this is more or less the same thing.  Its
difficult to debate this position other than to present the actual text of
the option. (BTW, this ambiguity about "type" was one of our reservations
about using option 61 in the first place to in some way "mean" an identifier
of type UUID.)

Ted's suggestion (below) to use the option number as the first field in the
LDAP entry makes a certain amount of sense if you accept that Option 61 has
no types.  However, if Option 61 were used as literally written (i.e. making
use of the "type" field for "types"), it seems to me there would be no
problem.  The first byte would be the "type of identifier and the rest would
be the identifier proper. UUID would be represented by it's Option 61 "type"
(I suggested 254 in the ID I believe) as would the various ARP types. This
seems simple. What am I missing?

BTW, I do not believe the "type" field should be interpreted as part of the
identifier proper.  Again, I think the Option 61 text is pretty clear on
this point (min length of 2, type field nomenclated differently than the
Identifier field).

 - Mike Henry

----Option 61 text----------

9.14. Client-identifier

   This option is used by DHCP clients to specify their unique
   identifier.  DHCP servers use this value to index their database of
   address bindings.  This value is expected to be unique for all
   clients in an administrative domain.

   Identifiers SHOULD be treated as opaque objects by DHCP servers.

   The client identifier MAY consist of type-value pairs similar to the
   'htype'/'chaddr' fields defined in [3]. For instance, it MAY consist
   of a hardware type and hardware address. In this case the type field
   SHOULD be one of the ARP hardware types defined in STD2 [22].  A
   hardware type of 0 (zero) should be used when the value field
   contains an identifier other than a hardware address (e.g. a fully
   qualified domain name).

   For correct identification of clients, each client's client-
   identifier MUST be unique among the client-identifiers used on the
   subnet to which the client is attached.  Vendors and system
   administrators are responsible for choosing client-identifiers that
   meet this requirement for uniqueness.

   The code for this option is 61, and its minimum length is 2.

   Code   Len   Type  Client-Identifier
   +-----+-----+-----+-----+-----+---
   |  61 |  n  |  t1 |  i1 |  i2 | ...
   +-----+-----+-----+-----+-----+---

> -----Original Message-----
> From: Ted Lemon [mailto:mellon@isc.org]
> Sent: Thursday, November 04, 1999 10:13 AM
> To: Multiple recipients of list
> Subject: Re: Possible ambiguity in definition of DHCP client 
> identifier
> op 
> 
> 
> 
> > If I am reading this right, it means that we always prefix a client
> > identifier
> > with 00: when stored as an LDAP attribute.
> 
> I think you and Bernie are saying the same thing, and I think it's not
> the right thing to do.  What you're saying is that if you use the
> client identifier as the key when you store it in LDAP, the first byte
> of this key is always zero.  If you use the MAC address, the first
> byte is the htype field of the DHCP packet and the subsequent bytes
> are the significant bytes of chaddr.
> 
> What is missing from this is the possibility of having a key that is
> derived from anything other than the client hardware address or the
> client identifier.  And yet we already have identified the UUID as a
> possible alternative option to use as a key, and some people use the
> hostname option as the key as well.  It would be nice to be able to
> represent these cleanly in an LDAP entry.
> 
> So I am proposing that we change the algorithm so that the first byte
> (or two bytes, preferably) are a type code, and subsequent bytes are
> the payload.  So if the client identifier is used as a key, we'd use a
> type code of 61 (btw, Barr and Nathan, you have this as 63 in your
> draft), and the rest of the key would be the payload in the client
> identifier option.  If we use the hardware address, we'd use a type
> code of 0, followed by the htype and hlen bytes of chaddr.  If we use
> the UUID, the type code would be the DHCP option code for UUID (I
> can't remember what it is) and the rest of the key would be the
> payload of the UUID option.  Likewise with the hostname option:
> 
> chaddr:		00 00 htype chaddr [0..hlen]
> client-id:	00 3D option payload
> uuid:		?? ?? option payload
> hostname:	00 12 option payload
> 
> By doing this, we maximize flexibility, at a fairly low cost.
> 
> 			       _MelloN_
> 


From dhcp-v4@bucknell.edu  Fri Nov  5 23:16:25 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA15691
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 23:16:25 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA18491;
	Fri, 5 Nov 1999 23:14:46 -0500 (EST)
Date: Fri, 5 Nov 1999 23:14:46 -0500 (EST)
Message-Id: <199911060331.WAA07811@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> While having the server interpret the identifier and base its response on
> the contents has some appeal, how do we specify a general method that can be
> easily implemented in the universe of clients and servers?

Barr, I think that you have mistaken the context of the suggestion
that I made to the fellow at Apple.   I am most definitely _not_
saying we should write a standard that says Apple must do what I
suggested.   I am not saying we should specify what a server should do
if a vendor uses a format like the one I suggested.

What I am saying is that if Apple were to choose, like sun, to put
hardware identifiers in their vendor class identifiers, it could be
useful.   It causes no harm.   It could be useful.   I think it's good
advice.   I am not proposing a standard.

So I think we are loudly in agreement, but I just want to be clear
that I'm just giving the guy at Apple some advice based on my
experience, not trying to mandate a new standard.   :')

			       _MelloN_


From dhcp-v4@bucknell.edu  Fri Nov  5 23:18:53 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA16915
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 23:18:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA09109;
	Fri, 5 Nov 1999 23:17:25 -0500 (EST)
Date: Fri, 5 Nov 1999 23:17:25 -0500 (EST)
Message-Id: <199911060341.WAA07849@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> Exactly what defines a "vendor option code space"
> embedded in a "vendor option"?

Vendor option code space refers to the codes from zero to 255 that are
embedded in a vendor-specific encapsulated options.  The server should
encode the vendor-specific encapsulated options using a different code
space depending on the value of the vendor class identifier that the
client sends.

> I doubt you mean the DHCP server has some method of parsing the
> "vendor option code space" (in any general way) out of the option
> code that also contains ad hoc fields of variant information with no
> generally defined boundary between the two.

I am saying that the DHCP server administrator might choose to write
some configuration language that parses the vendor classes in use at a
particular site, if he or she chooses to do so.  I already have an
example of this in the ISC DHCP server's manual page using Sun
options, and I imagine Sun's DHCP server has some way of doing it too.

> If the object is opaque you cannot embed information. If you embed
> information it implies the server is going to use it, or at least is
> potentially capable of using it.

Right, in a completely ad hoc way.   Making it readable for humans
could come in handy too.

> I would suggest that Option 60 be used to indicate the user *class*
> and that Option 43 be used to indicate *variant* information about
> that class.

This sounds reasonable in theory, but it's a big pain to implement on
the client side if the client side doesn't actually need to send
encapsulated options otherwise.   Just sending what, to the client, is
a fixed string in the vendor class identifier is a lot easier.
Remember, I am not proposing that we specify this - the message I sent
that has attracted so much criticism is just advice for what I, the
author of a DHCP server who has dealt a _lot_ with real-world users,
think Apple, a vendor of a DHCP client, should do.   I am not saying
we should require such a thing of any vendor, including Apple - I'm
just suggesting.

			       _MelloN_


From dhcp-v4@bucknell.edu  Fri Nov  5 23:20:36 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA17769
	for <dhc-archive@odin.ietf.org>; Fri, 5 Nov 1999 23:20:36 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA20173;
	Fri, 5 Nov 1999 23:19:06 -0500 (EST)
Date: Fri, 5 Nov 1999 23:19:06 -0500 (EST)
Message-Id: <199911060343.WAA07859@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> I expect one vendor to have many different keys. One per every device that
> has some interesting requirements. It is pure optimism to hope that vendors
> will follow some intelligent convention we can rely on to pull useful 
> encoded information from a key.

Yup.   Although there's nothing wrong with being optimistic.

			       _MelloN_


From dhcp-v6@bucknell.edu  Sat Nov  6 06:14:42 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id GAA09429
	for <dhc-archive@odin.ietf.org>; Sat, 6 Nov 1999 06:14:41 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id GAA10610;
	Sat, 6 Nov 1999 06:06:57 -0500 (EST)
Date: Sat, 6 Nov 1999 06:06:57 -0500 (EST)
Message-Id: <199911061053.FAA11478@uplink.net>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Agenda (w/o line wraps)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

Dynamic host configuration : DHCP reconfigure extension Yves T'Joens (Alcatel)
draft-schrijvp-dhcpv4-reconfigure-00.txt

Authentication for DHCP Messages                        Bill Arbaugh (Penn)
draft-ietf-dhc-authentication-12.txt

The Name Service Search Option for DHCP                 Carl Smith (Sun)
draft-ietf-dhc-nsso-01.txt

DHCP Failover Protocol                                  Kim Kinnear (Cisco)
draft-ietf-dhc-failover-05.txt

The User Class Option for DHCP                          Jerome Privat (BT)
draft-ietf-dhc-userclass-04.txt

Double Phase DHCP Configuration                         Jerome Privat (BT)
draft-privat-dhc-doublephase-01.txt

DHCP Relay Agent Information Option                     Mike Patrick (Motorola)
draft-ietf-dhc-agent-options-07.txt

Dynamic Registration and Configuration Protocol (DRCP)  Subir Das (Telcordia)
draft-itsumo-drcp-00.txt

DHCP Authentication Via Kerberos V                      Bernard Aboba (Microsoft)
draft-hornstein-dhc-kerbauth-01.txt

DHC load balancing algorithm                            Bernie Volz (Process Software)
draft-ietf-dhc-loadb-00.txt

The Server Selection Option for DHCP                    Bill Sommerfield (Epilogue)
draft-ietf-dhc-sso-03.txt

Dynamic Host Configuration Protocol (DHCP) Server MIB   Barr Hibbs (Pacbell)
draft-ietf-dhc-server-mib-04.txt

Interpreting Client Options for DHCP                    Barr Hibbs (Pacbell)
draft-ietf-dhc-client-options-00.txt

Option for NAT/IPSec interaction                        Jose Brustoloni (Bell Labs)


Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================

DHCP Schema for LDAP                                    Bernie Volz (Process Software)
draft-ietf-dhc-schema-01.txt

A DNS RR for DHCP information                           Andreas Gustafsson (Internet Engines
draft-ietf-dnsind-dhcp-rr-00.txt

Interaction between DHCP and DNS                        Mark Stapp (Cisco)
draft-ietf-dhc-dhcp-dns-11.txt

New Option Review Guidelines and                        Mike Carney (Sun)
  Additional Option Namespace
draft-ietf-dhc-option-review-and-namespace-01.txt

Connectathon                                            Mike Carney (Sun)


Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
Dynamic Host Configuration Protocol for IPv6 (DHCPv6)   Mike Carney (Sun)
draft-ietf-dhc-dhcpv6-14.txt                            Jim  Bound (Compaq)
Extensions for the DHCP for IPv6                        Charlie Perkins (Nokia)    
draft-ietf-dhc-v6exts-11.txt




From dhcp-v4@bucknell.edu  Sat Nov  6 06:18:35 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id GAA10999
	for <dhc-archive@odin.ietf.org>; Sat, 6 Nov 1999 06:18:34 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id GAA28403;
	Sat, 6 Nov 1999 06:16:59 -0500 (EST)
Date: Sat, 6 Nov 1999 06:16:59 -0500 (EST)
Message-Id: <199911061053.FAA11478@uplink.net>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Agenda (w/o line wraps)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


Tuesday, Nov 9 at 0900-1130 (DHCPv4)
====================================

Dynamic host configuration : DHCP reconfigure extension Yves T'Joens (Alcatel)
draft-schrijvp-dhcpv4-reconfigure-00.txt

Authentication for DHCP Messages                        Bill Arbaugh (Penn)
draft-ietf-dhc-authentication-12.txt

The Name Service Search Option for DHCP                 Carl Smith (Sun)
draft-ietf-dhc-nsso-01.txt

DHCP Failover Protocol                                  Kim Kinnear (Cisco)
draft-ietf-dhc-failover-05.txt

The User Class Option for DHCP                          Jerome Privat (BT)
draft-ietf-dhc-userclass-04.txt

Double Phase DHCP Configuration                         Jerome Privat (BT)
draft-privat-dhc-doublephase-01.txt

DHCP Relay Agent Information Option                     Mike Patrick (Motorola)
draft-ietf-dhc-agent-options-07.txt

Dynamic Registration and Configuration Protocol (DRCP)  Subir Das (Telcordia)
draft-itsumo-drcp-00.txt

DHCP Authentication Via Kerberos V                      Bernard Aboba (Microsoft)
draft-hornstein-dhc-kerbauth-01.txt

DHC load balancing algorithm                            Bernie Volz (Process Software)
draft-ietf-dhc-loadb-00.txt

The Server Selection Option for DHCP                    Bill Sommerfield (Epilogue)
draft-ietf-dhc-sso-03.txt

Dynamic Host Configuration Protocol (DHCP) Server MIB   Barr Hibbs (Pacbell)
draft-ietf-dhc-server-mib-04.txt

Interpreting Client Options for DHCP                    Barr Hibbs (Pacbell)
draft-ietf-dhc-client-options-00.txt

Option for NAT/IPSec interaction                        Jose Brustoloni (Bell Labs)


Tuesday, Nov 9 at 1300-1400 (DHCPv4)
====================================

DHCP Schema for LDAP                                    Bernie Volz (Process Software)
draft-ietf-dhc-schema-01.txt

A DNS RR for DHCP information                           Andreas Gustafsson (Internet Engines
draft-ietf-dnsind-dhcp-rr-00.txt

Interaction between DHCP and DNS                        Mark Stapp (Cisco)
draft-ietf-dhc-dhcp-dns-11.txt

New Option Review Guidelines and                        Mike Carney (Sun)
  Additional Option Namespace
draft-ietf-dhc-option-review-and-namespace-01.txt

Connectathon                                            Mike Carney (Sun)


Tuesday, Nov 9 at 1415-1515 (DHCPv6)
====================================
Dynamic Host Configuration Protocol for IPv6 (DHCPv6)   Mike Carney (Sun)
draft-ietf-dhc-dhcpv6-14.txt                            Jim  Bound (Compaq)
Extensions for the DHCP for IPv6                        Charlie Perkins (Nokia)    
draft-ietf-dhc-v6exts-11.txt




From dhcp-v4@bucknell.edu  Sat Nov  6 08:26:53 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id IAA02218
	for <dhc-archive@odin.ietf.org>; Sat, 6 Nov 1999 08:26:52 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA26109;
	Sat, 6 Nov 1999 08:25:14 -0500 (EST)
Date: Sat, 6 Nov 1999 08:25:14 -0500 (EST)
Message-Id: <199911061323.IAA09905@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Agenda (w/o line wraps) 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


Ralph,

I need an agenda that's encoded in Unicode.   Can you send one of
those too?   :')

			       _MelloN_


From dhcp-v4@bucknell.edu  Sat Nov  6 15:45:42 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA25816
	for <dhc-archive@odin.ietf.org>; Sat, 6 Nov 1999 15:45:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA10676;
	Sat, 6 Nov 1999 15:45:00 -0500 (EST)
Date: Sat, 6 Nov 1999 15:45:00 -0500 (EST)
Message-Id: <199911062025.MAA09730@luna.join.com>
Errors-To: droms@bucknell.edu
Reply-To: robs@join.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Rob Stevens <robs@join.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Agenda (w/o line wraps)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4

Yes,

I'd like mine in HTML with links pointing back to minues
from the Oslo IETF.  :')


From dhcp-v4@bucknell.edu  Sat Nov  6 16:57:09 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA22333
	for <dhc-archive@odin.ietf.org>; Sat, 6 Nov 1999 16:57:08 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA10011;
	Sat, 6 Nov 1999 16:56:38 -0500 (EST)
Date: Sat, 6 Nov 1999 16:56:38 -0500 (EST)
Message-Id: <199911062131.QAA06180@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Agenda (w/o line wraps)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Awright - anybody else wanna give me some smart-ass abuse?  Take a number
and get in line...

- Ralph <with a grin>

At 03:49 PM 11/6/99 -0500, you wrote:
>Yes,
>
>I'd like mine in HTML with links pointing back to minues
>from the Oslo IETF.  :')
> 


From dhcp-v4@bucknell.edu  Sun Nov  7 03:24:47 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id DAA02944
	for <dhc-archive@odin.ietf.org>; Sun, 7 Nov 1999 03:24:47 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id DAA12109;
	Sun, 7 Nov 1999 03:24:23 -0500 (EST)
Date: Sun, 7 Nov 1999 03:24:23 -0500 (EST)
Message-Id: <CF0A34F56241D211987F00805F8BA3974A21C6@dxbitcex01.hq.emirates.com>
Errors-To: droms@bucknell.edu
Reply-To: iqbalbm@emirates.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "M. Iqbal Bari" <iqbalbm@emirates.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

unsubscribe dhcp-v4
		


From dhcp-v4@bucknell.edu  Sun Nov  7 13:06:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA15732
	for <dhc-archive@odin.ietf.org>; Sun, 7 Nov 1999 13:06:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA32605;
	Sun, 7 Nov 1999 13:05:34 -0500 (EST)
Date: Sun, 7 Nov 1999 13:05:34 -0500 (EST)
Message-Id: <000801bf2940$baa00c00$ad7731cc@pro.bsc.net>
Errors-To: droms@bucknell.edu
Reply-To: pro@bsc.net
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "pro" <pro@bsc.net>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: Microsoft Outlook Express 4.72.3110.5
Content-Transfer-Encoding: 7bit
Content-Type: text/plain;
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

unsubscribe dhcp-v4




From dhcp-v4@bucknell.edu  Sun Nov  7 23:55:24 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA03686
	for <dhc-archive@odin.ietf.org>; Sun, 7 Nov 1999 23:55:24 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA12434;
	Sun, 7 Nov 1999 23:53:15 -0500 (EST)
Date: Sun, 7 Nov 1999 23:53:15 -0500 (EST)
Message-Id: <49256823.00181F0D.00@d22mta10.yamato.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: TOSHIS@jp.ibm.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Toshimasa Shimizu" <TOSHIS@jp.ibm.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0



unsubscribe dhcp-v4




From dhcp-v4@bucknell.edu  Mon Nov  8 00:35:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id AAA19812
	for <dhc-archive@odin.ietf.org>; Mon, 8 Nov 1999 00:35:36 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id AAA06183;
	Mon, 8 Nov 1999 00:34:23 -0500 (EST)
Date: Mon, 8 Nov 1999 00:34:23 -0500 (EST)
Message-Id: <004101bf29a6$1c31d100$0100a8c0@wrinkles>
Errors-To: droms@bucknell.edu
Reply-To: dmisdwb@mail.futuresouth.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Daniel Byrne" <dmisdwb@mail.futuresouth.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: Microsoft Outlook Express 5.00.2314.1300
Content-Transfer-Encoding: 7bit
Content-Type: text/plain;
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

 unsubscribe dhcp-v4




From dhcp-v4@bucknell.edu  Mon Nov  8 08:41:55 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id IAA10677
	for <dhc-archive@odin.ietf.org>; Mon, 8 Nov 1999 08:41:54 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA07173;
	Mon, 8 Nov 1999 08:41:09 -0500 (EST)
Date: Mon, 8 Nov 1999 08:41:09 -0500 (EST)
Message-Id: <3.0.6.32.19991108075620.009108f0@hawk.anselm.edu>
Errors-To: droms@bucknell.edu
Reply-To: dvigneau@anselm.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: David Vigneau <dvigneau@anselm.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Light Version 3.0.6 (32)

unsubscribe dhcp-v4

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  David Vigneau                            Network Systems Manager
  dvigneau@anselm.edu                         Saint Anselm College
  100 Saint Anselm Drive                      Manchester, NH 03102
  http://www.anselm.edu                       work: (603) 641-7129
                                               fax: (603) 641-7288
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


From dhcp-v6@bucknell.edu  Mon Nov  8 13:39:11 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA17561
	for <dhc-archive@odin.ietf.org>; Mon, 8 Nov 1999 13:39:10 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA18926;
	Mon, 8 Nov 1999 13:33:30 -0500 (EST)
Date: Mon, 8 Nov 1999 13:33:30 -0500 (EST)
Message-Id: <199911081722.MAA19730@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Leaving the mailing lists
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

I've seen a flurry of requests to unsubscribe from the dhcp-v4@bucknell.edu
and dhcp-v6@bucknell.edu mailing lists (some of these requests have leaked
through to the whole list).  Please send all admin requests, including
requests to unsubscribe from one of the lists, to listserv@bucknell.edu; e.g.:

unusbscribe dhcp-v4

- Ralph Droms



From dhcp-v4@bucknell.edu  Mon Nov  8 13:40:53 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA18254
	for <dhc-archive@odin.ietf.org>; Mon, 8 Nov 1999 13:40:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA30977;
	Mon, 8 Nov 1999 13:39:15 -0500 (EST)
Date: Mon, 8 Nov 1999 13:39:15 -0500 (EST)
Message-Id: <195339D90D04D3118B0E0090274EFB320129C7DD@msgcreo4.creo.bc.ca>
Errors-To: droms@bucknell.edu
Reply-To: PBaker@creo.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Peter Baker <PBaker@creo.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0

unsubscribe dhcp-v4




From dhcp-v4@bucknell.edu  Mon Nov  8 13:46:16 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA20456
	for <dhc-archive@odin.ietf.org>; Mon, 8 Nov 1999 13:46:15 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA05798;
	Mon, 8 Nov 1999 13:45:16 -0500 (EST)
Date: Mon, 8 Nov 1999 13:45:16 -0500 (EST)
Message-Id: <199911081722.MAA19730@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Leaving the mailing lists
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

I've seen a flurry of requests to unsubscribe from the dhcp-v4@bucknell.edu
and dhcp-v6@bucknell.edu mailing lists (some of these requests have leaked
through to the whole list).  Please send all admin requests, including
requests to unsubscribe from one of the lists, to listserv@bucknell.edu; e.g.:

unusbscribe dhcp-v4

- Ralph Droms



From dhcp-v4@bucknell.edu  Mon Nov  8 21:31:28 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA21485
	for <dhc-archive@odin.ietf.org>; Mon, 8 Nov 1999 21:31:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA23189;
	Mon, 8 Nov 1999 21:29:50 -0500 (EST)
Date: Mon, 8 Nov 1999 21:29:50 -0500 (EST)
Message-Id: <Pine.GSO.3.95.991108211453.2145A-100000@codex.cis.upenn.edu>
Errors-To: droms@bucknell.edu
Reply-To: waa@dsl.cis.upenn.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "William A. Arbaugh" <waa@dsl.cis.upenn.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: Authentication draft
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; charset=US-ASCII
Mime-Version: 1.0


> Per Ralph's request, here are some comments on the 
> http://www.ietf.org/internet-drafts/draft-ietf-dhc-authentication-12.txt.
> Most of these are more editorial type issues.
> 
	I will make the appropriate changes.

> 
> 4. In Appendix A, the key is generated using the "subnet address". Do you
> mean the [proposed] address for the client or the SUBNET address (ie,
> client's address ANDed with subnet mask).
> 
I didn't write this part of the draft, but I read it to mean the IP subnet
address, e.g. 192.168.1.0.  In my opinion, this is problematic. It may be
that we can get by with just the client identifier.  We can talk about
tomorrow in the meeting.

Bill



From dhcp-v6@bucknell.edu  Mon Nov  8 22:09:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA09410
	for <dhc-archive@odin.ietf.org>; Mon, 8 Nov 1999 22:09:02 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA05376;
	Mon, 8 Nov 1999 22:05:37 -0500 (EST)
Date: Mon, 8 Nov 1999 22:05:37 -0500 (EST)
Message-Id: <3829B36E.7EC6CE7D@east.sun.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Michael W. Carney" <michael.carney@east.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: List of DHCPv6 Issues with Responses, Part 2 of 2
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: multipart/mixed;
Mime-Version: 1.0
X-Mailer: Mozilla 4.6 [en] (X11; I; Linux 2.2.5-22 i686)

This is a multi-part message in MIME format.
--------------E33AC0E7F7964B030C3C010D
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Hi folks,
    Here is the balance of the issues with our responses for the input
we've received. We'll be discussing these tomorrow (along with the List
of DHCPv6 Issues with Responses, Part 1 of 2) during the DHCPv6
meeting tomorrow afternoon.

Enjoy,

--

Mike Carney



--------------E33AC0E7F7964B030C3C010D
Content-Type: text/plain; charset=us-ascii;
 name="11_08_99-issues+responses.txt"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="11_08_99-issues+responses.txt"

					bound/mwc/perkins	11/08/99

		List of DHCPv6 Issues with responses, part 2 of 2
		=================================================

DHCPv4 disadvantages to avoid:
==============================

	*	Lack of DHCP protocol versioning in original
		specification.

				We're adding a DHCPv6 extension for
				protocol version to the extensions
				draft.

	*	Changes to options in most cases necessitate changes to
		client and server implementations. Can we avoid this by
		carefully specifying suggested generic encoding
		mechanisms?

		Response:	Good point. We'll add text to the
				extensions document suggesting handling
				of data of different types. Note also
				that we'll draw in what was learned from
				the DHCPv4 futures panel regarding the
				review of options/extensions. This
				should help extension authors and
				the DHC community avoid unnecessary
				changes to implementations.

	*	Confusion regarding minimum packet size. Some say BOOTP
		size (300bytes) is the minimum, whereas others say the
		minimum is the fixed part of the BOOTP header.

		Response:	Appendix B notes that the IPv6 MTU is
				used, potential learned through Path MTU
				discovery. A client can signal to the
				agent its DHCPv6 MTU through the use of
				the Maximum DHCPv6 Message Size
				Extension. The client will learn the PMTU
				from the relay if necessary (DHCP relay
				ICMP error message).

	*	Change in RFC 1541 -> RFC 2131 INIT-REBOOT DHCPREQUEST
		format was not backward compatible.

		Response:	Adding a DHCPv6 protocol version
				extension discussed above will enable
				us to deal appropriately with
				incompatible changes in the future.


DHCPv4 advantages, DHCPv4-based questions
=========================================

	*	State transition diagram clarifies client / server
		interaction.

		Response:	We will produce client and server
				implementation state transition
				diagrams for DHCPv6 for inclusion in the
				appendix.

	*	Client version

		Response:	We're adding a DHCPv6 protocol version
				extension to carry the protocol version.
				Client implementation versions are
				handled by the implementors.

	*	DHCP server load balancing
		
		Response:	We will widen the specification to allow
				relay agents to relay to unicast and any
				other specified multicast address (not
				just the DHCP ones). We also suggest
				that the implementor of a server could
				use the preference field to signal when
				it was getting overloaded.


DHCPv6 draft issues
===================


	*	Consider merging two documents into one

		Response:	We plan on leaving the extensions
				definitions in the extensions draft. We
				will improve the protocol draft by
				adding more references to the extension
				draft where protocol-related extensions
				are discussed.

	*	What is the process for defining, reviewing, and
		acceptance of new extensions? Particularly when there are
		DHCPv4 options for the same item?

		Response:	We plan on incorporating the findings of
				the futures panel such that these
				processes are described where
				appropriate in the protocol document and
				the extensions document.

	*	There isn't a clear state machine defined.

		Response:	We will be adding client and server
				implementation state diagrams to the
				appendix of the protocol draft.

	*	No background explanations for complicated message
		exchanges.

		Response:	As part of clarifying the document,
				we'll expand the explanations to the
				WG's satisfaction.

	*	General lack of background for complication (seemingly)
		in protocol itself.

		Response:	We got no specific guidance from the WG
				on this point. We will add an
				introductory 'flow' section in an
				attempt to make the protocol clearer.
				There will also be client and server
				implementation state diagrams in the
				appendix.

	*	How does a client request addresses from multiple
		prefixes?

		Response:	We have two possibilities:

			     1) We will add a new "scope" field to the
				IP address extension for use by clients
				to request IP addresses of specific
				scope. The scope field consumes 4 bits
				of the reserved field as follows:

				The high order bit determines whether an
				address matching the scope is requested
				(0) or required (1). The following 3
				bits form a number (0-7) which have the
				following meanings:

				0	Globally scoped address
				1	Site local address
				2	reserved
				3	reserved
				4	reserved
				5	reserved
				6	reserved
				7	reserved

			     2) Produce an additional optional field in
			        the IP address extension called the
				scope field. A new bit (Scope=S) would
				be set when the scope field is present.
				This field carries a prefix
				specification.  Servers would use it in
				IP address extensions in advertise
				messages to communicate which prefixes
				were available and clients can use it in
				requests.

				Comments?
				
	*	How does a client request additional IP addresses? The
		specs aren't at all clear on how this would be done.

		Response:	When a client sends a DHCP request with
				the 'C' bit set, the client has
				'amnesia', and therefore requests the
				server to free up any allocated
				releasable resources. Thus, all IP
				addresses assigned to the client are
				released. Any IP address extensions are
				processed using #a and #c below. When a
				client sends a DHCP request without the
				'C' bit set, it is seeking one of the
				following, based upon the contents of
				the IP address extension:

				a) Requesting a particular IP address.

				b) Requesting an extension on the
				lifetimes of a existing address.

				c) Requesting a new IP address. Note
				that by using the proposed scope field
				in the IP address extension, the client
				can request exactly what prefix the
				client would like the IP address
				allocated from.

				If multiple IP address extensions are
				included in a request w/o the C bit set,
				the above possibilites (a-c) are
				considered by the server for each
				extension.


	*	Does the RECONFIGURE mechanism scale wrt ACK
		implosion? New RECONFIGURE reduces messages from 2N to
		N, but N may still be large.

		Response:	We believe our current mechanism is
				adequate, particularly since we haven't
				received an alternate suggested
				mechanism we find superior to the
				existing one.

	*	No concept of T1/T2 times (or % of lifetimes) for client
		use when deciding when and how often to contact server
		to extend various lifetimes for certain extensions, of
		which IP address is one such extension.

		Response:	We'll clarify the text such that a
				client requests a lifetime extension
				when an address becomes deprecated.

	*	Affect that 'A' and 'Q' bit (MUST update AAAA record
		before responding) has on retransmission not clear.

		Response:	We'll add text discussing the impact of
				that long-lived dns update transactions
				have on the client / server interaction.

	*	DHCPv6 is prone to denial of service attacks like the
		IPv4 icmp router discovery attack.

		Response:	To reduce the scope of such attacks,
				we'll change the text to specify that
				link local messages must set the hop
				limit to 255, which will limit a
				potential attack to the single link.

	*	Some extensions can be specified multiple times, whereas
		others cannot. Which alternative isn't always clear in
		the extensions draft.

		Response:	Add text to make extensions clear on
				this point.

	*	Release message and failover environments. Would be nice
		to permit messages to be sent to multicast address(es)
		from clients, thus any server will see the release.

		Response:	Allowing this prevents DHCPv6
				authentication from being used (an
				authentication header for a multicast
				address?)

	*	Can you map in the DHCPv4 option space into the DHCPv6
		extension space? This would simplify the LDAP schema.

		Response:	v4 and v6 are different protocols, and
				many of the options (which are even
				useless in v4 land) make no sense in a
				v6 environment.

	*	The reconfigure message semantics (N bit) are confusing;
		what about adding new messages to make it clearer?

		Response:	Background: A reconfigure with the N bit
				set contains any information a client
				needs for the reconfiguration. A
				reconfigure w/o the N bit set is used to
				inform the client to redo the
				request/reply exchange with the server.
				The reconfigure message in this case
				does not contain all of the necessary
				data for a reconfiguration.

				We will change the specs as follows: The
				reconfigure message maintains its
				current form (including the N bit).
				However, we will define a new client 
				message, the RECONFIGURE_REPLY message,
				which a client will use in response to a
				reconfigure message with the N bit set.
				This avoids overloading the DHCP request
				message as a "reconfigure reply"
				message, which has been a source of
				confusion within the WG.

	*	Setting address lifetimes to zero (in either
		reconfiguration messages or replies) is quite a denial of
		service attack. How is DHCPv6 going to deal with this?

		Response:	We'll change the spec so that it follows
				an algorithm similar to that in  Section 5.3
				of RFC 2462 (addrconf). If a lifetime is
				less than 2 hrs is received from an
				unauthenticated DHCP server, then a
				lifetime of 2 hrs is assumed. If a
				lifetime of less than 2 hrs is recieved
				from an authenticated DHCP server, then
				that lifetime is accepted.

--------------E33AC0E7F7964B030C3C010D--

-------


From dhcp-v4@bucknell.edu  Tue Nov  9 10:09:45 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA07385
	for <dhc-archive@odin.ietf.org>; Tue, 9 Nov 1999 10:09:40 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA21929;
	Tue, 9 Nov 1999 10:08:34 -0500 (EST)
Date: Tue, 9 Nov 1999 10:08:34 -0500 (EST)
Message-Id: <85256824.004F77C4.00@ismmta01.desjardins.com>
Errors-To: droms@bucknell.edu
Reply-To: Robert.Stringer@CCPEDQ.desjardins.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Robert.Stringer@CCPEDQ.desjardins.com
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0




unsubscribe dhcp-v4




From dhcp-v4@bucknell.edu  Wed Nov 10 17:48:22 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA02230
	for <dhc-archive@odin.ietf.org>; Wed, 10 Nov 1999 17:48:22 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA10801;
	Wed, 10 Nov 1999 17:47:33 -0500 (EST)
Date: Wed, 10 Nov 1999 17:47:33 -0500 (EST)
Message-Id: <009e01bf2bc7$1bc2c9e0$c9053b81@dycus.vanderbilt.edu.>
Errors-To: droms@bucknell.edu
Reply-To: mark.dycus@vanderbilt.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Mark Dycus" <mark.dycus@vanderbilt.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Utility to check if DHCP server is assigning addresses.
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: Microsoft Outlook Express 4.72.3110.5
Content-Transfer-Encoding: 7bit
Content-Type: text/plain;
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

I am looking for a utility which will check to see if 
a DHCP server is assigning addresses. 
I need this utility to be able setup something to
notify our NOC if there is a problem with the DHCP
server.  

Thanks

Mark Dycus
Network Engineer
Vanderbilt University



From dhcp-v6@bucknell.edu  Wed Nov 10 18:55:53 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA04931
	for <dhc-archive@odin.ietf.org>; Wed, 10 Nov 1999 18:55:52 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA15816;
	Wed, 10 Nov 1999 18:51:57 -0500 (EST)
Date: Wed, 10 Nov 1999 18:51:57 -0500 (EST)
Message-Id: <199911102306.SAA11436@atlantic.East.Sun.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: List of DHCPv6 Issues with Responses, Part 2 of 2
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Hi folks,
	Try again, mailer choked.
		

					bound/mwc/perkins	11/08/99

		List of DHCPv6 Issues with responses, part 2 of 2
		=================================================

DHCPv4 disadvantages to avoid:
==============================

	*	Lack of DHCP protocol versioning in original
		specification.

				We're adding a DHCPv6 extension for
				protocol version to the extensions
				draft.

	*	Changes to options in most cases necessitate changes to
		client and server implementations. Can we avoid this by
		carefully specifying suggested generic encoding
		mechanisms?

		Response:	Good point. We'll add text to the
				extensions document suggesting handling
				of data of different types. Note also
				that we'll draw in what was learned from
				the DHCPv4 futures panel regarding the
				review of options/extensions. This
				should help extension authors and
				the DHC community avoid unnecessary
				changes to implementations.

	*	Confusion regarding minimum packet size. Some say BOOTP
		size (300bytes) is the minimum, whereas others say the
		minimum is the fixed part of the BOOTP header.

		Response:	Appendix B notes that the IPv6 MTU is
				used, potential learned through Path MTU
				discovery. A client can signal to the
				agent its DHCPv6 MTU through the use of
				the Maximum DHCPv6 Message Size
				Extension. The client will learn the PMTU
				from the relay if necessary (DHCP relay
				ICMP error message).

	*	Change in RFC 1541 -> RFC 2131 INIT-REBOOT DHCPREQUEST
		format was not backward compatible.

		Response:	Adding a DHCPv6 protocol version
				extension discussed above will enable
				us to deal appropriately with
				incompatible changes in the future.


DHCPv4 advantages, DHCPv4-based questions
=========================================

	*	State transition diagram clarifies client / server
		interaction.

		Response:	We will produce client and server
				implementation state transition
				diagrams for DHCPv6 for inclusion in the
				appendix.

	*	Client version

		Response:	We're adding a DHCPv6 protocol version
				extension to carry the protocol version.
				Client implementation versions are
				handled by the implementors.

	*	DHCP server load balancing
		
		Response:	We will widen the specification to allow
				relay agents to relay to unicast and any
				other specified multicast address (not
				just the DHCP ones). We also suggest
				that the implementor of a server could
				use the preference field to signal when
				it was getting overloaded.


DHCPv6 draft issues
===================


	*	Consider merging two documents into one

		Response:	We plan on leaving the extensions
				definitions in the extensions draft. We
				will improve the protocol draft by
				adding more references to the extension
				draft where protocol-related extensions
				are discussed.

	*	What is the process for defining, reviewing, and
		acceptance of new extensions? Particularly when there are
		DHCPv4 options for the same item?

		Response:	We plan on incorporating the findings of
				the futures panel such that these
				processes are described where
				appropriate in the protocol document and
				the extensions document.

	*	There isn't a clear state machine defined.

		Response:	We will be adding client and server
				implementation state diagrams to the
				appendix of the protocol draft.

	*	No background explanations for complicated message
		exchanges.

		Response:	As part of clarifying the document,
				we'll expand the explanations to the
				WG's satisfaction.

	*	General lack of background for complication (seemingly)
		in protocol itself.

		Response:	We got no specific guidance from the WG
				on this point. We will add an
				introductory 'flow' section in an
				attempt to make the protocol clearer.
				There will also be client and server
				implementation state diagrams in the
				appendix.

	*	How does a client request addresses from multiple
		prefixes?

		Response:	We have two possibilities:

			     1) We will add a new "scope" field to the
				IP address extension for use by clients
				to request IP addresses of specific
				scope. The scope field consumes 4 bits
				of the reserved field as follows:

				The high order bit determines whether an
				address matching the scope is requested
				(0) or required (1). The following 3
				bits form a number (0-7) which have the
				following meanings:

				0	Globally scoped address
				1	Site local address
				2	reserved
				3	reserved
				4	reserved
				5	reserved
				6	reserved
				7	reserved

			     2) Produce an additional optional field in
			        the IP address extension called the
				scope field. A new bit (Scope=S) would
				be set when the scope field is present.
				This field carries a prefix
				specification.  Servers would use it in
				IP address extensions in advertise
				messages to communicate which prefixes
				were available and clients can use it in
				requests.

				Comments?
				
	*	How does a client request additional IP addresses? The
		specs aren't at all clear on how this would be done.

		Response:	When a client sends a DHCP request with
				the 'C' bit set, the client has
				'amnesia', and therefore requests the
				server to free up any allocated
				releasable resources. Thus, all IP
				addresses assigned to the client are
				released. Any IP address extensions are
				processed using #a and #c below. When a
				client sends a DHCP request without the
				'C' bit set, it is seeking one of the
				following, based upon the contents of
				the IP address extension:

				a) Requesting a particular IP address.

				b) Requesting an extension on the
				lifetimes of a existing address.

				c) Requesting a new IP address. Note
				that by using the proposed scope field
				in the IP address extension, the client
				can request exactly what prefix the
				client would like the IP address
				allocated from.

				If multiple IP address extensions are
				included in a request w/o the C bit set,
				the above possibilites (a-c) are
				considered by the server for each
				extension.


	*	Does the RECONFIGURE mechanism scale wrt ACK
		implosion? New RECONFIGURE reduces messages from 2N to
		N, but N may still be large.

		Response:	We believe our current mechanism is
				adequate, particularly since we haven't
				received an alternate suggested
				mechanism we find superior to the
				existing one.

	*	No concept of T1/T2 times (or % of lifetimes) for client
		use when deciding when and how often to contact server
		to extend various lifetimes for certain extensions, of
		which IP address is one such extension.

		Response:	We'll clarify the text such that a
				client requests a lifetime extension
				when an address becomes deprecated.

	*	Affect that 'A' and 'Q' bit (MUST update AAAA record
		before responding) has on retransmission not clear.

		Response:	We'll add text discussing the impact of
				that long-lived dns update transactions
				have on the client / server interaction.

	*	DHCPv6 is prone to denial of service attacks like the
		IPv4 icmp router discovery attack.

		Response:	To reduce the scope of such attacks,
				we'll change the text to specify that
				link local messages must set the hop
				limit to 255, which will limit a
				potential attack to the single link.

	*	Some extensions can be specified multiple times, whereas
		others cannot. Which alternative isn't always clear in
		the extensions draft.

		Response:	Add text to make extensions clear on
				this point.

	*	Release message and failover environments. Would be nice
		to permit messages to be sent to multicast address(es)
		from clients, thus any server will see the release.

		Response:	Allowing this prevents DHCPv6
				authentication from being used (an
				authentication header for a multicast
				address?)

	*	Can you map in the DHCPv4 option space into the DHCPv6
		extension space? This would simplify the LDAP schema.

		Response:	v4 and v6 are different protocols, and
				many of the options (which are even
				useless in v4 land) make no sense in a
				v6 environment.

	*	The reconfigure message semantics (N bit) are confusing;
		what about adding new messages to make it clearer?

		Response:	Background: A reconfigure with the N bit
				set contains any information a client
				needs for the reconfiguration. A
				reconfigure w/o the N bit set is used to
				inform the client to redo the
				request/reply exchange with the server.
				The reconfigure message in this case
				does not contain all of the necessary
				data for a reconfiguration.

				We will change the specs as follows: The
				reconfigure message maintains its
				current form (including the N bit).
				However, we will define a new client 
				message, the RECONFIGURE_REPLY message,
				which a client will use in response to a
				reconfigure message with the N bit set.
				This avoids overloading the DHCP request
				message as a "reconfigure reply"
				message, which has been a source of
				confusion within the WG.

	*	Setting address lifetimes to zero (in either
		reconfiguration messages or replies) is quite a denial of
		service attack. How is DHCPv6 going to deal with this?

		Response:	We'll change the spec so that it follows
				an algorithm similar to that in  Section 5.3
				of RFC 2462 (addrconf). If a lifetime is
				less than 2 hrs is received from an
				unauthenticated DHCP server, then a
				lifetime of 2 hrs is assumed. If a
				lifetime of less than 2 hrs is recieved
				from an authenticated DHCP server, then
				that lifetime is accepted.


From dhcp-v4@bucknell.edu  Wed Nov 10 20:51:51 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA24899
	for <dhc-archive@odin.ietf.org>; Wed, 10 Nov 1999 20:51:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA29117;
	Wed, 10 Nov 1999 20:51:25 -0500 (EST)
Date: Wed, 10 Nov 1999 20:51:25 -0500 (EST)
Message-Id: <199911110125.UAA09772@uplink.net>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Utility to check if DHCP server is assigning addresses.
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 05:56 PM 11/10/99 -0500, Mark Dycus wrote:
>I am looking for a utility which will check to see if 
>a DHCP server is assigning addresses. 

Check out http://www.eg.bucknell.edu/~jgoldsch/dhcp

- Ralph Droms





From dhcp-v4@bucknell.edu  Fri Nov 12 04:29:55 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id EAA27859
	for <dhc-archive@odin.ietf.org>; Fri, 12 Nov 1999 04:29:55 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id EAA19521;
	Fri, 12 Nov 1999 04:28:50 -0500 (EST)
Date: Fri, 12 Nov 1999 04:28:50 -0500 (EST)
Message-Id: <s82b77bf.021@prv-mail20.provo.novell.com>
Errors-To: droms@bucknell.edu
Reply-To: KNirupama@novell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Nirupma Kulshrestha" <KNirupama@novell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: New DHCP option
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset=US-ASCII
Mime-Version: 1.0
X-Mailer: Novell GroupWise Internet Agent 5.5.2.1
Content-Transfer-Encoding: 8bit
X-MIME-Autoconverted: from quoted-printable to 8bit by ietf.org id EAA27859

I would like to know the procedure for defining new DHCP options. I've gone through the RFC 2489 & 2434 ("Procedure for Defining New DHCP Options" & "Guidelines for Writing an IANA Considerations Section in RFCs") and prepared a draft. Where should i submit this draft and in which format this draft should be submitted ??

thanx in anticipation,
Nirupma Kulshreshtha.



From dhcp-v4@bucknell.edu  Fri Nov 12 12:25:17 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA21691
	for <dhc-archive@odin.ietf.org>; Fri, 12 Nov 1999 12:25:15 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA25657;
	Fri, 12 Nov 1999 12:24:26 -0500 (EST)
Date: Fri, 12 Nov 1999 12:24:26 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942426309.17378.mwc@atlantic>
Errors-To: droms@bucknell.edu
Reply-To: Michael.Carney@East.Sun.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: New DHCP option
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

Hi Nirupma,

	Go to www.ietf.org; click on Internet-drafts. Read the "Guidelines to authors
of internet drafts" and "submitting documents to the IESG". List the draft as
being of interest to:

Dynamic Host Configuration Working Group.

Good luck,


Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v4@bucknell.edu  Fri Nov 12 18:40:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA07216
	for <dhc-archive@odin.ietf.org>; Fri, 12 Nov 1999 18:40:04 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA31042;
	Fri, 12 Nov 1999 18:39:39 -0500 (EST)
Date: Fri, 12 Nov 1999 18:39:39 -0500 (EST)
Message-Id: <078292D50C98D2118D090008C7E9C6A605142762@STAY.platinum.corp.microsoft.com>
Errors-To: droms@bucknell.edu
Reply-To: annd@Exchange.Microsoft.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Ann Demirtjis (Exchange)" <annd@Exchange.Microsoft.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: NAT/IPSEC draft
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2232.9)
Mime-Version: 1.0


can someone point me to the location of the draft describing the option for
the NAT/IPSEC interaction (the one that was reviewed in DC)? 

I searched the draft library at www.ietf.org as well as the draft list at
www.dhcp.org and cannot find it anywhere.

Thanks.

Ann


From dhcp-v4@bucknell.edu  Fri Nov 12 22:50:23 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA09392
	for <dhc-archive@odin.ietf.org>; Fri, 12 Nov 1999 22:50:23 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA29671;
	Fri, 12 Nov 1999 22:50:12 -0500 (EST)
Date: Fri, 12 Nov 1999 22:50:12 -0500 (EST)
Message-Id: <Version.32.19991112221629.00e05ce0@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: droms@bucknell.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: NAT/IPSEC draft
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Ann - I don't believe Jose and Juan published a draft prior to the deadline
for DC.  Here's their contact information:

Juan A. Garay   garay@research.bell-labs.com
Jose Brustoloni jcb@research.bell-labs.com

- Ralph








From dhcp-v4@bucknell.edu  Mon Nov 15 12:35:08 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA21874
	for <dhc-archive@odin.ietf.org>; Mon, 15 Nov 1999 12:35:07 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA20215;
	Mon, 15 Nov 1999 12:33:53 -0500 (EST)
Date: Mon, 15 Nov 1999 12:33:53 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE53@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


yes, Ted, I think we are in agreement here:

(1) any class identifier (vendor or user) SHOULD be unambiguous, concise,
preferably a text string
(2) the format of a class identifier SHOULD NOT be specified by an RFC (that
is, it is an opaque string)
(3) the processing by clients and servers of class identifiers SHOULD NOT be
specified by an RFC -- however, the "promoter" of a class identifier MUST be
able to provide clear and complete guidance to administrators who must
configure their clients and servers to perform the intended processing

The latter point is the basis for differentiation among vendors of clients
and servers.  In the simplest case, class identifier "ZBT" might select one
or more option values to offer a client.  More complex cases could be built
if clients and servers had richer vocabularies of conditional actions for
processing requests and responses, but this should be entirely out of the
scope of an RFC, whose purpose is only to specify the code, length, and
description of a class identifier.

--Barr


> > While having the server interpret the identifier and base its response
> on
> > the contents has some appeal, how do we specify a general method that
> can be
> > easily implemented in the universe of clients and servers?
> 
> Barr, I think that you have mistaken the context of the suggestion
> that I made to the fellow at Apple.   I am most definitely _not_
> saying we should write a standard that says Apple must do what I
> suggested.   I am not saying we should specify what a server should do
> if a vendor uses a format like the one I suggested.
> 
> 


From dhcp-v4@bucknell.edu  Mon Nov 15 13:25:30 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA17636
	for <dhc-archive@odin.ietf.org>; Mon, 15 Nov 1999 13:25:25 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA16346;
	Mon, 15 Nov 1999 13:24:29 -0500 (EST)
Date: Mon, 15 Nov 1999 13:24:29 -0500 (EST)
Message-Id: <199911151755.MAA04087@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> (1) any class identifier (vendor or user) SHOULD be unambiguous, concise,
> preferably a text string

Okay.

> (2) the format of a class identifier SHOULD NOT be specified by an RFC (that
> is, it is an opaque string)

"Opaque string" != "not specified by RFC", which I think is one of the
causes of confusion over this point.   I think that it might be better
to say "is unspecified" than "is an opaque string".

> (3) the processing by clients and servers of class identifiers SHOULD NOT be
> specified by an RFC -- however, the "promoter" of a class identifier MUST be
> able to provide clear and complete guidance to administrators who must
> configure their clients and servers to perform the intended processing

mmm, I like this.   :')

			       _MelloN_


From dhcp-v4@bucknell.edu  Mon Nov 15 13:27:02 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA18312
	for <dhc-archive@odin.ietf.org>; Mon, 15 Nov 1999 13:27:01 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA31453;
	Mon, 15 Nov 1999 13:25:59 -0500 (EST)
Date: Mon, 15 Nov 1999 13:25:59 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE55@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP Vendor Class question 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0

> > (2) the format of a class identifier SHOULD NOT be specified by an RFC
> (that
> > is, it is an opaque string)
> 
> "Opaque string" != "not specified by RFC", which I think is one of the
> causes of confusion over this point.   I think that it might be better
> to say "is unspecified" than "is an opaque string".
> 
OKAY -- an important semantic point....  I was using the definition of
opaque string as "the contents and format of the string are not specified,"
so your clarification is right on target!

--Barr


From dhcp-v6@bucknell.edu  Tue Nov 16 13:55:24 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA28193
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 13:55:24 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA19696;
	Tue, 16 Nov 1999 13:50:16 -0500 (EST)
Date: Tue, 16 Nov 1999 13:50:16 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942776689.16960.mwc@atlantic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

Hi Folks,

	There were a number of discussion questions we covered in
DC. I'm going to send out each question with background for context in
separate email messages. Please feel free to jump in if the topic
interests you.

Here's the first question:

1) Seeking examples of releasable resources. An IP address is one. Others?
When would they be released?

EXPLANATION: What I was seeking here was an understanding of whether
we'd ever have resources other than IP addresses that would be
releasable. If the answer is NO, then Release becomes much easier. If
the answer is YES, then we have to do more work to identify those
resources which are releasable, but are opaque to the client (and
server) implementations.

In DC, we were really reaching for examples of releasable resources over
and above IP addresses.

My opinion: SLP is better at dealing with registering services, etc. We
have always thought that SLP would do more of the heavy lifting than it
did in the DHCPv4 environment. However, SLP still isn't deployed widely.
So, SLP should be used for releasable resources over and above what is
necessary for a host to achieve the ability to talk on the global
internet. Perhaps DHCPv6 can help SLP deployment by remaining true to
minimal stack-only configuration.


Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v6@bucknell.edu  Tue Nov 16 13:56:09 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA28585
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 13:56:09 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA18273;
	Tue, 16 Nov 1999 13:53:09 -0500 (EST)
Date: Tue, 16 Nov 1999 13:53:09 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942777701.21777.mwc@atlantic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: DC meeting: Discussion question #3, Reply to RECONFIGURE? (Final)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


3) Is a reply to a RECONFIGURE always desirable?

EXPLANATION: This question was intended to seek additional information
regarding Thomas Narten's concern about the RECONFIGURE_REPLY "implosion"
when a RECONFIGURE is multicast (or the server being hammered due to
RECONFIGURE_INIT message(s) being multicast). If scenario B described in
question #2 were used, and a RECONFIGURE message was sent out to change
the DNS server, one could argue that a RECONFIGURE_REPLY isn't necessary
or even preferred.

A server could also reduce its exposure to such "implosions" by sending
groups of clients different multicast RECONFIGURE addresses, ensuring that
the possible RECONFIGURE_REPLY/REQUESTs load is managable. Text describing
the problem and recommending a solution would need to be added to the draft.


Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v6@bucknell.edu  Tue Nov 16 14:58:46 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA26167
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 14:58:45 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA00393;
	Tue, 16 Nov 1999 14:54:13 -0500 (EST)
Date: Tue, 16 Nov 1999 14:54:13 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942777317.29902.mwc@atlantic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

Some background for this question:

In response to input by the WG, RECONFIGURE has been
clarified by adding additional messages:

Server:
=======
RECONFIGURE
RECONFIGURE_INIT

Client:
RECONFIGURE_REPLY

RECONFIGURE is used to send extensions to client(s) from the server,
to which a client would reply with a RECONFIGURE_REPLY message.

RECONFIGURE_INIT is used to notify clients that they are to perform
the REQUEST/REPLY exchange with the server.


2) Stateful autoconfiguration and multiple IP addresses. When would a
client need to request additional IP addresses over and above what the
administrator has configured through site policy?

EXPLANATION: This question can be rephrased as: "Can a client request IP
addresses over and above what has been configured by the server?"
Examples brought up at DC were somewhat far-fetched. However, in DHCPv4,
multiple IP addresses have been assigned to an interface simply by
specifying a different clientID. To the DHCPv4 server, it looks like
there are two distinct clients, when in fact there is a single client
managing the IP addresses for the interface. This is used to host a
service on different IP addresses (X.X.X.X.25)...

The IPng WG is working to clarify multihoming, which seems to point to
more interest in multiple addresses per interface.

Unless I've got this totally wrong (and I'm sure I'll be corrected if I
have it wrong), the questions regarding multiple IP addresses and
releasable resources are really parts of an overriding question: "Can
a client request parts of a configuration, or does a DHCP server reserve
and return a "configuration" for a client which contains IP address(es)
as well as other parameters, and that "configuration" is treated as a
black box by the client and the server?

Scenario A: Black box.

Server is configured with groupings of parameters which include IP
address(es) through server policy mechanisms. The combination of these
items is treated as one item by DHCP.

Scenario A ramifications:

Clients are really simple, since they simply request a "configuration" for
their link-local + prefix. There is no 'C' bit required for Solicit or
Request. Release refers to the client's "configuration" as defined by
the network administrator. A client might request an additional
"configuration" for an interface by specifying a different link-local
address w/ the same prefix (a'la DHCPv4). Reconfiguration refers to the
entire "configuration", not a subset of it. Thus, the concept of a
Reconfigure which contains all of the data in a "configuration" would
never be multicast, since it refers to data for specific link local
address(es) + prefix(es). Thus, one could argue that Reconfigures
containing data would never occur, and thus Reconfigure becomes
Reconfigure_init (which instructs the client to perform a Request/Reply
exchange with the server).

Note that if any of the data contained within a "black box" is changed,
then the server would need to generate Reconfigure message(s) to cause
the clients to resynchronize with the server.

Scenario B: Individual extensions

This scenario reflects the protocol as currently defined. Clients can
request individual extensions (how? no parameter request list extension)
which could include one or more IP address extensions. Assume for this
discussion that the protocol has been changed to include the following
change which I proposed as part of "List of DHCPv6 Issues with
responses, part 2 of 2": Advertise messages include IP address
extensions with Prefix (S bit set) specifications advertising which
prefixes are available and have IP addresses for assignment. The client
forms a Request which requests one or more IP addresses (which contain
Prefix specifications from the Advertisements) of the scope necessary.
Also included by the server are extensions configured (site policy)
which are informational in nature. Those extensions which identify
releasable resources (by a bit in the extension) are so marked. A
client can release all releasable resources if it develops amnesia by
generating Solicit/Requests with the C bit set. A client can release
certain releasable resources by including their extension(s) in
a Release. If a client wants to release all resources, it may generate a
Release with no extensions. A server can Reconfigure a client by generating a
RECONFIGURE message which will contain any combination of releasable or
informational extensions, to which a client would respond with a
RECONFIGURE_REPLY. For example, if a site changed the IP address of the
DNS server, the DHCP server could multicast out a RECONFIGURE with just
the extension for the DNS server with the new value to the clients.
Releasable resources would need to be Reconfigured by unicast
RECONFIGUREs or by Multicast RECONFIGURE_INITs, which would cause the
clients to perform the Request/Reply transaction with the server.

Scenario B ramifications:

Deciding whether or not to generate RECONFIGURE* messages in this
scenario is more difficult than Scenario A.

DISCUSSION:

The question to the WG is whether the additional complication of
the protocol described in Scenario B is worth the additional features
over and above what is possible with Scenario A.

Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v6@bucknell.edu  Tue Nov 16 14:58:47 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA26168
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 14:58:45 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA06580;
	Tue, 16 Nov 1999 14:52:47 -0500 (EST)
Date: Tue, 16 Nov 1999 14:52:47 -0500 (EST)
Message-Id: <199911161854.NAA0000016552@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: DHCPv6 Input from Francis Dupont
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Forwarded with permission from Francis...
Can we discuss this please...

thanks
the authors

------- Forwarded Message

Return-Path: Francis.Dupont@inria.fr
Received: from quarry.zk3.dec.com by iota.zk3.dec.com (8.7.6/UNX 1.7/1.1.20.3/24Apr98-0811AM)
	id KAA0000032657; Mon, 15 Nov 1999 10:39:51 -0500 (EST)
Received: from mail2.digital.com by quarry.zk3.dec.com (8.8.8/1.1.8.2/16Jan95-0946AM)
	id KAA0000001589; Mon, 15 Nov 1999 10:39:49 -0500 (EST)
Received: from concorde.inria.fr (concorde.inria.fr [192.93.2.39])
	by mail2.digital.com (8.9.2/8.9.2/WV2.0g) with ESMTP id HAA23424
	for <bound@zk3.dec.com>; Mon, 15 Nov 1999 07:39:16 -0800 (PST)
Received: from givry.inria.fr (givry.inria.fr [193.51.193.144])
	by concorde.inria.fr (8.8.7/8.8.7) with ESMTP id QAA29980;
	Mon, 15 Nov 1999 16:27:47 +0100 (MET)
Received: from givry.inria.fr (givry.inria.fr [193.51.193.144]) by givry.inria.fr (8.7.6/8.7.3) with ESMTP id QAA07608; Mon, 15 Nov 1999 16:27:46 +0100 (MET)
Message-Id: <199911151527.QAA07608@givry.inria.fr>
From: Francis Dupont <Francis.Dupont@inria.fr>
To: Jim Bound <bound@zk3.dec.com>, Charles Perkins <charliep@iprg.nokia.com>,
        Mike Carney <Michael.carney@east.sun.com>
Cc: Yanick Pouffary <pouffary@taec.enet.dec.com>
Subject: DHCPv6 flawns with fixes
Date: Mon, 15 Nov 1999 16:27:28 +0100
Sender: Francis.Dupont@inria.fr

I've read again technical reports from my students, current I-Ds
(dhcpv6-14 and v6exts-11) and ND/autoconf RFCs.
There are still minor problems in I-Ds, perhaps some of them are already
fixed, and a major problem with multiple prefixes (I believe I've found
a nice way to fix it).

First some obvious minor bugs in dhcpv6-14:
 - minimal MTU for IPv6 is 1280, not 1500 (annexe B, pages 35-36)
   (the discussion is still valid, only the value is wrong)
 - transaction ID threshold is 1024, not 1000, and clients use the >= 1024,
   not the <= 1000 (2.2 page 4, 4.3 page 12, 5.7 page 21)
   (fix 2.2)

Lacks of documentation:
 - dhcpv6-14 should explain which kinds of extensions can be in advertise
   messages, I believe only security extensions are useful there
 - DHCPv6 is link-oriented, ie. a multi-homed client should deal with
   autoconfiguration independently on each link
 - I'd like to get more text about renumbering and DHCPv6 servers,
   for instance I believe DHCPv6 servers should be targets of the router
   renumbering protocol/process (verify with Matt if this can be a problem,
   I believe this should work fine). Another interaction is for DHCPv6
   clients which are routers too.
 - v6exts-11 should mention address extension as the only defined releasable
   resource (then you can defer releable resource issue solution :-)

A potential problem in dhcpv6-14:
 - IPv6 address fields should be aligned on 64 bit boundaries in headers
 - the use of prefix-size is still not clear for me
 - I prefer to have a 32 bit interface index (relative to the agent)
   in place of the agent address + prefix-size. I am not convinced
   address + prefix-size is enough for all topologies, an interface index
   is clearer and easier to use (as a scope ID). This solves the alignement
   problem to if link-local addresses are replaced by interface index +
   link-local address.

Multiple prefix problem:
 - summary: my idea is to enforce one-for-one extension in request/reply
   and to use reconfigure in order to add extra prefixes.
 - the real source of the problem is in RFC 2462 5.5.3 "Router
   Advertisement Processing" page 17 where you can find after two paragraphs
   about M (managed) and O (other) bits this statement:
   "For each Prefix-Information option in the Router Advertisement: a-e)"
   obviously the text after this statement and the 5.6 "Configuration
   Consistency" section next page (with "hosts accept the union of all
   information ..." are misleading in the stateful autoconf context.
 - if the stateful autoconf *only* is used for address management then
   the good way to change a prefix on clients (for new prefixes, lifetime
   changes, ...) is to use the reconfiguration facility.

Proposed stateful address management:
 - at boot time a stateful client will request an address to its server.
 - a server should reserved an address per prefix to a new client,
   sends the first address in the reply and sends reconfigure messages
   for extra addresses. At the end of this phased an address per prefix
   is allocated/configurated.
 - on a prefix change, a server sends reconfigure messages to its clients
   (it may use multicast if available) in order to reflect the change.
 - a client can look at prefix-infos in router advertisements but should
   not use them directly if a stateful server is available. For instance
   if a router advertisement shows a change for a prefix, a stateful
   client MUST NOT apply the change but MAY send a request to its server
   in order to validate the change after a random delay (in order to avoid
   many requests at the same time and to give the time for a reconfigure).

Proposed new mode of autoconfiguration:
 - today we have stateless (use router advertisements) and stateful
   (use DHCPv6). If we make stateful stronger then we have to deal with
   the case where stateful is selected but no stateful service is available.
 - in this case (named ?) stateless rules will apply and from time to
   time an attempt should be done in order to find a stateful server.
 - if one is found the standard revalidation process should be done
   (same case than the current server is lost) and the client should
   enter in the "strong" stateful mode.
 - the same process should apply when the managed flag changes from
   false to true the first time.
 - another view of this can be:
   * strong stateful is "stateful enabled, stateless disabled"
   * weak stateful is "stateful and stateless enabled", ie.
     when the stateful mode is enabled and a stateful service is available
     then the stateless mode is disabled.
   this can make the wording of RFC 2462 update easier, ie. the only thing
   needed is an addition of "when the stateless mode is available" at
   the middle of 5.5.3 section (For each Prefix-Information option...).
   The conflicting 5.6 will need near no clarification too!

Regards

Francis.Dupont@inria.fr

PS: I can send Jean-Jacques'report (in French) to Yannick if you'd like,
his major comment is the current ID should enforce the "no user config
on client" principle, ie. the state should be on servers.

------- End of Forwarded Message



From dhcp-v6@bucknell.edu  Tue Nov 16 18:10:11 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA28095
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 18:10:10 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA31348;
	Tue, 16 Nov 1999 18:05:44 -0500 (EST)
Date: Tue, 16 Nov 1999 18:05:44 -0500 (EST)
Message-Id: <199911162211.RAA13768@rotala.raleigh.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM> writes:

> 1) Seeking examples of releasable resources. An IP address is one. Others?
> When would they be released?

> EXPLANATION: What I was seeking here was an understanding of whether
> we'd ever have resources other than IP addresses that would be
> releasable. If the answer is NO, then Release becomes much easier. If
> the answer is YES, then we have to do more work to identify those
> resources which are releasable, but are opaque to the client (and
> server) implementations.

Right. There are semantics associated with the current definition of
Release that make it more complex for releasable resources. For
example, a client is only allowed to release releasable resources via
the same server that originally allocated them. That means a client
has to keep track of that and can't just release stuff if it switches
DHCP server. It also means that (say) if it gets a Reconfigure from
another server, it can't release that resource to it. I don't fully
understand why things are defined this way, but they are.

> My opinion: SLP is better at dealing with registering services, etc. We
> have always thought that SLP would do more of the heavy lifting than it
> did in the DHCPv4 environment. However, SLP still isn't deployed widely.
> So, SLP should be used for releasable resources over and above what is
> necessary for a host to achieve the ability to talk on the global
> internet. Perhaps DHCPv6 can help SLP deployment by remaining true to
> minimal stack-only configuration.

I'm not sure I follow the comparison with SLP. SLP is for locating
servers and services. How you actually access a service is a secondary
issue and not SLP specific. Releasable resources are a specific type
of resource that the current DHCP spec allows a server to hand out to
clients. I.e., it is a way of allowing DHCP servers to assign specific
resources to a client.

My basic question on releasable resources has to do with cost
(complexity) vs. benefits (what will they really be used for).

Thomas


From dhcp-v6@bucknell.edu  Tue Nov 16 18:10:18 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA28156
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 18:10:18 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA08000;
	Tue, 16 Nov 1999 18:07:44 -0500 (EST)
Date: Tue, 16 Nov 1999 18:07:44 -0500 (EST)
Message-Id: <199911162221.RAA13784@rotala.raleigh.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #3, Reply to RECONFIGURE? (Final) 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

> 3) Is a reply to a RECONFIGURE always desirable?

> EXPLANATION: This question was intended to seek additional information
> regarding Thomas Narten's concern about the RECONFIGURE_REPLY "implosion"
> when a RECONFIGURE is multicast (or the server being hammered due to
> RECONFIGURE_INIT message(s) being multicast). If scenario B described in
> question #2 were used, and a RECONFIGURE message was sent out to change
> the DNS server, one could argue that a RECONFIGURE_REPLY isn't necessary
> or even preferred.

> A server could also reduce its exposure to such "implosions" by sending
> groups of clients different multicast RECONFIGURE addresses, ensuring that
> the possible RECONFIGURE_REPLY/REQUESTs load is managable. Text describing
> the problem and recommending a solution would need to be added to the draft.

I've been concerned for some time with the basic idea that the server
multicasts (widely) to a set of clients, and the clients are required
to ack. This results in ack implosion and there is no easy way around
the problem. I.e., this is the reliable multicast problem, which we
don't have good general solutions for.

I also think the same effect can be achieved by changing the semantics
of RECONFIGURE. That is, instead of having the RECONFIGURE mean "send
me an ACK", have it instead tell the client "the server has new info,
use your normal DHCP protocol exchanges to get that info
reliably". This message can be multicast a number of times, with the
client ignoring subsequent retransmissions. That way the server can
retransmit multicast RECONFIGUREs to ensure that everyone gets a copy,
and it can also check for subseqent client REQUEST messages if it
really wants to know who has received its RECONFIGURE. I.e., the
subsequent REQUESTs are the ACK the server is looking for.

Thomas


From dhcp-v6@bucknell.edu  Tue Nov 16 20:06:27 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA09192
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 20:06:26 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA28216;
	Tue, 16 Nov 1999 20:04:13 -0500 (EST)
Date: Tue, 16 Nov 1999 20:04:13 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942795473.30019.mwc@atlantic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

> Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM> writes:
> 
> > 1) Seeking examples of releasable resources. An IP address is one. Others?
> > When would they be released?
> 
> > EXPLANATION: What I was seeking here was an understanding of whether
> > we'd ever have resources other than IP addresses that would be
> > releasable. If the answer is NO, then Release becomes much easier. If
> > the answer is YES, then we have to do more work to identify those
> > resources which are releasable, but are opaque to the client (and
> > server) implementations.
> 
> Right. There are semantics associated with the current definition of
> Release that make it more complex for releasable resources. For
> example, a client is only allowed to release releasable resources via
> the same server that originally allocated them. That means a client
> has to keep track of that and can't just release stuff if it switches
> DHCP server. It also means that (say) if it gets a Reconfigure from
> another server, it can't release that resource to it. I don't fully
> understand why things are defined this way, but they are.

Any releaseable resource would need to have some lifetime or lease associated
with it in order to automatically reclaim it. That lifetime (and the fact that
a particular extension carried releaseable data) would need to be communicated 
to the client in a non-extension-specific manner (So that clients would not
have to change if new extensions are added which contain releaseable data).
Folks at Minneapolis requested such a mechanism. If we decide that supporting
releaseable/leasable resources is important, we'd need to add supporting text.

But before we drive down into solving the semantics of releaseable resources,
we still have to ask whether the complexity of managing releaseable/leased
resources is worth it given we don't have examples of such resources (other
than IP addresses).

> 
> > My opinion: SLP is better at dealing with registering services, etc. We
> > have always thought that SLP would do more of the heavy lifting than it
> > did in the DHCPv4 environment. However, SLP still isn't deployed widely.
> > So, SLP should be used for releasable resources over and above what is
> > necessary for a host to achieve the ability to talk on the global
> > internet. Perhaps DHCPv6 can help SLP deployment by remaining true to
> > minimal stack-only configuration.
> 
> I'm not sure I follow the comparison with SLP. SLP is for locating
> servers and services. How you actually access a service is a secondary
> issue and not SLP specific. Releasable resources are a specific type
> of resource that the current DHCP spec allows a server to hand out to
> clients. I.e., it is a way of allowing DHCP servers to assign specific
> resources to a client.

There was no comparison intended between DHCPv6 and SLP. If we were to decide
that releaseable/leaseable resources other than IP addresses weren't something
we'd deal with in DHCPv6, then we'd need some other method of managing them,
and SLP seems to fit.

> 
> My basic question on releasable resources has to do with cost
> (complexity) vs. benefits (what will they really be used for).

And in order to answer this question we need to determine whether there
are examples of resources we think DHCPv6 should manage which are
releaseable/leaseable.

> 
> Thomas



Mike Carney
SNT Internet Engineering
Sun Microsystems, Inc.
Mailstop UBUR02-212
One Network Drive
Burlington, MA 01803




From dhcp-v6@bucknell.edu  Tue Nov 16 22:11:03 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA02734
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 22:11:02 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA22932;
	Tue, 16 Nov 1999 22:07:47 -0500 (EST)
Date: Tue, 16 Nov 1999 22:07:47 -0500 (EST)
Message-Id: <199911170237.VAA05260@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: dhcp-v6@bucknell.edu (by way of Ralph Droms <dhcp-v6@bucknell.edu>)
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


message sent to dhcp-v6@bucknell.edu by CARL.SMITH@ENG.SUN.COM follows.
-----------------------------------------------------------------------
> But before we drive down into solving the semantics of releaseable
resources,
> we still have to ask whether the complexity of managing releaseable/leased
> resources is worth it given we don't have examples of such resources (other
> than IP addresses).

	No, I believe we have to know what semantics need to be supported
in order to properly evaluate how complex the management solution needs to
be.  Of course the lack of examples of releasable resources makes it even
more difficult for us to try to infer appropriate semantics.

			Carl




From dhcp-v6@bucknell.edu  Tue Nov 16 22:47:09 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA20256
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 22:47:09 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA19510;
	Tue, 16 Nov 1999 22:44:51 -0500 (EST)
Date: Tue, 16 Nov 1999 22:44:51 -0500 (EST)
Message-Id: <199911170310.WAA0000018292@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #3, Reply to RECONFIGURE? (Final) 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>I also think the same effect can be achieved by changing the semantics
>of RECONFIGURE. That is, instead of having the RECONFIGURE mean "send
>me an ACK", have it instead tell the client "the server has new info,
>use your normal DHCP protocol exchanges to get that info
>reliably". This message can be multicast a number of times, with the
>client ignoring subsequent retransmissions. That way the server can
>retransmit multicast RECONFIGUREs to ensure that everyone gets a copy,
>and it can also check for subseqent client REQUEST messages if it
>really wants to know who has received its RECONFIGURE. I.e., the
>subsequent REQUESTs are the ACK the server is looking for.

I view it as ACK explosion (client exploding acks towards the servers)
but no big deal. 

A compromise might be to attach a suffix to this reconfigure type like
RCONFIGURE-INIT and have it mean as you say do a request via dhcp.
Then down the road if we figure out scalable multicast we add a new
message type called RECONFIGURE-INIT-W/ACK or some such name and new msg
type.

If we did this then would you and others be OK with how we would do the
hack now to prevent clients from responding to multiple multicasts of
this nature? 

/jim



From dhcp-v6@bucknell.edu  Tue Nov 16 22:48:40 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA20967
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 22:48:39 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA21194;
	Tue, 16 Nov 1999 22:46:06 -0500 (EST)
Date: Tue, 16 Nov 1999 22:46:06 -0500 (EST)
Message-Id: <199911170332.WAA0000022374@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>Right. There are semantics associated with the current definition of
>Release that make it more complex for releasable resources. For
>example, a client is only allowed to release releasable resources via
>the same server that originally allocated them. That means a client
>has to keep track of that and can't just release stuff if it switches
>DHCP server. It also means that (say) if it gets a Reconfigure from
>another server, it can't release that resource to it. I don't fully
>understand why things are defined this way, but they are.

I tend to think the only real releasable resources are the Addresses.
I am beginning to think that a releaseable resource is one which the
server can use elsewhere?  The only ones I can come up with are
addresses?  

But I want to make sure we get to all the issues in this discussion too.

When you say semantics can you be specific?  Is it the resource-server
associations statements?  The "D" bit?  etc...  Just so we don't miss
anything in the syntax either that may imply semantics?

Or is it just a matter of saying that clients only release addresses?
And its that simple?

As far as the multiple server issue that I think is a separate issue?

To release an address for a client (assuming for this discusssion thats
all we are talking about) to a server, when the server gets the release
it will have to do a search on that client and locate the entry for that
client on some non-volatile storage medium I will assume (sure it could
be a cache in a good fail-over server) then do the delete and reuse or
wait for XX time before reusing.

Today we have no mechanism for servers to know about each others
databases and entries.  So we assume that servers can't do this and say
a client can keep the address of a server who gave them their address,
because we know new servers can't "today" know about each others
databases in a "standard" way.  

But..tomorrow it may be there is a standard way for servers to know
about each other and clients can clearly just send the release to any
server if that exists in Theory....if the standard exists.  

Here is a suggestion for discussion:

1.  Lets make the default that the client has no knowledge of servers
    and can release addresses to the next server it hears (e.g. 
    reconfigure, reply, advertise [from reboot]).

2.  But put a MAY in the protocol that permits clients to keep
    the address of servers to later release addresses from where
    it got them.

#2 will permit mobile nodes to release adddresses that roam around
before we get a server-to-server protocol deployed in the market.

thanks
/jim



From dhcp-v6@bucknell.edu  Tue Nov 16 23:30:08 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA07983
	for <dhc-archive@odin.ietf.org>; Tue, 16 Nov 1999 23:30:08 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA06827;
	Tue, 16 Nov 1999 23:26:32 -0500 (EST)
Date: Tue, 16 Nov 1999 23:26:32 -0500 (EST)
Message-Id: <Version.32.19991116224027.00e02b90@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 10:11 PM 11/16/99 -0500, dhcp-v6@bucknell.edu wrote:
>
>message sent to dhcp-v6@bucknell.edu by CARL.SMITH@ENG.SUN.COM follows.
>-----------------------------------------------------------------------
>> But before we drive down into solving the semantics of releaseable
>resources,
>> we still have to ask whether the complexity of managing releaseable/leased
>> resources is worth it given we don't have examples of such resources (other
>> than IP addresses).
>
>	No, I believe we have to know what semantics need to be supported
>in order to properly evaluate how complex the management solution needs to
>be.  Of course the lack of examples of releasable resources makes it even
>more difficult for us to try to infer appropriate semantics.

I might ask a slightly different series of questions:

1. Should we take on the exercise of defining the semantics of releasable
resources, given that we don't have examples of such resources?

2. If the answer to (1) is yes, then - after defining the semantics of
releasable resources - is there reason to develop a mechanism for managing
releasable resources DHCPv6?

3. If the answer to (2) is yes, then is the cost of defining and
implementing a mechanism for managing releasable resources (and that cost
may be 0) worth the benefit of being able to manage those resources through
DHCPv6?

- Ralph



From dhcp-v6@bucknell.edu  Wed Nov 17 06:27:00 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id GAA02450
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 06:27:00 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id GAA31755;
	Wed, 17 Nov 1999 06:24:44 -0500 (EST)
Date: Wed, 17 Nov 1999 06:24:44 -0500 (EST)
Message-Id: <Version.32.19991116224950.00e044e0@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

So, let me answer (1) with "yes", and take a stab at defining "releasable
resources".

1. Should we take on the exercise of defining the semantics of releasable
resources, given that we don't have examples of such resources?

Here's a working definition for a "releasble resource": a resource that is
uniquely assigned to a client and that can be reassigned to a different
client at some point in the future.

Discussion: If I remember correctly, the idea of "releasable resources"
first surfaced in the Minneapolis WG meeting.  We generalized the idea from
IP addresses, in a discussion about where a client should send a RELEASE
message.  I don't remember that we had any specific examples of releasable
resources othee than IP addresses.  Our experience with DHCPv4 is that IPv4
addresses are "releasable resources" while everything else managed through
DHCPv4 is a "configuration parameter".

Based on the working definition, here are some preliminary observations
about managing releasable resources:

Servers will:
  - Implement assignment policy determined by DHCP admin
  - Assign resources as part of the configuration requested by clients
  - Track assignment of resources to clients
  - Reassume responsibility for resources released by clients
  - Recover resources (presumably through leasing mechanism)

Clients will:
  - Request configuration from servers
  - Keep track of resources and leases on those resources
  - Release resources




From dhcp-v6@bucknell.edu  Wed Nov 17 08:46:18 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id IAA05395
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 08:46:17 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA15384;
	Wed, 17 Nov 1999 08:42:00 -0500 (EST)
Date: Wed, 17 Nov 1999 08:42:00 -0500 (EST)
Message-Id: <199911171305.IAA31922@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: dhcp-v6@bucknell.edu (by way of Ralph Droms <dhcp-v6@bucknell.edu>)
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


message sent to dhcp-v6@bucknell.edu by NARTEN@HYGRO.ADSL.DUKE.EDU follows.
---------------------------------------------------------------------------
Jim Bound <bound@zk3.dec.com> writes:


> >Right. There are semantics associated with the current definition of
> >Release that make it more complex for releasable resources. For
> >example, a client is only allowed to release releasable resources via
> >the same server that originally allocated them. That means a client
> >has to keep track of that and can't just release stuff if it switches
> >DHCP server. It also means that (say) if it gets a Reconfigure from
> >another server, it can't release that resource to it. I don't fully
> >understand why things are defined this way, but they are.

> I tend to think the only real releasable resources are the Addresses.
> I am beginning to think that a releaseable resource is one which the
> server can use elsewhere?  The only ones I can come up with are
> addresses?  

> But I want to make sure we get to all the issues in this discussion too.

> When you say semantics can you be specific?  Is it the resource-server
> associations statements?  The "D" bit?  etc...  Just so we don't miss
> anything in the syntax either that may imply semantics?

The current draft has pretty specific text about how a client releases
resources:

>    a DHCP Request message with the `C' bit set and no extensions.  If a
>    client wishes to retain some of its network configuration parameters,
>    but determines that others are no longer needed, it SHOULD enable
>    the server to release allocated resources which are no longer in
>    use by sending a DHCP Release message to the server, and including
>    extensions to identify the unneeded items.  The client consults its
>    list of resource-server associations in order to determine which
>    server should receive the desired Release message.  The Release
>    MUST be transmitted to the server that provided the configuration
>    parameters.

Note that I'm assuming that the only resources a client would release
are "releasable resources", i.e., addresses that have explicit
leases. There is no need to release other types of parameters. Right?

So, I guess my previous mail wasn't quite clear. Requiring that
clients rememember which server they got a resource from and then
specifying that it can only release resources by contacting the same
server again (in the MUST sense) seems a bit restrictive. If we
imagine a server-server protocol becoming available, being able to
release a resource to any of the available servers seems desirable. It
might still make sense (as an optimization) to have clients first
attempt to release a resource via the same server it got it from, but
if that fails, wouldn't it make sense to allow it to try releasing
resources via another server? Or more generally, perhaps it makes
sense to say a client always prefers a particular server, the one it
most recently got parameters/resources from. That way, a new server
could (say) send a reconfigure that causes the client to switch to the

new server and then use that new server for subsequent communication,
regardless of where it got resources from previously.

> Or is it just a matter of saying that clients only release addresses?
> And its that simple?

I think this is in essence the question about whether we should
generalize addresses into releasable resources. Certainly, we'll want
to be able to release addresses. 

> As far as the multiple server issue that I think is a separate issue?

> To release an address for a client (assuming for this discusssion thats
> all we are talking about) to a server, when the server gets the release
> it will have to do a search on that client and locate the entry for that
> client on some non-volatile storage medium I will assume (sure it could
> be a cache in a good fail-over server) then do the delete and reuse or
> wait for XX time before reusing.

Agreed.

> Today we have no mechanism for servers to know about each others
> databases and entries.  So we assume that servers can't do this and say
> a client can keep the address of a server who gave them their address,
> because we know new servers can't "today" know about each others
> databases in a "standard" way.  

> But..tomorrow it may be there is a standard way for servers to know
> about each other and clients can clearly just send the release to any
> server if that exists in Theory....if the standard exists.  

> Here is a suggestion for discussion:

> 1.  Lets make the default that the client has no knowledge of servers
>     and can release addresses to the next server it hears (e.g. 
>     reconfigure, reply, advertise [from reboot]).

> 2.  But put a MAY in the protocol that permits clients to keep
>     the address of servers to later release addresses from where
>     it got them.

This seems reasonable to me. I might even go a bit further and say
that a client SHOULD (?) keep track of one server, the one it most
recently got resources/parameters from and use that one first when
releasing resources. If that fails, fall back to finding and trying a
different server. I think this retains the spirit of what is on the
current draft but also provides more flexibility in the event that a
server-server protocol comes along.

> #2 will permit mobile nodes to release adddresses that roam around
> before we get a server-to-server protocol deployed in the market.

Thomas



From dhcp-v6@bucknell.edu  Wed Nov 17 09:52:00 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA07261
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 09:52:00 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA24711;
	Wed, 17 Nov 1999 09:47:38 -0500 (EST)
Date: Wed, 17 Nov 1999 09:47:38 -0500 (EST)
Message-Id: <199911171351.IAA17247@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: dhcp-v6@bucknell.edu (by way of Ralph Droms <dhcp-v6@bucknell.edu>)
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #3, Reply to RECONFIGURE?
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


message sent to dhcp-v6@bucknell.edu by NARTEN@HYGRO.ADSL.DUKE.EDU follows.
---------------------------------------------------------------------------
Jim Bound <bound@zk3.dec.com> writes:

> >I also think the same effect can be achieved by changing the semantics
> >of RECONFIGURE. That is, instead of having the RECONFIGURE mean "send
> >me an ACK", have it instead tell the client "the server has new info,
> >use your normal DHCP protocol exchanges to get that info
> >reliably". This message can be multicast a number of times, with the
> >client ignoring subsequent retransmissions. That way the server can
> >retransmit multicast RECONFIGUREs to ensure that everyone gets a copy,
> >and it can also check for subseqent client REQUEST messages if it
> >really wants to know who has received its RECONFIGURE. I.e., the
> >subsequent REQUESTs are the ACK the server is looking for.

> I view it as ACK explosion (client exploding acks towards the servers)
> but no big deal.

The common term that describes this situation (single multicast packet
resulting in a flood of return ACKs) is ack implosion, even though
explosion might seem like a more appropriate description.

But I do think that this is a big deal. If you have ten nodes, no big
deal. But if you have 100? or 1000? Such a large number of ACKs sent
at the same time will almost certainly result in congestion and lost
packets. Even worse, if the request packet is retransmitted, the exact
same thing will likely happen again (i.e., slow servers lose), with it
fairly likely that the *exact* same packets that were lost the first
time being lost again. Hence, retrasmitting request messages is much
less effective than in the unicast case.

> A compromise might be to attach a suffix to this reconfigure type like
> RCONFIGURE-INIT and have it mean as you say do a request via dhcp.
> Then down the road if we figure out scalable multicast we add a new
> message type called RECONFIGURE-INIT-W/ACK or some such name and new msg
> type.

> If we did this then would you and others be OK with how we would do the
> hack now to prevent clients from responding to multiple multicasts of
> this nature?

I think so. One other detail I didn't mention before. It may also make
sense to have the reconfigure message include a delay time value in
it. Clients don't respond immediately to the Reconfigure, but delay a
random amount of time between 0 and the delay time value in the
received message. That way a server can arrange that clients contact
it over a period of (say) ten minutes rather than all at once (which
could result in similar problems as ACK implosion).

Thomas



From dhcp-v6@bucknell.edu  Wed Nov 17 12:41:02 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA27582
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 12:40:58 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA12836;
	Wed, 17 Nov 1999 12:36:38 -0500 (EST)
Date: Wed, 17 Nov 1999 12:36:38 -0500 (EST)
Message-Id: <199911171624.LAA15723@rotala.raleigh.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #3, Reply to RECONFIGURE? (Final) 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM> writes:

> 3) Is a reply to a RECONFIGURE always desirable?

Addressing just this specific question, I have to wonder whether such
a Reconfigure would be useful in practice. If I were a site
administrator, the idea that a change in the configuration would not
be reliably communicated to all clients is troubling. If you have new
information that clients should use immediately and needs to be sent
out ASAP, presumably it is important for nodes to get that information
and in particular stop using the older information.

One possible argument for having a non-reliable notification is that
maybe the information isn't really that important, and its OK for
clients that miss the Reconfigure to get it the next time they contact
the DHCP server for other reasons (e.g., to extend an address
lifetime). But if that is the case, does the information need to be
pushed to the clients in the first place?

Thomas


From dhcp-v6@bucknell.edu  Wed Nov 17 12:42:16 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA28161
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 12:42:14 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA19739;
	Wed, 17 Nov 1999 12:38:53 -0500 (EST)
Date: Wed, 17 Nov 1999 12:38:53 -0500 (EST)
Message-Id: <199911171644.LAA15760@rotala.raleigh.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Mike Carney - SNT Internet Engineering <Michael.Carney@East.Sun.COM> writes:

> Some background for this question:

> In response to input by the WG, RECONFIGURE has been
> clarified by adding additional messages:

> Server:
> =======
> RECONFIGURE
> RECONFIGURE_INIT

> Client:
> RECONFIGURE_REPLY

> RECONFIGURE is used to send extensions to client(s) from the server,
> to which a client would reply with a RECONFIGURE_REPLY message.

> RECONFIGURE_INIT is used to notify clients that they are to perform
> the REQUEST/REPLY exchange with the server.


> 2) Stateful autoconfiguration and multiple IP addresses. When would a
> client need to request additional IP addresses over and above what the
> administrator has configured through site policy?

> EXPLANATION: This question can be rephrased as: "Can a client request IP
> addresses over and above what has been configured by the server?"
> Examples brought up at DC were somewhat far-fetched. However, in DHCPv4,
> multiple IP addresses have been assigned to an interface simply by
> specifying a different clientID. To the DHCPv4 server, it looks like
> there are two distinct clients, when in fact there is a single client
> managing the IP addresses for the interface. This is used to host a
> service on different IP addresses (X.X.X.X.25)...

This approach would certainly also be available for DHCPv6. I.e., use
different link-local addresses when additional addresses are
needed. This has the advantage that it doesn't require the basic
DHCPv6 protocol to deal with the multiple sets of addresses problem,
yet doesn't rule out doing this sometime later if applications find it
useful.

> The IPng WG is working to clarify multihoming, which seems to point to
> more interest in multiple addresses per interface.

The ongoing multihoming work should have no impact on DHCP that I can
see at present. The existing v6 model is that a multihomed site has
multiple prefixes. This is supported by the existing stateless
addrconf (i.e., no protocol changes are needed) and is also supported
by the "basic" DHCP model (called scenario A below) where the server
tells a client all of its prefixes.

I don't believe the ongoing multihoming work has any implications for
DHCP at present.

> Unless I've got this totally wrong (and I'm sure I'll be corrected if I
> have it wrong), the questions regarding multiple IP addresses and
> releasable resources are really parts of an overriding question: "Can
> a client request parts of a configuration, or does a DHCP server reserve
> and return a "configuration" for a client which contains IP address(es)
> as well as other parameters, and that "configuration" is treated as a
> black box by the client and the server?

I'm not sure the above captures the question completely. Regarding
whether a client requests part of a configuration: if a client is
allowed to ask for additional resources, that in effect means that the
configuration is dynamic and changes depending to what the client has
asked for. I.e., each time the client asks for more resources, the
configuration assigned to the client has changed.  Thus, another way
to characterize the difference is that in one model the server has a
configuration for the client (addresses and parameters) that is in
effect static (e.g., doesn't change unless the set of prefixes in the
site changes). In the other model the configuration is dynamic, with
the exact configuration changing over time depending on what the
client has requested.

> Scenario A: Black box.

> Server is configured with groupings of parameters which include IP
> address(es) through server policy mechanisms. The combination of these
> items is treated as one item by DHCP.

> Scenario A ramifications:

> Clients are really simple, since they simply request a "configuration" for
> their link-local + prefix. There is no 'C' bit required for Solicit or
> Request. Release refers to the client's "configuration" as defined by
> the network administrator. A client might request an additional
> "configuration" for an interface by specifying a different link-local
> address w/ the same prefix (a'la DHCPv4). Reconfiguration refers to the
> entire "configuration", not a subset of it. Thus, the concept of a
> Reconfigure which contains all of the data in a "configuration" would
> never be multicast, since it refers to data for specific link local
> address(es) + prefix(es). Thus, one could argue that Reconfigures
> containing data would never occur, and thus Reconfigure becomes
> Reconfigure_init (which instructs the client to perform a Request/Reply
> exchange with the server).

> Note that if any of the data contained within a "black box" is changed,
> then the server would need to generate Reconfigure message(s) to cause
> the clients to resynchronize with the server.

Another simplification of this model is that the binding cache (which
deals with retransmitted client requests) is no longer needed, as all
requests are idempotent. In model B (below) the server (and client in
the case of reconfigures) needs to distinguish a first request from
subsequent retransmissions and respond appropriately.

Thomas


From dhcp-v6@bucknell.edu  Wed Nov 17 14:16:24 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA12802
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 14:16:23 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA04340;
	Wed, 17 Nov 1999 14:11:37 -0500 (EST)
Date: Wed, 17 Nov 1999 14:11:37 -0500 (EST)
Message-Id: <3832F6DE.42C7A62F@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Ralph Droms wrote:

> Based on the working definition, here are some preliminary observations
> about managing releasable resources:
> 
> Servers will:
>   - Implement assignment policy determined by DHCP admin
>   - Assign resources as part of the configuration requested by clients
>   - Track assignment of resources to clients
>   - Reassume responsibility for resources released by clients
>   - Recover resources (presumably through leasing mechanism)
> 
> Clients will:
>   - Request configuration from servers
>   - Keep track of resources and leases on those resources
>   - Release resources

I think this is very well stated, and matches my understanding
about what is needed.

It seems to me that one of the sticking points in this discussion
is some difficulty in imagining other releasable resources besides
IP addresses.  I'd like to address that issue in two ways:

1. I can imagine a _lot_ of releasable resources in the future, but
   it would be a mistake to try to specify them before we can have
   DHCPv6 as a Proposed Standard.  After that time, new extensions
   to specify new releasable resources will come up for discussion
   in a natural way, as long as we have a natural mechanism specified
   for handling them.  Some possible examples of releasable resources:
   = disk space
   = network memory
   = I/O devices
   = Licensed software
   = time-metered equipment
   = ports
   = protocol translators

2. However, in a way, imagining cool resources as in (1) even misses
   the point.  I think the point is that DHCP is a way to configure
   computers, and it seems clear to me that computers (especially
   computers on networks) can be considered to be composed of allocated
   resources structured according to parametric values.  DHCPv6 makes
   a nice way to collect such parameters and resources, after some
   years of work.  I vote for keeping it that way.

There is another point regarding the distinction between DHCP and SLP.
SLP is good for applications that need access to services that can be
described by attribute/value pairs.  DHCP is good for configuration
needs that do not need such precise, attribute-driven specification.
The dividing line is perhaps fuzzy between these two areas of need,
but it is not invisible.

>From discussion by Thomas Narten and Jim Bound on this subject:

> > == Jim Bound

>   == Thomas Narten

> > Today we have no mechanism for servers to know about each others
> > databases and entries.  So we assume that servers can't do this and say
> > a client can keep the address of a server who gave them their address,
> > because we know new servers can't "today" know about each others
> > databases in a "standard" way.
> 
> > But..tomorrow it may be there is a standard way for servers to know
> > about each other and clients can clearly just send the release to any
> > server if that exists in Theory....if the standard exists.
> 
> > Here is a suggestion for discussion:
> 
> > 1.  Lets make the default that the client has no knowledge of servers
> >     and can release addresses to the next server it hears (e.g.
> >     reconfigure, reply, advertise [from reboot]).

I think it would be better for clients, by default, to release resources
to the server that allocated them.  We've already got this mechanism
specified in the draft now.

> > 2.  But put a MAY in the protocol that permits clients to keep
> >     the address of servers to later release addresses from where
> >     it got them.
> 
> This seems reasonable to me. I might even go a bit further and say
> that a client SHOULD (?) keep track of one server, the one it most
> recently got resources/parameters from and use that one first when
> releasing resources. If that fails, fall back to finding and trying a
> different server. I think this retains the spirit of what is on the
> current draft but also provides more flexibility in the event that a
> server-server protocol comes along.

To me it seems natural to let clients release things to the server
that allocated them by default, and if that isn't possible then the
client MAY release them to the next server it knows about.

When a server-server protocol is designed, this can be evaluated again,
but in the meantime it would be better to make things work in the
logical way without such server coordination.


> > #2 will permit mobile nodes to release adddresses that roam around
> > before we get a server-to-server protocol deployed in the market.

I don't see how this follows at all, especially if the two servers
involved are in different domains.  We haven't even started to figure
out that problem yet.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Wed Nov 17 15:35:15 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA21244
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 15:35:12 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA16082;
	Wed, 17 Nov 1999 15:31:10 -0500 (EST)
Date: Wed, 17 Nov 1999 15:31:10 -0500 (EST)
Message-Id: <199911171922.OAA16409@rotala.raleigh.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Charles Perkins <charliep@iprg.nokia.com> writes:

> > > 1.  Lets make the default that the client has no knowledge of servers
> > >     and can release addresses to the next server it hears (e.g.
> > >     reconfigure, reply, advertise [from reboot]).

> I think it would be better for clients, by default, to release resources
> to the server that allocated them.  We've already got this mechanism
> specified in the draft now.

Note: the current draft (my recollection of it anyway) says that
releasable resources can ONLY be released via the original server.
Are you saying this is not the case, or are you suggesting that
relaxing this would be OK?

Also, I'd prefer SHOULD to MAY because if it's a MAY some
implementations won't do it. Either we think all clients should try a
fallback server or we don't. I don't see the value in a MAY
here. Making it a MAY will result in non-uniform behavior, which will
likely cause problems for the sys admins using DHCP (i.e., not have a
uniform predicatable behavior across client implementations).

Thomas


From dhcp-v6@bucknell.edu  Wed Nov 17 16:40:30 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA17324
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 16:40:26 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA13525;
	Wed, 17 Nov 1999 16:35:19 -0500 (EST)
Date: Wed, 17 Nov 1999 16:35:19 -0500 (EST)
Message-Id: <38331D8A.FEECB3A5@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Thomas Narten wrote:

> Note: the current draft (my recollection of it anyway) says that
> releasable resources can ONLY be released via the original server.
> Are you saying this is not the case, or are you suggesting that
> relaxing this would be OK?

How about if we say that the client should try to release to the
original server, and if that fails try something else?

> Also, I'd prefer SHOULD to MAY because if it's a MAY some
> implementations won't do it. Either we think all clients should try a
> fallback server or we don't. I don't see the value in a MAY
> here. Making it a MAY will result in non-uniform behavior, which will
> likely cause problems for the sys admins using DHCP (i.e., not have a
> uniform predicatable behavior across client implementations).

I agree with this.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Wed Nov 17 17:57:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA18383
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 17:57:01 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA28572;
	Wed, 17 Nov 1999 17:54:00 -0500 (EST)
Date: Wed, 17 Nov 1999 17:54:00 -0500 (EST)
Message-Id: <38331914.F8F30E0B@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Mike,


> Mike Carney - SNT Internet Engineering wrote:
>
> 2) Stateful autoconfiguration and multiple IP addresses. When would a
> client need to request additional IP addresses over and above what the
> administrator has configured through site policy?
>
>         .........  the questions regarding multiple IP addresses and
> releasable resources are really parts of an overriding question: "Can
> a client request parts of a configuration, or does a DHCP server reserve
> and return a "configuration" for a client which contains IP address(es)
> as well as other parameters, and that "configuration" is treated as a
> black box by the client and the server?

Phrased this way, I think it is clear that some configuration
details should not be controlled so tightly by the server as to
preclude additional client interaction later in time.  The main
reason is that I think that some of the configuration needs for
a computer are controlled by the applications that are to be run
on that computer.  It may be impossible for the server to decide
which applications are likely to be run on a particular computer.
More to the point, I think it would be a terrible mistake to legislate
that servers always have to be configured by an administrator to
take care of all possible application configurations on all the
existing computing platforms in any sizable enterprise computing
environment.

Briefly, I believe that DHCP should let applications initiate
transactions as needed.

This should occur without requiring that applications be known
beforehand
and tightly bound to particular platforms.

Especially because there isn't any compelling reason for this overeager
form of rigid configuration.  The protocol has already been designed
to offer the appropriate amount of flexibility for creating applications
that can get configuration information as they need it.

The specific considerations for IP addresses, in this context, are
merely special cases of the same idea, but I will try to lay them
out separately in another note.


> Scenario B: Individual extensions
> 
> This scenario reflects the protocol as currently defined. Clients can
> request individual extensions (how? no parameter request list extension)

- A client should be able to request particular extensions by including
  those extensions in the DHCP request.
- No one ever asked for a "parameter request list" extension in all the
  years of discussion on the mailing list.  However, if there is general
  agreement that such an extension is needed, I don't see any technical
  barrier to putting it into the extensions draft.

> which could include one or more IP address extensions. Assume for this
> discussion that the protocol has been changed to include the following
> change which I proposed as part of "List of DHCPv6 Issues with
> responses, part 2 of 2": Advertise messages include IP address
> extensions with Prefix (S bit set) specifications advertising which
> prefixes are available and have IP addresses for assignment.

I thought it would be a good idea for Router Advertisements to have
prefix information (even for managed links) and then DHCP clients would
know what prefixes they need IP addresses for.

After re-reading part of RFC 2462, I see that this is possible but not
mandatory.  Perhaps we should make it mandatory.


Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Wed Nov 17 17:58:32 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA18981
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 17:58:28 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA27332;
	Wed, 17 Nov 1999 17:57:34 -0500 (EST)
Date: Wed, 17 Nov 1999 17:57:34 -0500 (EST)
Message-Id: <199911172245.RAA0000023067@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #3, Reply to RECONFIGURE? 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>> I view it as ACK explosion (client exploding acks towards the servers)
>> but no big deal.
>
>The common term that describes this situation (single multicast packet
>resulting in a flood of return ACKs) is ack implosion, even though
>explosion might seem like a more appropriate description.

I know I have seen this before.  I don't agree with the "common term".
But not asking anyone to change it.  Just making my view known.

>But I do think that this is a big deal. If you have ten nodes, no big
>deal. But if you have 100? or 1000? Such a large number of ACKs sent
>at the same time will almost certainly result in congestion and lost
>packets. Even worse, if the request packet is retransmitted, the exact
>same thing will likely happen again (i.e., slow servers lose), with it
>fairly likely that the *exact* same packets that were lost the first
>time being lost again. Hence, retrasmitting request messages is much
>less effective than in the unicast case.

My mail was not clear.  I mean't no big deal about whether its
"implosion" or "explosion" I agree with you on the above.
Funny this thing called "email" ----).............

>> A compromise might be to attach a suffix to this reconfigure type like
>> RCONFIGURE-INIT and have it mean as you say do a request via dhcp.
>> Then down the road if we figure out scalable multicast we add a new
>> message type called RECONFIGURE-INIT-W/ACK or some such name and new msg
>> type.
>
>> If we did this then would you and others be OK with how we would do the
>> hack now to prevent clients from responding to multiple multicasts of
>> this nature?

>I think so. One other detail I didn't mention before. It may also make
>sense to have the reconfigure message include a delay time value in
>it. Clients don't respond immediately to the Reconfigure, but delay a
>random amount of time between 0 and the delay time value in the
>received message. That way a server can arrange that clients contact
>it over a period of (say) ten minutes rather than all at once (which
>could result in similar problems as ACK implosion).

Thats a good idea.  Would be easy to add to the draft and easy to
implement.

/jim


From dhcp-v6@bucknell.edu  Wed Nov 17 17:58:48 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA19093
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 17:58:45 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA29395;
	Wed, 17 Nov 1999 17:56:00 -0500 (EST)
Date: Wed, 17 Nov 1999 17:56:00 -0500 (EST)
Message-Id: <199911172241.RAA0000021844@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>>    a DHCP Request message with the `C' bit set and no extensions.  If a
>>    client wishes to retain some of its network configuration parameters,
>>    but determines that others are no longer needed, it SHOULD enable
>>    the server to release allocated resources which are no longer in
>>    use by sending a DHCP Release message to the server, and including
>>    extensions to identify the unneeded items.  The client consults its
>>    list of resource-server associations in order to determine which
>>    server should receive the desired Release message.  The Release
>    MUST be transmitted to the server that provided the configuration
>>    parameters.

This is with a request.  I was speaking of the Release secton.  But this
is valid we would have to synch up the C bit with Release if we change.

>Note that I'm assuming that the only resources a client would release
>are "releasable resources", i.e., addresses that have explicit
>leases. There is no need to release other types of parameters. Right?

I agree.  None that I know of now.

>So, I guess my previous mail wasn't quite clear. Requiring that
>clients rememember which server they got a resource from and then
>specifying that it can only release resources by contacting the same
>server again (in the MUST sense) seems a bit restrictive. If we
>imagine a server-server protocol becoming available, being able to
>release a resource to any of the available servers seems desirable. It
>might still make sense (as an optimization) to have clients first
>attempt to release a resource via the same server it got it from, but
>if that fails, wouldn't it make sense to allow it to try releasing
>resources via another server? Or more generally, perhaps it makes
>sense to say a client always prefers a particular server, the one it
>most recently got parameters/resources from. That way, a new server
>could (say) send a reconfigure that causes the client to switch to the
>new server and then use that new server for subsequent communication,
>regardless of where it got resources from previously.

Agreed.

>> Or is it just a matter of saying that clients only release addresses?
>> And its that simple?

>I think this is in essence the question about whether we should
>generalize addresses into releasable resources. Certainly, we'll want
>to be able to release addresses. 

Agreed.

>> 1.  Lets make the default that the client has no knowledge of servers
>>     and can release addresses to the next server it hears (e.g. 
>>     reconfigure, reply, advertise [from reboot]).
>
>> 2.  But put a MAY in the protocol that permits clients to keep
>>     the address of servers to later release addresses from where
>>     it got them.

>This seems reasonable to me. I might even go a bit further and say
>that a client SHOULD (?) keep track of one server, the one it most
>recently got resources/parameters from and use that one first when
>releasing resources. If that fails, fall back to finding and trying a
>different server. I think this retains the spirit of what is on the
>current draft but also provides more flexibility in the event that a
>server-server protocol comes along.

Agreed.

/jim 


From dhcp-v6@bucknell.edu  Wed Nov 17 19:00:11 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA13051
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 19:00:10 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA24871;
	Wed, 17 Nov 1999 18:57:11 -0500 (EST)
Date: Wed, 17 Nov 1999 18:57:11 -0500 (EST)
Message-Id: <199911172304.SAA0000026047@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Thomas,

Only going to address your statements below as the issue you bring up
needs to be clarified.  I will address any issues from this mail in
other space in another mail thread.  

>Another simplification of this model is that the binding cache (which
>deals with retransmitted client requests) is no longer needed, as all
>requests are idempotent. In model B (below) the server (and client in
>the case of reconfigures) needs to distinguish a first request from
>subsequent retransmissions and respond appropriately.

The binding cache in DHCPv6 is there regardless of the model because as
DHCPv4, DHCPv6 needs to handle retransmission of requests.

To verify the request is a retransmit so we keep a cache as an
optimization to avoid a complete client database lookup.  We do not need
this because of the model.  

/jim




From dhcp-v6@bucknell.edu  Wed Nov 17 19:42:53 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA29318
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 19:42:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA11724;
	Wed, 17 Nov 1999 19:40:29 -0500 (EST)
Date: Wed, 17 Nov 1999 19:40:29 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE72@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releasable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain;
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


[Charlie Perkins]

> It seems to me that one of the sticking points in this discussion
> is some difficulty in imagining other releasable resources besides
> IP addresses.  I'd like to address that issue in two ways:
> 
> 1. I can imagine a _lot_ of releasable resources in the future, but
>    it would be a mistake to try to specify them before we can have
>    DHCPv6 as a Proposed Standard.  After that time, new extensions
>    to specify new releasable resources will come up for discussion
>    in a natural way, as long as we have a natural mechanism specified
>    for handling them.  Some possible examples of releasable resources:
>    = disk space
>    = network memory
>    = I/O devices
>    = Licensed software
>    = time-metered equipment
>    = ports
>    = protocol translators
> 
[Barr Hibbs]

I can think of a few interesting cases where parametric information supplied
by a server (which, by definition, is NOT a releasable resource, as it is
not "owned" by the DHCP server) might require treatment in a similar
fashion.

Consider a list of DNS server addresses.  Slow or no response from a
particular DNS server MAY be a good reason for a client to abandon use of
that server, and for a client to attempt to "replace" that address with
another one (especially if the server "list" contained only one address.)
In this case, the client is essentially saying to the server, "I can't use
an address you supplied me:  give me a replacement," which is not quite the
same thing as saying, "Please reconfigure me."

[Charlie Perkins]

> 2. However, in a way, imagining cool resources as in (1) even misses
>    the point.  I think the point is that DHCP is a way to configure
>    computers, and it seems clear to me that computers (especially
>    computers on networks) can be considered to be composed of  allocated
>    resources structured according to parametric values.  DHCPv6 makes
>    a nice way to collect such parameters and resources, after some
>    years of work.  I vote for keeping it that way.
> 
[Barr Hibbs]

This does seem to be a valid use of a DHCP server, so the question really
becomes "Is the proposed release mechanism appropriate for the [one known]
resource and likely parameters?"  Given that an answer is forthcoming, we
will also need to ask "What are the proposed semantics?" in order to
determine the [probable] complexity of implementations, which drives the
cost/benefit analysis.


[Jim Bound]

> > > Today we have no mechanism for servers to know about each others
> > > databases and entries.  So we assume that servers can't do this and
> say
> > > a client can keep the address of a server who gave them their address,
> > > because we know new servers can't "today" know about each others
> > > databases in a "standard" way.
> > 
> > > But..tomorrow it may be there is a standard way for servers to know
> > > about each other and clients can clearly just send the release to any
> > > server if that exists in Theory....if the standard exists.
> > 
> > > Here is a suggestion for discussion:
> > 
> > > 1.  Lets make the default that the client has no knowledge of servers
> > >     and can release addresses to the next server it hears (e.g.
> > >     reconfigure, reply, advertise [from reboot]).
> 
[Charlie Perkins]

> I think it would be better for clients, by default, to release resources
> to the server that allocated them.  We've already got this mechanism
> specified in the draft now.
> 
[Barr Hibbs]

While a server-to-server protocol for DHCPv6 MAY be developed in the future,
I think we are doing the larger community a huge disservice to predicate the
behavior of a DHCPv6 client and server today on the basis of unknown future
developments.  We need to specify something that will work without a
server-to-server protocol, with an eye towards its possible future
implementation, but not hold these drafts for months or years waiting on an
as-yet-to-be-started design effort.

[Jim Bound]

> > > 2.  But put a MAY in the protocol that permits clients to keep
> > >     the address of servers to later release addresses from where
> > >     it got them.
>  
[Thomas Narten]

> > This seems reasonable to me. I might even go a bit further and say
> > that a client SHOULD (?) keep track of one server, the one it most
> > recently got resources/parameters from and use that one first when
> > releasing resources. If that fails, fall back to finding and trying a
> > different server. I think this retains the spirit of what is on the
> > current draft but also provides more flexibility in the event that a
> > server-server protocol comes along.
> 
[Charlie Perkins]

> To me it seems natural to let clients release things to the server
> that allocated them by default, and if that isn't possible then the
> client MAY release them to the next server it knows about.
> 
> When a server-server protocol is designed, this can be evaluated again,
> but in the meantime it would be better to make things work in the
> logical way without such server coordination.
> 
[Barr Hibbs]

Absolutely!  Given that no inter-server protocol currently exists, I believe
that a client MUST release resources [well, the one we can agree on, anyway]
to the server who originally granted rights to it.  Otherwise, why even
bother with a release message?  I also believe that it is completely
appropriate to reconsider this approach in the future when an inter-server
protocol exists.




From dhcp-v6@bucknell.edu  Wed Nov 17 20:15:37 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA09895
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 20:15:31 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA20311;
	Wed, 17 Nov 1999 20:12:42 -0500 (EST)
Date: Wed, 17 Nov 1999 20:12:42 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE73@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releasable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


[Jim Bound]

> > Today we have no mechanism for servers to know about each others
> > databases and entries.  So we assume that servers can't do this and say
> > a client can keep the address of a server who gave them their address,
> > because we know new servers can't "today" know about each others
> > databases in a "standard" way.  
> 
> > But..tomorrow it may be there is a standard way for servers to know
> > about each other and clients can clearly just send the release to any
> > server if that exists in Theory....if the standard exists.  
> 
> > Here is a suggestion for discussion:
> 
> > 1.  Lets make the default that the client has no knowledge of servers
> >     and can release addresses to the next server it hears (e.g. 
> >     reconfigure, reply, advertise [from reboot]).
> 
> > 2.  But put a MAY in the protocol that permits clients to keep
> >     the address of servers to later release addresses from where
> >     it got them.
> 
[Thomas Narten]

> This seems reasonable to me. I might even go a bit further and say
> that a client SHOULD (?) keep track of one server, the one it most
> recently got resources/parameters from and use that one first when
> releasing resources. If that fails, fall back to finding and trying a
> different server. I think this retains the spirit of what is on the
> current draft but also provides more flexibility in the event that a
> server-server protocol comes along.
> 
[Barr Hibbs]

I am definitely missing something in this discussion.  Why would you permit
a client to release a resource to a server that might have no knowledge of
it, and could simply discard the release message because it has no way to
determine/communicate with the owner of the resource?  How does permitting a
client to send a release message to the "next server that happens along"
represent an improvement over dropping the resource on the floor?




From dhcp-v6@bucknell.edu  Wed Nov 17 20:15:44 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA09952
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 20:15:43 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA11425;
	Wed, 17 Nov 1999 20:14:41 -0500 (EST)
Date: Wed, 17 Nov 1999 20:14:41 -0500 (EST)
Message-Id: <199911180056.QAA00795@feller.mentat.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Tim Hartrick <tim@mentat.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.





All,

> 
> >From discussion by Thomas Narten and Jim Bound on this subject:
> 
> > > == Jim Bound
> 
> >   == Thomas Narten
> 
> > > Today we have no mechanism for servers to know about each others
> > > databases and entries.  So we assume that servers can't do this and say
> > > a client can keep the address of a server who gave them their address,
> > > because we know new servers can't "today" know about each others
> > > databases in a "standard" way.
> > 
> > > But..tomorrow it may be there is a standard way for servers to know
> > > about each other and clients can clearly just send the release to any
> > > server if that exists in Theory....if the standard exists.
> > 
> > > Here is a suggestion for discussion:
> > 
> > > 1.  Lets make the default that the client has no knowledge of servers
> > >     and can release addresses to the next server it hears (e.g.
> > >     reconfigure, reply, advertise [from reboot]).
> 
> I think it would be better for clients, by default, to release resources
> to the server that allocated them.  We've already got this mechanism
> specified in the draft now.
>

If we are going to have releaseable resources and we aren't going to
specify a server-to-server protocol to coordinate resource allocation then
I don't see how the default behavior can be anything but to release the
resources to the server which allocated them.  The alternative of releasing
a resource to any old server could result in servers believing resources
are in use when they are not and resources being released to servers which
aren't managing the resource being released.  This is bad.

This appears at first blush to require more complexity in the client since
the client will need to maintain a fair amount of state in order to make
sure that resources are released to the servers which allocated them.  This
in contrast to a "black-box" of configuration which is passed out to the client
once and only released by server request or lease expiration.

Since we expect many more clients to be multi-homed in the future I believe
that we can also expect that more clients will be interacting with more than
one DHCPv6 server and these multiple DHCPv6 servers could be completely un-
related to each other.  That is, the servers could be managing address space or
other resources from two different providers without any knowledge of the other servers' presence.

If we believe that this is going to be a common configuration then it might
be best to just bite the bullet and deal with the complexity that will allow
clients to handle getting resources from multiple, possibly unrelated, servers.

Does this seem like a reasonable way to view the environment in which DHCPv6
is going to be deployed?  I am not sure, but it seems completely plausible
if we consider mobile IPv6 clients roaming about within various unrelated
adminstrative domains.

> > > 2.  But put a MAY in the protocol that permits clients to keep
> > >     the address of servers to later release addresses from where
> > >     it got them.
> > 
> > This seems reasonable to me. I might even go a bit further and say
> > that a client SHOULD (?) keep track of one server, the one it most
> > recently got resources/parameters from and use that one first when
> > releasing resources. If that fails, fall back to finding and trying a
> > different server. I think this retains the spirit of what is on the
> > current draft but also provides more flexibility in the event that a
> > server-server protocol comes along.
> 
> To me it seems natural to let clients release things to the server
> that allocated them by default, and if that isn't possible then the
> client MAY release them to the next server it knows about.
> 
> When a server-server protocol is designed, this can be evaluated again,
> but in the meantime it would be better to make things work in the
> logical way without such server coordination.
>

As noted above, I think Charlie is right.  It is hard to see what good could
come from releasing resources to a server that didn't allocate them in the
first place.  I can imagine circumstances in which it would work but they
all require thinking about server-to-server protocols which is a rat-hole
that we desperately want to avoid I think.
 
> 
> > > #2 will permit mobile nodes to release adddresses that roam around
> > > before we get a server-to-server protocol deployed in the market.
> 
> I don't see how this follows at all, especially if the two servers
> involved are in different domains.  We haven't even started to figure
> out that problem yet.
> 

Once again I think Charlie has it right.  There are two cases.  The case
where there are multiple servers within one administrative domain handing
out a shared block of resources and the case where there are multiple servers
from possibly seperate administrative domains handing out items from completely
seperate blocks of resources.  The former will probably require a server-to-
server protocol to coordinate things.  The latter doesn't need a server-to-
server protocol since the resources being managed are disjoint.  A mobile node
is likely to find itself dealing with both of these circumstances.  I think we
need to make them both work, or at least make it possible for them to work in
the future, without running down the server-to-server protocol design rat-hole.




Mind you that I haven't completely bought off going with something that is
not the "black-box" approach that has been suggested.  However, if we go down
the path that allows clients to have more freedom then we need to develop
something that works properly in absence of a server-to-server protocol.
The only way I see that happening is if releasible resources are released
to the server from which they were allocated.

In the future we can develop a server-to-server protocol which will allow
clients to release these resources to a number of different servers.  How
the client learns about these other confederate servers is TBD.



Tim Hartrick
Mentat Inc.




From dhcp-v6@bucknell.edu  Wed Nov 17 20:17:32 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA10515
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 20:17:28 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA28098;
	Wed, 17 Nov 1999 20:15:49 -0500 (EST)
Date: Wed, 17 Nov 1999 20:15:49 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE74@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


[Mike Carney]

> > 2) Stateful autoconfiguration and multiple IP addresses. When would a
> > client need to request additional IP addresses over and above what the
> > administrator has configured through site policy?
> >
> >         .........  the questions regarding multiple IP addresses and
> > releasable resources are really parts of an overriding question: "Can
> > a client request parts of a configuration, or does a DHCP server reserve
> > and return a "configuration" for a client which contains IP address(es)
> > as well as other parameters, and that "configuration" is treated as a
> > black box by the client and the server?
> 
[Charlie Perkins]

> Phrased this way, I think it is clear that some configuration
> details should not be controlled so tightly by the server as to
> preclude additional client interaction later in time.  The main
> reason is that I think that some of the configuration needs for
> a computer are controlled by the applications that are to be run
> on that computer.  It may be impossible for the server to decide
> which applications are likely to be run on a particular computer.
> More to the point, I think it would be a terrible mistake to legislate
> that servers always have to be configured by an administrator to
> take care of all possible application configurations on all the
> existing computing platforms in any sizable enterprise computing
> environment.
> 
> Briefly, I believe that DHCP should let applications initiate
> transactions as needed.
> 
> This should occur without requiring that applications be known
> beforehand
> and tightly bound to particular platforms.
> 
> Especially because there isn't any compelling reason for this overeager
> form of rigid configuration.  The protocol has already been designed
> to offer the appropriate amount of flexibility for creating applications
> that can get configuration information as they need it.
> 
[Barr Hibbs]

Consider a RAS server.  It may not be started at the time a client requests
configuration data from a server, so if it is later started, the only means
for it to obtain the IP addresses it needs is to request them.  Requesting
multiple addresses in a single request message would be highly desirable.

As to the packaged configuration suggested by Mike as one interpretation of
the protocol behavior, let me observe that the v4 model works quite well,
permitting a client to request any option it wishes (or not), while a server
will deliver (without being asked to do so) the options specified by host
requirements, DHCP protocol, and/or local administrative site policy.  Since
it works well, why change the mechanism?


[Mike Carney]

> > Scenario B: Individual extensions
> > 
> > This scenario reflects the protocol as currently defined. Clients can
> > request individual extensions (how? no parameter request list extension)
> 
[Charlie Perkins]

> - A client should be able to request particular extensions by including
>   those extensions in the DHCP request.
> - No one ever asked for a "parameter request list" extension in all the
>   years of discussion on the mailing list.  However, if there is general
>   agreement that such an extension is needed, I don't see any technical
>   barrier to putting it into the extensions draft.
> 
[Barr Hibbs]

There are really two issues here:

(1) should a client be able to request multiple IP addresses (perhaps in
multiple requests)?

(2) should clients be able to request specific extensions at all?

A "yes" answer to either presupposes a mechanism to specify what is
requested, so a parameter request list is needed.


--Barr



From dhcp-v6@bucknell.edu  Wed Nov 17 23:16:28 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA26719
	for <dhc-archive@odin.ietf.org>; Wed, 17 Nov 1999 23:16:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA31046;
	Wed, 17 Nov 1999 23:10:23 -0500 (EST)
Date: Wed, 17 Nov 1999 23:10:23 -0500 (EST)
Message-Id: <199911180353.EAA02121@givry.inria.fr>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Francis Dupont <Francis.Dupont@inria.fr>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

 In your previous mail you wrote:

   I thought it would be a good idea for Router Advertisements to have
   prefix information (even for managed links) and then DHCP clients would
   know what prefixes they need IP addresses for.
   
=> I believe stateless prefixes (in router advertisements) will be a
subset of stateful prefixes (managed by DHCP) and in some case a
proper subset then DHCP clients could not rely on router advertisements
in order to know the list of prefixes.

   After re-reading part of RFC 2462, I see that this is possible but not
   mandatory.  Perhaps we should make it mandatory.
   
=> to keep a stateless service will give a fallback in case of a
temporary failure of the stateful service then I agree we should
make the announce of at least one prefix with the largest scope
(ie. a global prefix in common cases) by router advertisements
mandatory.

Regards

Francis.Dupont@inria.fr


From dhcp-v6@bucknell.edu  Thu Nov 18 11:19:54 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA17123
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 11:19:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA06530;
	Thu, 18 Nov 1999 11:12:55 -0500 (EST)
Date: Thu, 18 Nov 1999 11:12:55 -0500 (EST)
Message-Id: <199911181526.KAA27763@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: dhcp-v6@bucknell.edu (by way of Ralph Droms <dhcp-v6@bucknell.edu>)
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


(From Thomas Narten)
--------------------
Tim,

> If we are going to have releaseable resources and we aren't going to
> specify a server-to-server protocol to coordinate resource allocation then
> I don't see how the default behavior can be anything but to release the
> resources to the server which allocated them.  The alternative of releasing
> a resource to any old server could result in servers believing resources
> are in use when they are not and resources being released to servers which
> aren't managing the resource being released.  This is bad.

Let me clarify. I agree that releasing resources via the server one
got the resources from makes sense.  Having that be the default seems
fine.  However, the current draft goes further and essentially says:
you can't release via another server, period. The middle ground that
has been suggested is to first try releasing via the server one got
the resource from in the first place.  If that fails, however, then
fall back by trying other servers. Having a fall back might make
things work in the case that another server was available. This could
happen via the server-server protocol, but could also happen if a DHCP
server was retired and replaced by another with configuation
information migrated from one server to the other manually.  Thus,
there is no requirement that a server-server protocol be in place, but
we'd be leaving in hooks to make things work better in such an
environment.

I also agree that we shouldn't assume a server-server protocol is
present (since it doesn't quite exist yet), but we should also keep in
mind that changing clients later when/if we do have a server-server
protocol will be very difficult (i.e., we may not have the opporunity
of DHCPv6 clients are already deployed). Thus, anything we can do now
that helps the server-server case should be considered.

> Since we expect many more clients to be multi-homed in the future I
> believe that we can also expect that more clients will be
> interacting with more than one DHCPv6 server and these multiple
> DHCPv6 servers could be completely un- related to each other.  That
> is, the servers could be managing address space or other resources
> from two different providers without any knowledge of the other
> servers' presence.

I don't see right off why multi-homed sites implies multiple DHCP
servers.  Could you further elaborate please?

> If we believe that this is going to be a common configuration then
> it might be best to just bite the bullet and deal with the
> complexity that will allow clients to handle getting resources from
> multiple, possibly unrelated, servers.

If you think that the normal situation would be for a client to need
to talk to multiple DHCP servers to get its configuration, the current
protocol may well be inadequate for that. It has the clear notion that
you locate one DHCP server and then get all your configuration
information from it. There is no notion in the protocol that one

should be conversing with multiple servers simultaneously.  Do you
think this is a requirement?

> Does this seem like a reasonable way to view the environment in
> which DHCPv6 is going to be deployed?  I am not sure, but it seems
> completely plausible if we consider mobile IPv6 clients roaming
> about within various unrelated adminstrative domains.

Roaming about as in the Mobile IP (the protocol) sense? That might be
a somewhat different case. I.e., I can imagine that nodes supporting
Mobile IP might have special requirements to handle the case where
they need to talk to multiple DHCP servers (the one at the site they
are visiting, plus the one back home).  But we need to distinguish
between what is a general requirement for all nodes vs. those that
will be mobile. In addition, does the mobile case require any changes
to the current DHCP protocol? If so, we need to understand those
requirements.

Thomas



From dhcp-v6@bucknell.edu  Thu Nov 18 11:19:56 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA17147
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 11:19:55 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA14789;
	Thu, 18 Nov 1999 11:11:01 -0500 (EST)
Date: Thu, 18 Nov 1999 11:11:01 -0500 (EST)
Message-Id: <38341A2F.76334C09@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Francis Dupont wrote:

> => I believe stateless prefixes (in router advertisements) will be a
> subset of stateful prefixes (managed by DHCP) and in some case a
> proper subset then DHCP clients could not rely on router advertisements
> in order to know the list of prefixes.

The proper subset relationship may be true, but can still be handled by
allowing clients to request IP addresses based on what is advertised.
If a request arrives from a network using an advertised prefix,
the server could just as well supply additional IP addresses from
prefixes that were not advertised in the same DHCP Reply message.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Thu Nov 18 12:45:23 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA25411
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 12:45:21 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA26120;
	Thu, 18 Nov 1999 12:24:48 -0500 (EST)
Date: Thu, 18 Nov 1999 12:24:48 -0500 (EST)
Message-Id: <383430CF.5C4FB865@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Thomas,

> (From Thomas Narten)
> --------------------

> > Since we expect many more clients to be multi-homed in the future I
> > believe that we can also expect that more clients will be
> > interacting with more than one DHCPv6 server and these multiple
> > DHCPv6 servers could be completely un- related to each other.  That
> > is, the servers could be managing address space or other resources
> > from two different providers without any knowledge of the other
> > servers' presence.
> 
> I don't see right off why multi-homed sites implies multiple DHCP
> servers.  Could you further elaborate please?

I don't think that multi-homing _requires_ multiple DHCP servers,
but it certainly suggests the possibility of it.  This would be
particularly true if the multi-homing were for unrelated ISPs.


> If you think that the normal situation would be for a client to need
> to talk to multiple DHCP servers to get its configuration, the current
> protocol may well be inadequate for that.

I hope not!  See below...

>                                          It has the clear notion that
> you locate one DHCP server and then get all your configuration
> information from it.

With the exception of the (now likely to be changed) overspecified
behavior for DHCP Release, I don't remember any place where the existing
protocol is inadequate for dealing with multiple DHCP servers.  Thus,
I disagree with your statement.  I hope that you will help identify any
places within the protocol specification that might make such misleading
suggestions, so that we can repair the wording as soon as possible.

>                       There is no notion in the protocol that one
> should be conversing with multiple servers simultaneously.  Do you
> think this is a requirement?

Jim and I have, all along, tried to make sure that the DHCP protocol
does exactly provide for the possibility that a client will have
(perhaps simultaneous) transactions with multiple DHCP servers.
This is enabled by the transaction ID, the minimal linkage between
advertisement and request, the separation of control flow from
data items in the extensions, the modular isolation of state
information from the various control flows, and the avoidance of
requiring state information in the relays.

I think it should be a requirement that the protocol support access
to multiple DHCP servers simultaneously.  The client gets to decide
which servers to query, and the servers get to implement enough policy
information to help them serve the clients according to rational
administration.




> > Does this seem like a reasonable way to view the environment in
> > which DHCPv6 is going to be deployed?  I am not sure, but it seems
> > completely plausible if we consider mobile IPv6 clients roaming
> > about within various unrelated adminstrative domains.
> 
> Roaming about as in the Mobile IP (the protocol) sense? That might be
> a somewhat different case. I.e., I can imagine that nodes supporting
> Mobile IP might have special requirements to handle the case where
> they need to talk to multiple DHCP servers (the one at the site they
> are visiting, plus the one back home).

Please send any thoughts about support requirements for mobile nodes
to the mailing list.  Since the Internet will largely be composed of
mobile nodes by the time IPv6 is dominant, we'd better make sure they
get the support they need from DHCPv6.

>                                     But we need to distinguish
> between what is a general requirement for all nodes vs. those that
> will be mobile.

I am curious about what kind of distinction you would like to see
put into place.  I'd rather that distinctions not be made, except
in cases which are purely related to requirements from Mobile IP(v6).

>                 In addition, does the mobile case require any changes
> to the current DHCP protocol? If so, we need to understand those
> requirements.

I'd like to suggest rethinking the retransmission timeout and
exponential
backoff policy in light of the needs of mobile hosts.  But I won't
suggest
it, for the obvious reasons.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Thu Nov 18 12:46:03 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA25742
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 12:45:54 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA07461;
	Thu, 18 Nov 1999 12:23:10 -0500 (EST)
Date: Thu, 18 Nov 1999 12:23:10 -0500 (EST)
Message-Id: <199911181646.LAA13902@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2,
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0


(From Thomas Narten)
--------------------
Jim Bound <bound@zk3.dec.com> writes:

> >Another simplification of this model is that the binding cache (which
> >deals with retransmitted client requests) is no longer needed, as all
> >requests are idempotent. In model B (below) the server (and client in
> >the case of reconfigures) needs to distinguish a first request from
> >subsequent retransmissions and respond appropriately.

> The binding cache in DHCPv6 is there regardless of the model because as
> DHCPv4, DHCPv6 needs to handle retransmission of requests.

> To verify the request is a retransmit so we keep a cache as an
> optimization to avoid a complete client database lookup.  We do not need
> this because of the model.  

That is not what I understand from the draft. The binding cache is
needed not just as an optimization, but for correctness. Specifically,
it is needed so that a server (in the case of Request messages) can
distinguish between an initial (i.e., "new") request and a
retransmission. Request messages are not idempotent in the current
draft due to the the model. If a client makes three requests for
addresses, does the server send back the same three answers, or is
each Request message treated as a separate request requiring a
different set of addresses returned? The document specifically says:

>   While processing the Request, the server MUST first determine whether
>   or not the Request is a retransmission of an earlier DHCP Request
>   from the same client.  This is done by comparing the transaction-ID
>   to all those transaction-IDs in the XID_TIMEOUT binding cache
>   received from the same client during the previous XID_TIMEOUT
>   seconds.  If the transaction-ID is the same as one received during
>   that time, the server MUST take the same action (e.g., retransmit
>   the same DHCP Reply to the client) as it did after processing the
>   previous DHCP Request with the same transaction-ID.

In contrast, in DHCPv4, requests are idempotent and there is no need
(for correctness) for the server to distinguish between
retransmissions and "new" requests. 

Thomas



From dhcp-v6@bucknell.edu  Thu Nov 18 15:17:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA09612
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 15:17:37 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA28478;
	Thu, 18 Nov 1999 15:12:54 -0500 (EST)
Date: Thu, 18 Nov 1999 15:12:54 -0500 (EST)
Message-Id: <199911182003.MAA00984@feller.mentat.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Tim Hartrick <tim@mentat.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.





Thomas,

> 
> I don't see right off why multi-homed sites implies multiple DHCP
> servers.  Could you further elaborate please?
>

As an example, lets assume that I have a "standard" cable modem in my
house.  This modem and my cable company allow me to choose from any number
of providers or multiple providers.  If I did indeed have multiple
providers then I could be interacting with two different DHCPv6 servers
managed by those two providers.  They don't know anything about each
other and it makes no sense for me to attempt to release an address
that was allocated by provider A to the server administered by provider
B.  This is a unlikely to be common but given appropriate competition in
the market it won't be impossible.

There is also the mobility case.  I have a mobile node which has been
assigned an address from its home network via DHCPv6.  I now pick up this
mobile node and take it to some other administrative domain.  I need an
address from the new network and that address could well be assigned by
a local DHCPv6 server.  Once again I have scenario where the client needs
to interact will multiple DHCPv6 servers but those servers are completely
unrelated.  Attempting to release an address from my home network to
the DHCPv6 server of the local network is not what you want.

I agree that we probably don't want to say that the client MUST release
the resources to the original server.  That precludes the possibility of
designing server-to-server and server redundancy protocols.  All I am
saying is that in order to get things to work correctly in the more common
case we need to say that the client SHOULD release resources to the
original server and if that fails then they MAY release them to some other
known server with the understanding that absent some adminstrative co-
ordination, the description of which is out of scope for the draft, the
release may not work.  That's it.  Nothing more complicated than that.

> > If we believe that this is going to be a common configuration then
> > it might be best to just bite the bullet and deal with the
> > complexity that will allow clients to handle getting resources from
> > multiple, possibly unrelated, servers.
> 
> If you think that the normal situation would be for a client to need
> to talk to multiple DHCP servers to get its configuration, the current
> protocol may well be inadequate for that. It has the clear notion that
> you locate one DHCP server and then get all your configuration
> information from it. There is no notion in the protocol that one
> 
> should be conversing with multiple servers simultaneously.  Do you
> think this is a requirement?
>

I am not sure exactly.  It seems as though it has the potential to be
more common in the IPv6 world.  Is it an absolute requirement that we make
that sort of interaction bullet proof before we go to proposed standard.
I doubt it.
 
> > Does this seem like a reasonable way to view the environment in
> > which DHCPv6 is going to be deployed?  I am not sure, but it seems
> > completely plausible if we consider mobile IPv6 clients roaming
> > about within various unrelated adminstrative domains.
> 
> Roaming about as in the Mobile IP (the protocol) sense? That might be
> a somewhat different case. I.e., I can imagine that nodes supporting
> Mobile IP might have special requirements to handle the case where
> they need to talk to multiple DHCP servers (the one at the site they
> are visiting, plus the one back home).  But we need to distinguish
> between what is a general requirement for all nodes vs. those that
> will be mobile. In addition, does the mobile case require any changes
> to the current DHCP protocol? If so, we need to understand those
> requirements.
> 

Yes, in the mobile IPv6 sense.  I agree that mobility may well be a
special case that can be handled with mobility specific extensions to
ND, DHCPv6 or mobile IPv6.  However, if we are going down the path of
a relatively complex protocol for managing releasible resources then I
think we should try and do it in a general way.



Tim Hartrick
Mentat Inc.


From dhcp-v4@bucknell.edu  Thu Nov 18 19:08:12 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA28092
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 19:08:11 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA30287;
	Thu, 18 Nov 1999 19:07:22 -0500 (EST)
Date: Thu, 18 Nov 1999 19:07:22 -0500 (EST)
Message-Id: <199911182338.PAA76354@vihar.engr.sgi.com>
Errors-To: droms@bucknell.edu
Reply-To: eddiem@vihar.engr.sgi.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Edward Mascarenhas <eddiem@vihar.engr.sgi.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: DHCP server/Client behavior when options change while lease is valid
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: exmh version 2.0.2 2/24/98


If a client has a valid lease of an IP address but some of 
the options change (the subnet mask, for example), what 
should the server do when the client attempts to renew the 
lease. Should the server renew the lease returning 
the options requested (some of which may have changed) or 
cause the client to reinitialize.

Should the client check the set of options received 
each time a lease is renewed and take appropriate action if 
the some of the options have changed? 

Edward



From dhcp-v6@bucknell.edu  Thu Nov 18 19:51:52 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA16110
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 19:51:51 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA00755;
	Thu, 18 Nov 1999 19:49:16 -0500 (EST)
Date: Thu, 18 Nov 1999 19:49:16 -0500 (EST)
Message-Id: <199911190021.TAA0000028703@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Thomas,

Just to make sure.  You don't have an issue with clients requesting
things it needs even after it has its addresses.  It does not all have
to happen with one transaction.  DHCPv4 supports this today too and all
client server apps I am aware of do too.  So maybe I missing your point
on this particular issue?

thanks
/jim


From dhcp-v6@bucknell.edu  Thu Nov 18 19:52:45 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA16494
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 19:52:44 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA18323;
	Thu, 18 Nov 1999 19:50:59 -0500 (EST)
Date: Thu, 18 Nov 1999 19:50:59 -0500 (EST)
Message-Id: <199911190031.TAA0000022788@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Charlie,

>> > #2 will permit mobile nodes to release adddresses that roam around
>> > before we get a server-to-server protocol deployed in the market.
>
>I don't see how this follows at all, especially if the two servers
>involved are in different domains.  We haven't even started to figure
>out that problem yet.

I "think" you misread what I am saying above.  Number #2 is that the
node would take the server that gave the node the resource and be able
to release them.

Here is #2:

> > 2.  But put a MAY in the protocol that permits clients to keep
> >     the address of servers to later release addresses from where
> >     it got them.

It states that clients may be able to keep the addresses of servers to
release parameters.  Mobile nodes will need this until we do have
server-to-server protocol.

I think we are in violent agreement?  Except about the default behavior
and usage of RFC 2119.

thanks
/jim


From dhcp-v4@bucknell.edu  Thu Nov 18 19:55:13 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA17495
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 19:55:12 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA01466;
	Thu, 18 Nov 1999 19:54:50 -0500 (EST)
Date: Thu, 18 Nov 1999 19:54:50 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE78@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: RBHIBBS@msg.pacbell.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP server/Client behavior when options change while lease i
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> From: 	Edward Mascarenhas[SMTP:eddiem@vihar.engr.sgi.com]
> Sent: 	Thursday, November 18, 1999 4:08 PM
> 
> If a client has a valid lease of an IP address but some of 
> the options change (the subnet mask, for example), what 
> should the server do when the client attempts to renew the 
> lease. Should the server renew the lease returning 
> the options requested (some of which may have changed) or 
> cause the client to reinitialize.
> 
--using DHCPv4, the server has no way to force a client to reinitialize, the
server must wait until the client initiates a protocol exchange.  At that
time, the server's behavior should be a local policy decision, it is not
specified in the RFC.  It is quite common for a server to supply the most
current configuration data for all subsequent requests, but there is nothing
to prevent an implementation from caching the data supplied previously, and
simply replaying it.

> Should the client check the set of options received 
> each time a lease is renewed and take appropriate action if 
> the some of the options have changed? 
> 
Again, that is an implementation decision.  Consider the things most likely
to change in practice:  router addresses, DNS server addresses, and subnet
masks are changed by an administrator from time to time.  Since it is
possible for option values to change, a competent client implementation
probably anticipate this and recognize changes, taking an appropriate
action.  On the other hand, a client could choose not to request options
after receiving an IP address lease, deferring changes until the next
discover/offer cycle begins.

--Barr



From dhcp-v4@bucknell.edu  Thu Nov 18 19:56:03 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA17843
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 19:56:02 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA19740;
	Thu, 18 Nov 1999 19:55:37 -0500 (EST)
Date: Thu, 18 Nov 1999 19:55:37 -0500 (EST)
Message-Id: <14388.38628.313990.468056@spg-east-006>
Errors-To: droms@bucknell.edu
Reply-To: Peter.Memishian@East.Sun.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Peter Memishian <Peter.Memishian@East.Sun.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: DHCP server/Client behavior when options change while lease is valid
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: VM 6.43 under 20.4 "Emerald" XEmacs  Lucid
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit


 > If a client has a valid lease of an IP address but some of 
 > the options change (the subnet mask, for example), what 
 > should the server do when the client attempts to renew the 
 > lease. Should the server renew the lease returning 
 > the options requested (some of which may have changed) or 
 > cause the client to reinitialize.

there are a few controversial issues here.  first, while the rfc
permits the server to send a NAK to a renewing client, and permits the
renewing client to restart lease negotiation upon receipt of such a
NAK, i do not see the rfc state that the client is required to
restart.  at the very least, the NAK may have been lost in transit.
perhaps, however, the client has chosen to ignore the NAK and continue
to use the remainder of its existing lease.  so i do not see how the
server can "cause the client to reinitialize".

also, i'm of the opinion that a change in the subnet mask is similar
to a change in ip address, since the subnet mask directly affects the
client's ability to communicate using its ip address.  since it seems
to me that by granting the initial lease, the server has given the
client the right to use its ip address (which includes the related
broadcast and network addresses) for the duration of the lease, the
client's subnet mask should not be (technically) allowed to change
during that lease.  in practice, i'd question how often the subnet
mask could really change without being forced to renumber the client's 
ip addresses as well.

however, like i said, this is one man's opinion ;-)

 > Should the client check the set of options received each time a
 > lease is renewed and take appropriate action if the some of the
 > options have changed?

for portability, i'd say so.

-- 
meem


From dhcp-v6@bucknell.edu  Thu Nov 18 21:09:54 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA23914
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 21:09:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA30047;
	Thu, 18 Nov 1999 21:07:00 -0500 (EST)
Date: Thu, 18 Nov 1999 21:07:00 -0500 (EST)
Message-Id: <199911190152.UAA0000001044@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>> >Another simplification of this model is that the binding cache (which
>> >deals with retransmitted client requests) is no longer needed, as all
>> >requests are idempotent. In model B (below) the server (and client in
>> >the case of reconfigures) needs to distinguish a first request from
>> >subsequent retransmissions and respond appropriately.
>
>> The binding cache in DHCPv6 is there regardless of the model because as
>> DHCPv4, DHCPv6 needs to handle retransmission of requests.
>
>> To verify the request is a retransmit so we keep a cache as an
>> optimization to avoid a complete client database lookup.  We do not need
>> this because of the model.  

>That is not what I understand from the draft. The binding cache is
>needed not just as an optimization, but for correctness. Specifically,
>it is needed so that a server (in the case of Request messages) can
>distinguish between an initial (i.e., "new") request and a
>retransmission. Request messages are not idempotent in the current
>draft due to the the model. If a client makes three requests for
>addresses, does the server send back the same three answers, or is
>each Request message treated as a separate request requiring a
>different set of addresses returned? The document specifically says:

The purpose of the cache is first to distinguish btw retransmit and
request.  The issue of the model is pure serendipity.
But it is also an optimization to avoid a full database search at least
for XID_TIMEOUT.

Once that TIMEOUT happens the server will reply to the client with the
same information as we treat it as a new DHCPv6 Request so it is
idempotent.

Requests are idempotent in DHCPv6 by design, because we designed a
transaction-based protocol.  You MUST verify it is not a retransmit but
only for XID_TIMEOUT.  This is the norm for transaction based processing
since IBM invented it in 1979 with CICS as far as I am concerned. 
The DHCPv6 request could be reprocessed and an implementation can just
respond again.  But then we are violating the principles of transaction
based systems.  So we don't.

>>   While processing the Request, the server MUST first determine whether
>>   or not the Request is a retransmission of an earlier DHCP Request
>>   from the same client.  This is done by comparing the transaction-ID
>>   to all those transaction-IDs in the XID_TIMEOUT binding cache
>>   received from the same client during the previous XID_TIMEOUT
>>   seconds.  If the transaction-ID is the same as one received during
>>   that time, the server MUST take the same action (e.g., retransmit
>>   the same DHCP Reply to the client) as it did after processing the
>>   previous DHCP Request with the same transaction-ID.

>In contrast, in DHCPv4, requests are idempotent and there is no need
>(for correctness) for the server to distinguish between
>retransmissions and "new" requests. 

So we built a different protocol!!! But DHCPv4 handles retransmits too.
But I don't agree with your statements about idempotency.                    

We put the above section in the spec as an optimization not because we
had to because the Request can't be idempotent.  You are making
assumptions about the draft but not backing it up with detail in your
response.  So I am not sure how to respond other than if we took this
optimization out the protocol would still work.  But we want the
optimization in the protocol spec.  If you disagree thats fine thats why
we are discussing it on the mailing list.

Any protocol that does not differentiate btw new messages and
retransmits is a bad protocol, it makes no sense to me to have the
processing take place if its a retransmit in the same manner as it would
with a new message in any protocol.  It may be we have a computer
science disagreement here on correct design and behavior. And in fact
the DHCPv6 protocol will act like you want once the XID_TIMEOUT timer
goes off.

thanks
/jim


From dhcp-v4@bucknell.edu  Thu Nov 18 21:11:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA24792
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 21:11:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA03026;
	Thu, 18 Nov 1999 21:11:14 -0500 (EST)
Date: Thu, 18 Nov 1999 21:11:14 -0500 (EST)
Message-Id: <199911190138.UAA22275@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while lease i 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> On the other hand, a client could choose not to request options
> after receiving an IP address lease, deferring changes until the next
> discover/offer cycle begins.

Of course, such a client would probably find itself on the network
administration's Do Not Buy In the Future list very quickly.

			       _MelloN_


From dhcp-v6@bucknell.edu  Thu Nov 18 22:22:37 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA29251
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 22:22:37 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA12829;
	Thu, 18 Nov 1999 22:20:12 -0500 (EST)
Date: Thu, 18 Nov 1999 22:20:12 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942979940.24740.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DHCPv6 Input from Francis Dupont
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


>  - I prefer to have a 32 bit interface index (relative to the agent)
>    in place of the agent address + prefix-size. I am not convinced
>    address + prefix-size is enough for all topologies, an interface index
>    is clearer and easier to use (as a scope ID). This solves the alignement
>    problem to if link-local addresses are replaced by interface index +
>    link-local address.

If we go the above path this implies that a client will be inentified
by the ifindex on the relay plus the client's link-local address.
This implies that if a relay fails (and there is a second relay on the link)
the client will be viewed as a new client - it will have a new identity from
the perspective of the server. Likewise if the relay reboots since interface
indicies are not required to be the same after a reboot.

Thus I think we need to first decide what the desired behavior is
in the above two cases and from that derive what should identify the client.
Then we can decide what should go in the packet.

>  - the real source of the problem is in RFC 2462 5.5.3 "Router
>    Advertisement Processing" page 17 where you can find after two paragraphs
>    about M (managed) and O (other) bits this statement:
>    "For each Prefix-Information option in the Router Advertisement: a-e)"
>    obviously the text after this statement and the 5.6 "Configuration
>    Consistency" section next page (with "hosts accept the union of all
>    information ..." are misleading in the stateful autoconf context.

What is misleading?
The intent is that if the M bit is set and the router advertisements
contain e.g. 2 prefixes with the A flag set then the host will 
use stateless addrconf to form two addresses, and invoke DHCP to ask for
addresses and other configuration.

But typically I assume that if the routers send the M bit they wouldn't
set the A flag in any prefix advertisements meaning that all addresses
come from DHCP.

>  - another view of this can be:
>    * strong stateful is "stateful enabled, stateless disabled"
>    * weak stateful is "stateful and stateless enabled", ie.
>      when the stateful mode is enabled and a stateful service is available
>      then the stateless mode is disabled.
>    this can make the wording of RFC 2462 update easier, ie. the only thing
>    needed is an addition of "when the stateless mode is available" at
>    the middle of 5.5.3 section (For each Prefix-Information option...).

But isn't "stateless mode is available" the same as "one or more prefixes
have the A bit set"?
Thus I think we already have the necessary knobs.

   Erik



From dhcp-v6@bucknell.edu  Thu Nov 18 23:34:08 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA04658
	for <dhc-archive@odin.ietf.org>; Thu, 18 Nov 1999 23:34:01 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA05785;
	Thu, 18 Nov 1999 23:31:22 -0500 (EST)
Date: Thu, 18 Nov 1999 23:31:22 -0500 (EST)
Message-Id: <199911190416.XAA0000023151@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DHCPv6 Input from Francis Dupont 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>>  - I prefer to have a 32 bit interface index (relative to the agent)
>>    in place of the agent address + prefix-size. I am not convinced
>>    address + prefix-size is enough for all topologies, an interface index
>>    is clearer and easier to use (as a scope ID). This solves the alignement
>>    problem to if link-local addresses are replaced by interface index +
>>    link-local address.

>If we go the above path this implies that a client will be inentified
>by the ifindex on the relay plus the client's link-local address.
>This implies that if a relay fails (and there is a second relay on the link)
>the client will be viewed as a new client - it will have a new identity from
>the perspective of the server. Likewise if the relay reboots since interface
>indicies are not required to be the same after a reboot.

This is a problem.  We need to have as consistent state for a client in
the server database as possible to find them.

   A server maintains a collection of client records, called
   ``bindings''.  Each binding is uniquely identifiable by the ordered
   pair <link-local address, agent-address>, since the link-local
   address is guaranteed to be unique [20] on the link identified by the
  agent-address and prefix.  An implementation MUST support bindings
   consisting of at least a client's link-local address, agent-address,
  preferred lifetime and valid lifetime [20] for each client address.

If the agent address is replaced by interface-id how do we make sure it is 
unique on another lan where the link-local address is duplicated and in
fact the interface ID?  I don't think this will work and we should leave
the bindings to be what they are now.

But this may raise a bug in the specs.  It could be we need to state
that the agent-address-prefix is a global one as even a site-local
agent address prefix could theoretically be duplicated though not probable.
We can fix this by simply saying the agent-address must be a global
address.

>Thus I think we need to first decide what the desired behavior is
>in the above two cases and from that derive what should identify the client.
>Then we can decide what should go in the packet.

I think first it should be discussed what is wrong with the spec first
before we say it needs changing.  I don't think Francis was saying there
is an issue with the specs behavior but you are I think?

Can you articulate what is wrong with the spec as is presently on this
topic?

>>  - the real source of the problem is in RFC 2462 5.5.3 "Router
>>    Advertisement Processing" page 17 where you can find after two paragraphs
>>    about M (managed) and O (other) bits this statement:
>>    "For each Prefix-Information option in the Router Advertisement: a-e)"
>>    obviously the text after this statement and the 5.6 "Configuration
>>    Consistency" section next page (with "hosts accept the union of all
>>    information ..." are misleading in the stateful autoconf context.
>
>What is misleading?
>The intent is that if the M bit is set and the router advertisements
>contain e.g. 2 prefixes with the A flag set then the host will 
>use stateless addrconf to form two addresses, and invoke DHCP to ask for
>addresses and other configuration.
>
>But typically I assume that if the routers send the M bit they wouldn't
>set the A flag in any prefix advertisements meaning that all addresses
>come from DHCP.

If they don't do this typically it is an implementation issue how
statless and stateful addresses are managed on the end node and this has
always been the case in discussions in addrconf.  I asked for this to be
resolved as you may recall but we decided to permit the behavior of the
union Francis is speaking of and I agree with him, but believe we must
just live with it.  In my approach I will mask stateless and stateful
addresses as stateless or stateful and will take care of it as an end
node.  So I think Francis is correct, but I don't think its misleading
its just the way the addrconf spec left the issue.  Which is fine with
me and I think most implementors of Ipv6 stateless addrconf.

thanks
/jim



From dhcp-v6@bucknell.edu  Fri Nov 19 01:12:48 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id BAA22444
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 01:12:47 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id BAA30052;
	Fri, 19 Nov 1999 01:10:49 -0500 (EST)
Date: Fri, 19 Nov 1999 01:10:49 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942991720.7689.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #3, Reply to RECONFIGURE? (Final) 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


> I've been concerned for some time with the basic idea that the server
> multicasts (widely) to a set of clients, and the clients are required
> to ack. This results in ack implosion and there is no easy way around
> the problem. I.e., this is the reliable multicast problem, which we
> don't have good general solutions for.

Multicast protocols tend to deal with this by having the sender
of the multicast packet include a parameter to control over how long time
the receivers should do a random delay. That way, since the DHCP server knows
who many clients are listening on each multicast address, it can 
suggest a reasonable random delay to avoid overload.

Presumably such a mechanism is needed for both the RECONFIGURE/RECONFIGURE_ACK
and the RECONFIGURE_INIT/REQUEST/REPLY exchanges since they can both
cause implosion when the first message is multicast.

> I also think the same effect can be achieved by changing the semantics
> of RECONFIGURE. That is, instead of having the RECONFIGURE mean "send
> me an ACK", have it instead tell the client "the server has new info,
> use your normal DHCP protocol exchanges to get that info
> reliably". This message can be multicast a number of times, with the
> client ignoring subsequent retransmissions. That way the server can
> retransmit multicast RECONFIGUREs to ensure that everyone gets a copy,
> and it can also check for subseqent client REQUEST messages if it
> really wants to know who has received its RECONFIGURE. I.e., the
> subsequent REQUESTs are the ACK the server is looking for.

Making the RECONFIGURE* message(s) idenpotent seems like a good idea.
But even with a RECONFIGURE_INIT/REQUEST/REPLY exchange we need random
delays to avoid a request implosion.

  Erik



From dhcp-v6@bucknell.edu  Fri Nov 19 02:20:34 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id CAA07544
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 02:20:33 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id CAA22291;
	Fri, 19 Nov 1999 02:18:39 -0500 (EST)
Date: Fri, 19 Nov 1999 02:18:39 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.942995049.30439.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


Tim Hartrick wrote:

> Since we expect many more clients to be multi-homed in the future I believe
> that we can also expect that more clients will be interacting with more than
> one DHCPv6 server and these multiple DHCPv6 servers could be completely un-
> related to each other.  That is, the servers could be managing address space
> or other resources from two different providers without any knowledge of the
> other servers' presence.

Dealing with multi-homed machines, or to be more specific, multi-interfaced
or multi-linked machines doesn't need anything more than stating that
DHCPv6 operates on a per-interface basis plus some way to resolve
differences for single-valued parameters.
For instance, if the client receives time zone extensions from multiple
servers on different interfaces it must pick one time zone value somehow.
But if the client receives multiple NTP server extensions, it can
either internally have a list of all NTP servers on all interfaces,
or it can have NTP software that runs independently on each interface
(the latter doesn't make much sense for NTP, but it might make sense
for other extensions which contain lists of things).

Is there something more you believe DHCPv6 needs to do to handle
multi-interfaced clients?

To me this seems like a different beast than tracking releaseable 
resources per server on a single interface, especially if only IP
addresses are releaseable since IP addresses are inherently associated
with interfaces.

   Erik



From dhcp-v4@bucknell.edu  Fri Nov 19 03:24:27 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id DAA07565
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 03:24:24 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id DAA24708;
	Fri, 19 Nov 1999 03:23:42 -0500 (EST)
Date: Fri, 19 Nov 1999 03:23:42 -0500 (EST)
Message-Id: <4.2.0.58.19991118234727.00a8ea90@mail.join.com>
Errors-To: droms@bucknell.edu
Reply-To: robs@join.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Rob Stevens <robs@join.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"; format=flowed
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 

At 07:55 PM 11/18/99 -0500, Peter Memishian wrote:

>
>  > Should the client check the set of options received each time a
>  > lease is renewed and take appropriate action if the some of the
>  > options have changed?
>
>for portability, i'd say so.
>
>--

I think that is too sweeping, and that it is entirely
a matter for the client device. For dumb client
devices, which only know how to do "one thing"
it might well make sense to check parameters
which may have changed....but

.at the other end of the scale, unix workstations
are unlikely to be able to make use of much
of the information. reconfiguring DNS is not
going to change application which have already
initialised. Changing IP masks is going to
breka existing connections.. And how does
the DHCP client know which apps are interested
in which parameters..

basically, I think it is useless for desk top
computers to bother with changes to parameters
unless they reboot.

I guess that means that I wouldn't agree with Ted's
comment either.








From dhcp-v6@bucknell.edu  Fri Nov 19 03:55:41 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id DAA21276
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 03:55:39 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id DAA08117;
	Fri, 19 Nov 1999 03:54:26 -0500 (EST)
Date: Fri, 19 Nov 1999 03:54:26 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.943000285.16433.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


> [Barr Hibbs]
> 
> As to the packaged configuration suggested by Mike as one interpretation of
> the protocol behavior, let me observe that the v4 model works quite well,
> permitting a client to request any option it wishes (or not), while a server
> will deliver (without being asked to do so) the options specified by host
> requirements, DHCP protocol, and/or local administrative site policy.  Since
> it works well, why change the mechanism?

I've heard of interoperability problems in DHCPv4 since it isn't clear
what options the client need to explicitly ask for and what the server will
automatically provide. Thus a client might not need to ask for an option
for some server implementations but other server implementations might insist
on being asked to provide the option.

Isn't this something we should try to avoid in DHCPv6?

   Erik



From dhcp-v6@bucknell.edu  Fri Nov 19 07:31:44 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id HAA02141
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 07:31:44 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id HAA26236;
	Fri, 19 Nov 1999 07:27:17 -0500 (EST)
Date: Fri, 19 Nov 1999 07:27:17 -0500 (EST)
Message-Id: <199911191204.HAA01789@hygro.adsl.duke.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Charles Perkins <charliep@iprg.nokia.com> writes:

> > I don't see right off why multi-homed sites implies multiple DHCP
> > servers.  Could you further elaborate please?

> I don't think that multi-homing _requires_ multiple DHCP servers,
> but it certainly suggests the possibility of it.  This would be
> particularly true if the multi-homing were for unrelated ISPs.

"Possible" doesn't imply likely or desirable. What would be the
benefit of such a configuration? At the least, it would seem to
complicate client implementations (see below).

> > If you think that the normal situation would be for a client to need
> > to talk to multiple DHCP servers to get its configuration, the current
> > protocol may well be inadequate for that.

> I hope not!  See below...

> >                                          It has the clear notion that
> > you locate one DHCP server and then get all your configuration
> > information from it.

> With the exception of the (now likely to be changed) overspecified
> behavior for DHCP Release, I don't remember any place where the existing
> protocol is inadequate for dealing with multiple DHCP servers.  Thus,
> I disagree with your statement.  I hope that you will help identify any
> places within the protocol specification that might make such misleading
> suggestions, so that we can repair the wording as soon as possible.

Note: I am assuming that for the purposes of this dicussion, multiple
DHCP servers means one server has one prefix, another server has a
different prefix. I.e., the two servers do not have the same
information about what prefixes (or resources/parameters in general)
to give out and a client must contact both of them to get all of the
addresses/parameters it needs. That is what I understood Tim to be
getting at.

Where in the current spec does it call out for a client to ask for
addresses/parameters from more than one dhcp server?  My read of the
spec is that it says you pick one server and get all your resources
from it.

In a theroretical sense, the protocol doesn't disallow getting
resources from multiple servers in the above case. But because there
is no text calling for clients to actually do this (and the current
text certainly implies that one get all resources from one server), I
believe it is fair to say that DHCP as currently written does not
support getting resources from multiple servers in the case Tim
mentions.

> >                       There is no notion in the protocol that one
> > should be conversing with multiple servers simultaneously.  Do you
> > think this is a requirement?

> Jim and I have, all along, tried to make sure that the DHCP protocol
> does exactly provide for the possibility that a client will have
> (perhaps simultaneous) transactions with multiple DHCP servers.

Unless the draft clearly spells out when and under what conditions
simultaneous transactions with multiple servers are to take place,
clients won't implement the feature. 

> I think it should be a requirement that the protocol support access
> to multiple DHCP servers simultaneously.  The client gets to decide
> which servers to query, and the servers get to implement enough policy
> information to help them serve the clients according to rational
> administration.

Please point out specific text in the current draft that makes it
clear under what conditions clients are to expected to be interacting
with multiple servers simultaneously. I don't recall there being much
in the way of specific requirements here.

> >                                     But we need to distinguish
> > between what is a general requirement for all nodes vs. those that
> > will be mobile.

> I am curious about what kind of distinction you would like to see
> put into place.  I'd rather that distinctions not be made, except
> in cases which are purely related to requirements from Mobile
> IP(v6).

I suspect (but am not entirely sure) that nodes implementing mobile IP
may have special requirements with regard to DHCP (e.g., a need to
interact with servers at both the local and home site
simultaneously). In such cases, it would probably make sense to have
mobile nodes (as part of the mobile IP spec) specify those
requirements as part of what implementing mobile IP means. It is not
clear to me that all clients would need to implement this. In any
case, we need to talk specifics here in order to decide whether
somethign is a general requirement or one appropriate for mobile
nodes.

> >                 In addition, does the mobile case require any changes
> > to the current DHCP protocol? If so, we need to understand those
> > requirements.

> I'd like to suggest rethinking the retransmission timeout and
> exponential backoff policy in light of the needs of mobile hosts.
> But I won't suggest it, for the obvious reasons.

The IESG message is clear here. Protocols must be congestion
friendly. The base case against which comparisons are made is TCP
(which has expential backoff, slow start, etc.)  If a protocol wants
to deviate from this, the burden is on the protocol to make the case
for why it's congestion strategy is different and why its deviation
from the norm does is acceptable for the environment into which it
will be deployed.

Thomas


From dhcp-v6@bucknell.edu  Fri Nov 19 08:00:43 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id HAA02142
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 07:31:44 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id HAA30982;
	Fri, 19 Nov 1999 07:29:35 -0500 (EST)
Date: Fri, 19 Nov 1999 07:29:35 -0500 (EST)
Message-Id: <199911191225.HAA01855@hygro.adsl.duke.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Tim,

> > I don't see right off why multi-homed sites implies multiple DHCP
> > servers.  Could you further elaborate please?
> >

> As an example, lets assume that I have a "standard" cable modem in my
> house.  This modem and my cable company allow me to choose from any number
> of providers or multiple providers.  If I did indeed have multiple
> providers then I could be interacting with two different DHCPv6 servers
> managed by those two providers.

My expectation would be that in this case the connectivity provider
would arrange to have your DHCP packets delivered to the DHCP server
of your ISP and NOT to the other DHCP servers. Or, the various DHCP
servers would get the client requests, but only the client's selected
ISP would respond. In both cases, the client would appear to be
interacting with only 1 DHCP server, not multiple ones, and I think
think this should work fine today.

> They don't know anything about each other and it makes no sense for
> me to attempt to release an address that was allocated by provider A
> to the server administered by provider B. This is a unlikely to be
> common but given appropriate competition in the market it won't be
> impossible.

Agreed. But the server can simply return a "no binding" error if a
client attempts to release a resource and the server has no record of
that binding. And in the case where the connectivity provider arranges
to have all DHCP packets delivered only to the client's ISP, this
would not be an issue.

As a data point, I have an ADSL connection to a telco (GTE). GTE also
provides ISP services. However, my ISP is not GTE, it is someone
else. I use DHCP to get addresses and configuration info. It just
works without any special configuration on my part. GTE arranges to
tunnel all of my non-local traffic via bridging to my ISP.

> There is also the mobility case.  I have a mobile node which has been
> assigned an address from its home network via DHCPv6.  I now pick up this
> mobile node and take it to some other administrative domain.  I need an
> address from the new network and that address could well be assigned by
> a local DHCPv6 server.  Once again I have scenario where the client needs
> to interact will multiple DHCPv6 servers but those servers are completely
> unrelated.  Attempting to release an address from my home network to
> the DHCPv6 server of the local network is not what you want.

I view this is a special case that applies to mobile IP nodes, and not
to (stationary) clients in general. This case is also different than
the one described above because I believe the node will be aware in
this case that it is operating in two realms (the local one and the
home one) and needs to do special things in some contexts because of
that (e.g., it knows to distinguish interactions with its home site
from those at the local site).

> I agree that we probably don't want to say that the client MUST release
> the resources to the original server.  That precludes the possibility of
> designing server-to-server and server redundancy protocols.  All I am
> saying is that in order to get things to work correctly in the more common
> case we need to say that the client SHOULD release resources to the
> original server and if that fails then they MAY release them to some other
> known server with the understanding that absent some adminstrative co-
> ordination, the description of which is out of scope for the draft, the
> release may not work.  That's it.  Nothing more complicated than
> that.

I think we are in general agreement here.

Thomas


From dhcp-v6@bucknell.edu  Fri Nov 19 08:04:59 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id IAA19405
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 08:04:59 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA32014;
	Fri, 19 Nov 1999 08:02:39 -0500 (EST)
Date: Fri, 19 Nov 1999 08:02:39 -0500 (EST)
Message-Id: <199911191228.HAA01867@hygro.adsl.duke.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

> Just to make sure.  You don't have an issue with clients requesting
> things it needs even after it has its addresses.  It does not all have
> to happen with one transaction.  DHCPv4 supports this today too and all
> client server apps I am aware of do too.  So maybe I missing your point
> on this particular issue?

I don't have an issue with this. I.e., it is fine for a client to
later ask for (say) an option requesting a list of its web proxies.

But that is a different question than the one about requesting
additional addresses.

Thomas


From dhcp-v6@bucknell.edu  Fri Nov 19 08:05:18 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id IAA19580
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 08:05:17 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA01722;
	Fri, 19 Nov 1999 08:04:44 -0500 (EST)
Date: Fri, 19 Nov 1999 08:04:44 -0500 (EST)
Message-Id: <199911191247.HAA02068@hygro.adsl.duke.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Jim,

> The purpose of the cache is first to distinguish btw retransmit and
> request.

Agreed.

> The issue of the model is pure serendipity. But it is also an
> optimization to avoid a full database search at least for
> XID_TIMEOUT.

Disagree.

If the binding cache is just an optimization, then I should be able to
remove it and have the protocol work correctly.

But consider two request messages from a client requesting additional
addresses. If they are unrelated (i.e., the second is not a
retransmission but a different request), the server would respond with
different answers to the two requests. If the second message was a
retransmission, the server would need to respond to the retransmission
with exactly the same information as it responded to the first
message.

My understanding of what idempotent means is that if you reexecute the
same request multiple times, you get the right answer. But in the case
of requesting additional addresses, the answer you get very much
depends on what previous requests have already been processed. So it
is very important for the server to distinguish original requests from
retransmissions and process them differently.

Consider DHCPv4 as a comparison. No binding cache is needed because if
a client asks for an address from the server 3 times, it gets back the
same 3 answers. The server doesn't need to check whether a request is
new or a retransmission, it processes it in the same way and gives
back the proper answer. (It might want a cache as a performance
optimization, but it is not required for correctness.)

Thus, I assert that removal of the binding cache in DHCPv6 would
result in the server returning differnt answers than it would
otherwise. Hence, it is not just a performance optimization, it is
needed for correctness.

> Once that TIMEOUT happens the server will reply to the client with the
> same information as we treat it as a new DHCPv6 Request so it is
> idempotent.

It seems to me that the addition of the transaction ID and the binding
cache is what makes the transaction idempotent. My point is that the
address allocation model makes the high-level request itself
non-idempotent. I agree that the cache takes care of the problem with
this model (but that it is required, not optional).

> Any protocol that does not differentiate btw new messages and
> retransmits is a bad protocol, it makes no sense to me to have the
> processing take place if its a retransmit in the same manner as it would
> with a new message in any protocol.

I don't agree with this as a blanket statement. Consider a time of day
server. The cost of looking up the time on most machines is trivial,
and the cost of reexecuting retransmitted requests is small. Having a
cache of recent response also has complexity and cost (e.g., requires
memory), and this cost is not necessarily lower than reexecuting a
retransmitted request.

Thomas


From dhcp-v4@bucknell.edu  Fri Nov 19 09:52:53 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA09278
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 09:52:49 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA01464;
	Fri, 19 Nov 1999 09:51:55 -0500 (EST)
Date: Fri, 19 Nov 1999 09:51:55 -0500 (EST)
Message-Id: <19991119134921.22174.rocketmail@web1904.mail.yahoo.com>
Errors-To: droms@bucknell.edu
Reply-To: anassar2000@yahoo.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ayman Nassar <anassar2000@yahoo.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Dynamic assignment of IP address to a relay agent interface
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0


I was reading draft-ietf-dhc-agent-options-07.txt
"DHCP Relay agent Information Option" . Is it possible
for a relay agent with a WAN interface like ATM, with
no IP address set to dynamically assign an IP address
to one of its subinterfaces that are connected to a
LAN (DHCP clients) via a PVC.

Under section 3.1 "Agent circuit ID suboption",
(option 82) the client could receive IP configuration
information from the DHCP server via the relay agent
using an identifier, which in this case could be
either the router sub-interface number or the ATM VC
number. 

Is it also possible for the agent itself, to receive
IP info about itself from the DHCP server. If so will
the agent act as both a client and an agent, when it
acts as a client, does it have to (must) add a
loopback address of itself to be identified by the
DHCP server, and it will also probably have to utilize
option 82 again, to define which sub-interface is
requesting the IP address.

Am I missing something here ?

Thanks

=====
Ayman Nassar
Product Manager
Ericsson IP Infrastructure
301-586-8200 ext 644
__________________________________________________
Do You Yahoo!?
Bid and sell for free at http://auctions.yahoo.com


From dhcp-v4@bucknell.edu  Fri Nov 19 09:53:52 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA09738
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 09:53:49 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA28839;
	Fri, 19 Nov 1999 09:53:35 -0500 (EST)
Date: Fri, 19 Nov 1999 09:53:35 -0500 (EST)
Message-Id: <199911191418.JAA24967@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


I don't want to create needless controversy, but I'd like to point out
a couple of things:

> reconfiguring DNS is not going to change application which have
> already initialised.

This depends on the application, and the resolver.   The BIND 8
resolver does indeed appear cache the contents of the resolv.conf
file, but this is only a problem for long-running applications -
anything that runs for short periods of time isn't going to have a
problem with that.   And I'd argue that it would probably be better if
the resolver statted the file once in a while to see if it's changed.
I'm not sure what the BIND 9 resolver does.   Personally, the idea
that you have to restart applications to see a new resolv.conf file
seems like a bug to me.

> Changing IP masks is going to breka existing connections..

I can't speak for all unix implementations here, but certainly
implementations that are based on the 4.4BSD stack do not have this
problem - if your IP address stays the same and your netmask changes,
connections will continue.  In fact, you can completely lose your IP
address, and as long as you get the same one back before the next time
a particular TCB is used, the connection will not be broken.

The real problem with the 4.4BSD stack is route caching - TCBs have
cached routes, so if the routing changes in a way that matters to a
particular connection, the connection doesn't switch to the new
route.   IMHO, though, this is also a bug.

> And how does the DHCP client know which apps are interested in which
> parameters..

It shouldn't matter to the DHCP client - the client should just do
what the server tells it, and let the applications decide whether or
not they care.

Basically, what I'm saying here is that the DHCP client should *try*
to do the right thing, and if applications do the wrong thing, that's
a problem the authors of the applications should solve.   In the
NetBSD community, we certainly have the intention of getting to the
point where the kernel and applications are maximally responsive to 
changes like this.

That's not to say your take on things is wrong for all situations, but
I think that in the cases where what you have said is correct, the
host is probably not using DHCP anyway.  For those sites where it is
considered preferable to have even Unix servers use DHCP to get their
IP addresses, I think what you've said is probably true, and a good
Unix DHCP client should probably have a switch that lets you control
what it does when parameters change.

The ISC DHCP client actually lets you completely override what the
server sends if you want, and also lets you override what the server
has sent in a conditional way, based on what the server sent (e.g.,
you can say "use 127.0.0.1 for the name server unless the domain-name
option is "firewalled.com", because I happen to know that
firewalled.com has a firewall that blocks DNS, so my local name server
won't work - only when I'm at firewalled.com, use the name server list
the DHCP server supplies).

			       _MelloN_


From dhcp-v6@bucknell.edu  Fri Nov 19 10:58:42 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA07514
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 10:58:40 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA04273;
	Fri, 19 Nov 1999 10:55:03 -0500 (EST)
Date: Fri, 19 Nov 1999 10:55:03 -0500 (EST)
Message-Id: <38357236.2A3D1A31@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Thomas Narten wrote:
> 
> Charles Perkins <charliep@iprg.nokia.com> writes:

> > I don't think that multi-homing _requires_ multiple DHCP servers,
> > but it certainly suggests the possibility of it.  This would be
> > particularly true if the multi-homing were for unrelated ISPs.
> 
> "Possible" doesn't imply likely or desirable. What would be the
> benefit of such a configuration? At the least, it would seem to
> complicate client implementations (see below).

DHCP(v6) is at the service of network computers and applications.
DHCP is not the application.  It is constructed to offer benefits
to applications.  If an application needs to get configuration
information from each of two DHCP servers at different ISPs, then
it is the job of DHCP to make that happen.  The benefit of having
DHCP work to serve that possibility is to simplify application
programming.  Insofar as applications care about multi-homing and
multiple ISPs, I think it is quite clear that they will need such
features from DHCP.  If you want to argue that applications should
not ever care about multi-homing and ISPs, then I can't add anything
more to that discussion except my intuition that it would be a
terrible mistake to remove the relevant features from DHCP.


> Where in the current spec does it call out for a client to ask for
> addresses/parameters from more than one dhcp server?  My read of the
> spec is that it says you pick one server and get all your resources
> from it.

Then you are reading something into the spec that isn't anywhere in
there.  The job of a specification is to make protocol operations
possible,
not to tell clients when to use the protocol.  The specification also
does not say what kind of routers to use, nor how many bits per second
of bandwidth is required, nor whether mobile computers can use the
messages, nor whether a collision-detection scheme is needed at
layer-2, nor whether the server must store its data using LDAP, nor
any of a thousand other details.  All of these thousand other details
are ABSOLUTELY CRUCIAL for lots of implementations, and yet the
specification does not discuss them.

I claim this is a very good feature.

I claim this very good feature opens up quite a broad system
design space for application developers and network administrators.

The specification does not have to tell clients when to use the
protocol.

> In a theroretical sense, the protocol doesn't disallow getting
> resources from multiple servers in the above case.

Your use of the word theoretical here would indicate that you do
not believe that practice follows theory.  On the other hand,
I believe that it is crucial to have a realistic theory with
clear rules for implementation in any protocol specification.
Let's make the rules clear and give as much latitude as possible
for implementations , and, subsequently, applications using the
implementations.

It is a mistake to follow the road of prescribing every possible
application interaction with a protocol.  Not to mention time-consuming,
tedious, and self-limiting.  Such an authoritarian view of protocol
design is almost always self-defeating, I think.

>                                                    But because there
> is no text calling for clients to actually do this (and the current
> text certainly implies that one get all resources from one server), I
> believe it is fair to say that DHCP as currently written does not
> support getting resources from multiple servers in the case Tim
> mentions.

By now, I hope my reasons for disagreeing totally with this paragraph
are very clear.  It is not fair to say that our specification does
not support Cisco routers just because we don't say how to use them.
It is not fair to say that our spec does not support mobile computers
just because we don't say how to use them.  I think your contention here
is, I wish I could say it more tactfully but I don't know how, totally
wrong.


> Unless the draft clearly spells out when and under what conditions
> simultaneous transactions with multiple servers are to take place,
> clients won't implement the feature.

Again, I think this statement is totally wrong.

As another example, our specification does not say how to a server
should operate Mobile IP at the same time it operates DHCP.

Perhaps this is a danger of thinking in state machines.  If that is
the sticking point, then I want to be sure that our protocol is able
to be understood without reference to state machines, even if a state
machine is helpful for the purposes of implementation.

> > I think it should be a requirement that the protocol support access
> > to multiple DHCP servers simultaneously.  The client gets to decide
> > which servers to query, and the servers get to implement enough policy
> > information to help them serve the clients according to rational
> > administration.
> 
> Please point out specific text in the current draft that makes it
> clear under what conditions clients are to expected to be interacting
> with multiple servers simultaneously. I don't recall there being much
> in the way of specific requirements here.

The protocol is specifically designed to support interactions with
multiple servers.  We could insert that statement.  Where do you
think it belongs?


> I suspect (but am not entirely sure) that nodes implementing mobile IP
> may have special requirements with regard to DHCP (e.g., a need to
> interact with servers at both the local and home site
> simultaneously). In such cases, it would probably make sense to have
> mobile nodes (as part of the mobile IP spec) specify those
> requirements as part of what implementing mobile IP means. It is not
> clear to me that all clients would need to implement this. In any
> case, we need to talk specifics here in order to decide whether
> somethign is a general requirement or one appropriate for mobile
> nodes.

Well, I'm ready to talk specifics as soon as anything becomes known.

We implemented Mobile IP with DHCPv4 more than once.  The problem we
had was with the state machine implementation of DHCPv4, and the
drastic inconvenience or impossibility of starting multiple state
machines.
That problem has been avoided in DHCPv6, to the best of my current
understanding.

Cbarlie P.


From dhcp-v6@bucknell.edu  Fri Nov 19 11:56:37 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA00724
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 11:56:35 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA15557;
	Fri, 19 Nov 1999 11:52:47 -0500 (EST)
Date: Fri, 19 Nov 1999 11:52:47 -0500 (EST)
Message-Id: <38357464.D95B1183@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Tim Hartrick wrote:

> > There is also the mobility case.  I have a mobile node which has been
> > assigned an address from its home network via DHCPv6.  I now pick up this
> > mobile node and take it to some other administrative domain.  I need an
> > address from the new network and that address could well be assigned by
> > a local DHCPv6 server.  Once again I have scenario where the client needs
> > to interact will multiple DHCPv6 servers but those servers are completely
> > unrelated.  Attempting to release an address from my home network to
> > the DHCPv6 server of the local network is not what you want.

Thomas Narten wrote (in response to Tim Hartrick):

> I view this is a special case that applies to mobile IP nodes, and not
> to (stationary) clients in general. This case is also different than
> the one described above because I believe the node will be aware in
> this case that it is operating in two realms (the local one and the
> home one) and needs to do special things in some contexts because of
> that (e.g., it knows to distinguish interactions with its home site
> from those at the local site).

This again relates to the discussion about whether clients should
be instructed when to use a protocol.  Since you see here how a Mobile
IP client might use two different servers, it makes sense to you.  Yet,
you do not allow for the operation to make sense _in the abstract_
(or, perhaps, "theoretical") case.  I claim that it is better to get
the protocol right in the abstract, knowing that there will be specific
cases.  I do not wish to get involved with listing the special cases
whereby clients might wish to interact with multiple DHCP servers.

Charlie P.


From dhcp-v6@bucknell.edu  Fri Nov 19 11:57:17 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA00955
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 11:57:15 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA25001;
	Fri, 19 Nov 1999 11:55:29 -0500 (EST)
Date: Fri, 19 Nov 1999 11:55:29 -0500 (EST)
Message-Id: <199911191623.LAA0000014560@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Thomas,

>> The issue of the model is pure serendipity. But it is also an
>> optimization to avoid a full database search at least for
>> XID_TIMEOUT.

>Disagree.

Do you disagree it optimizes retransmission?

Or do you disagree that I am telling you that we did not put this in
here because the protocol would not work?  Thats not what charlie and I
did..

>If the binding cache is just an optimization, then I should be able to
>remove it and have the protocol work correctly.

It will work and I rechecked that last night.

>But consider two request messages from a client requesting additional
>addresses. If they are unrelated (i.e., the second is not a
>retransmission but a different request), the server would respond with
>different answers to the two requests. If the second message was a
>retransmission, the server would need to respond to the retransmission
>with exactly the same information as it responded to the first
>message.

Not true.  If we had no binding cache the server would return the same
information simply because it had the information stored on the
database.  If the client requested new information we did not give it
previously the server would provide new information.

>My understanding of what idempotent means is that if you reexecute the
>same request multiple times, you get the right answer. But in the case
>of requesting additional addresses, the answer you get very much
>depends on what previous requests have already been processed. So it
>is very important for the server to distinguish original requests from
>retransmissions and process them differently.

Agree on idempotent.  1 OR 1 does not change 1.  

If the client provides 3 address extensions and in the database the
server has given them 3 then 3 will be returned.

if the client asks for 4 addresses and in the database the server only
has 3 addresses and the client can have 4 addresses the server will
return the original 3 addresses and an additional address. 

But now we head into the area I was waiting to come up and the real
issue for this multiple addresses and I think we do have a bug in the
spec we did not fix since our holding pattern since minneapolis meeting.
but I will bring it up now but was hoping you would directly.

If the client requests 1 address now continuing the discussion I would
return the client one of the addresses I have already given it as a
server implementor.  But this is not clear in the spec and if I don't do
it (implement) this way the REQUEST would not be idempotent if I give
the client a new address.

That is why we need to do one of the following things as I see it in the
protocol:
  
  1.  Add a bit in the request that is a "pure" request for a brand 
      new address or addresses and addr extentions would follow.

  2.  The client must ask  for all its addresses back and an additional
      extension if it wants a new one.  This would be done by the 
      client providing multiple addr extentions in the request.

  3.  Or we don't permit this behavior and all the client does
      is have a bit set in the request header "give me my addresses"
      an no addr extensions are provided by the client request msg.

I would like to see us adopt #2 above and is the compromise.

>Consider DHCPv4 as a comparison. No binding cache is needed because if
>a client asks for an address from the server 3 times, it gets back the
>same 3 answers. The server doesn't need to check whether a request is
>new or a retransmission, it processes it in the same way and gives
>back the proper answer. (It might want a cache as a performance
>optimization, but it is not required for correctness.)

We did not build the binding cache for this reason but I see your point.
See my above statements.  What I am saying is a correct server
implementor would not permit the idempotent problem. But there is a bug
and it could.  But its an implementation protocol clarification we need
or do as DHCPv4 does which I find to limiting for DHCPv6. 
 
>Thus, I assert that removal of the binding cache in DHCPv6 would
>result in the server returning differnt answers than it would
>otherwise. Hence, it is not just a performance optimization, it is
>needed for correctness.

I don't agree see above.  Also the once the binding cache times out the
problem happens anyway.  We don't keep the cache forever.  

>> Once that TIMEOUT happens the server will reply to the client with the
>> same information as we treat it as a new DHCPv6 Request so it is
>> idempotent.

>It seems to me that the addition of the transaction ID and the binding
>cache is what makes the transaction idempotent. My point is that the
>address allocation model makes the high-level request itself
>non-idempotent. I agree that the cache takes care of the problem with
>this model (but that it is required, not optional).

Not once the cache times out.
If the cache times out and the client retransmits then the server must
respond by whats in the database.  This is probably not likely but could
happen.  Ergo why I say above there is bug here we need to fix and I
provided my view of that fix.

>> Any protocol that does not differentiate btw new messages and
>> retransmits is a bad protocol, it makes no sense to me to have the
>> processing take place if its a retransmit in the same manner as it would
>> with a new message in any protocol.

>I don't agree with this as a blanket statement. Consider a time of day
>server. The cost of looking up the time on most machines is trivial,
>and the cost of reexecuting retransmitted requests is small. Having a
>cache of recent response also has complexity and cost (e.g., requires
>memory), and this cost is not necessarily lower than reexecuting a
>retransmitted request.

I understand that argument.  But I am biased as you say the memory look
ups is always a cost and has to be re-evaluated often in systems as they
grow and change.  Its also a maintenance headache.  If I can avoid a
lookup its always a good thing IMO.

/jim


From dhcp-v6@bucknell.edu  Fri Nov 19 11:57:19 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA00966
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 11:57:17 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA22197;
	Fri, 19 Nov 1999 11:54:23 -0500 (EST)
Date: Fri, 19 Nov 1999 11:54:23 -0500 (EST)
Message-Id: <383577B8.6046FD60@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2,
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Thomas Narten wrote, in response to Jim Bound:

>> The issue of the model is pure serendipity. But it is also an
>> optimization to avoid a full database search at least for
>> XID_TIMEOUT.
> 
> Disagree.
> 
> If the binding cache is just an optimization, then I should be able to
> remove it and have the protocol work correctly.
> 
> But consider two request messages from a client requesting additional
> addresses. If they are unrelated (i.e., the second is not a
> retransmission but a different request), the server would respond with
> different answers to the two requests. If the second message was a
> retransmission, the server would need to respond to the retransmission
> with exactly the same information as it responded to the first
> message.

I don't want to get deeply involved with this discussion, but
I'm worried that the above might be misinterpreted.

It is the transaction ID that allows retransmission to be
idempotent.  The specification requires, I think, that a server
transmit the same DHCP Reply after a retransmitted DHCP Request
from a client.

The binding cache makes that easier to do.  Other possibilities
exist; we couldn't think of any easier way to do it than using
a binding cache.  Strictly speaking, it should be an implementation
choice; at this stage I am not in any way suggesting that we
make changes to the specification for the purpose of reaching
that level of purity.

But it's the transaction ID that is the determinant here.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Fri Nov 19 12:52:55 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA24974
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 12:52:54 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA00598;
	Fri, 19 Nov 1999 12:46:23 -0500 (EST)
Date: Fri, 19 Nov 1999 12:46:23 -0500 (EST)
Message-Id: <199911191708.MAA0000032433@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Thomas,

>> With the exception of the (now likely to be changed) overspecified
>> behavior for DHCP Release, I don't remember any place where the existing
>> protocol is inadequate for dealing with multiple DHCP servers.  Thus,
>> I disagree with your statement.  I hope that you will help identify any
>> places within the protocol specification that might make such misleading
>> suggestions, so that we can repair the wording as soon as possible.

>Note: I am assuming that for the purposes of this dicussion, multiple
>DHCP servers means one server has one prefix, another server has a
>different prefix. I.e., the two servers do not have the same
>information about what prefixes (or resources/parameters in general)
>to give out and a client must contact both of them to get all of the
>addresses/parameters it needs. That is what I understood Tim to be
>getting at.

You are viewing this different from us I think and its good this is
exposed to the WG cause we need to hear from those WG members speaking 
now (thank you please stick with us as we discuss this).  Nothing in
DHCPv6 prohibits the use of multiple servers on different links and just
one client process running or multiple client processes running.  It
really is an implementation choice and the vendors will decide in the
open competitive market how the implementation will proceed.  

As far as the protocol I think you should tell us exactly where you
think DHCPv6 breaks the use of multiple servers?  I don't see where it
does off the top of my head.

Nor do I think this spec needs to define that with any text so I
strongly disagree.  But I assume right now you are participating as a
working group member not as an Area Director?  Right.

>Where in the current spec does it call out for a client to ask for
>addresses/parameters from more than one dhcp server?  My read of the
>spec is that it says you pick one server and get all your resources
>from it.

Your read of the spec is an interpretation of it what it can do  but not
what it can't do.  There is nothing in the protocol that prohibits
multiple servers. How the client does this is the implementors business.
If folks want an appendix to describe an implementation scenario where:

  1. a new interface comes up on the same link (same sever though)
     this is easy.
  2. a new interface comes up on a link that did not exist and 
     may be a new server after the solicit goes out.

As an implementor if I get a new link but dhcpv6 client is already
running for another link AND I want to use the same dhcpv6 client
process I may just try a REQUEST first with the server address I have on
the other link and if it don't work go back to solicit.

Or I may decide the best way to do this for my customers as a client
implementor is fork a new dhcpv6 client when the M bit gets set on the
new link from addrconf.

Either way the protocol will work.

Whether an implementation starts a new dhcpv6 client process or figures
out the software to use the same process is totally not our business in
the IETF.  

If the protocol will not support either case or one of the cases then I
think that is a bug in the spec.  I don't see one and would like to see
the bug revealed if there is one in the protocol.

As a note this is where Charlie and I consistenly disagree with Thomas
on issues.  But we really need to expose all these issues and hear from
the group.

>In a theroretical sense, the protocol doesn't disallow getting
>resources from multiple servers in the above case. But because there
>is no text calling for clients to actually do this (and the current
>text certainly implies that one get all resources from one server), I

I disagree with that.  We talk about a server.  

>believe it is fair to say that DHCP as currently written does not
>support getting resources from multiple servers in the case Tim
>mentions.

I think your premises to you conclusion are invalid hence your
conclusion is invalid.

>> >                       There is no notion in the protocol that one
>> > should be conversing with multiple servers simultaneously.  Do you
>> > think this is a requirement?
>
>> Jim and I have, all along, tried to make sure that the DHCP protocol
>> does exactly provide for the possibility that a client will have
>> (perhaps simultaneous) transactions with multiple DHCP servers.

>Unless the draft clearly spells out when and under what conditions
>simultaneous transactions with multiple servers are to take place,
>clients won't implement the feature. 

I disagree with you premise and conclusions again.  If an implementor is
that unaware to not catch this they and the company deserve the
competitive lashing they will get in the market for missing this as
implementors.  They will fix it in their next release.

Do you think implementors are not capable of dealing with a protocol
that is idempotent to this issue?  I think they are and trust them more
than the IETF to put this stuff in specs.

>> I think it should be a requirement that the protocol support access
>> to multiple DHCP servers simultaneously.  The client gets to decide
>> which servers to query, and the servers get to implement enough policy
>> information to help them serve the clients according to rational
>> administration.

>Please point out specific text in the current draft that makes it
>clear under what conditions clients are to expected to be interacting
>with multiple servers simultaneously. I don't recall there being much
>in the way of specific requirements here.

We don't and did not have to do that the protocol is written so that
client can communicate with a server or server's its that good that
implementors can expound upon the use of that as an implementation
choice.

I can't find any bugs in the spec that cause such a problem and it
appears you can't either.  If you can we will fix them.

It could be when I implement this completely that we will find bugs and
I think we all will.  What I don't think is possible is try to find
every bug in this spec before it goes to PS and that is impossible IMO.
To find the bugs left (other than Mikes issues he is raising) we need to
go implement.

This is how IPv6 base spec, stateless addrconf, neighbor discovery et al
all have worked with IPv6.  I know cause I implemennted a lot of it and
we changed all those specs a few times after they were proposed
standard.

If we are to attempt to find every bug in these specs without going to
PS we will not be successful be here for another 4 years discussing
cause none of us are implememnting this spec until it goes to PS.

>> >                                     But we need to distinguish
>> > between what is a general requirement for all nodes vs. those that
>> > will be mobile.
>
>> I am curious about what kind of distinction you would like to see
>> put into place.  I'd rather that distinctions not be made, except
>> in cases which are purely related to requirements from Mobile
>> IP(v6).
>
>I suspect (but am not entirely sure) that nodes implementing mobile IP
>may have special requirements with regard to DHCP (e.g., a need to
>interact with servers at both the local and home site
>simultaneously). In such cases, it would probably make sense to have
>mobile nodes (as part of the mobile IP spec) specify those
>requirements as part of what implementing mobile IP means. It is not
>clear to me that all clients would need to implement this. In any
>case, we need to talk specifics here in order to decide whether
>somethign is a general requirement or one appropriate for mobile
>nodes.

I also would like to say that this is not needed if it is needed to move
this spec to PS.  And I would like to hear from others on this besides
the Area Directors and us authors?

/jim


From dhcp-v6@bucknell.edu  Fri Nov 19 13:55:41 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA23243
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 13:55:40 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA22848;
	Fri, 19 Nov 1999 13:50:32 -0500 (EST)
Date: Fri, 19 Nov 1999 13:50:32 -0500 (EST)
Message-Id: <199911191748.MAA0000011217@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

charlie,

the issue with the transation id is that if it wraps hence they are not
part of the server database entry in section 6.  the reason we did this
was because of the wrapping.  but one can find the client.

once the retransmit is over via XID_TIMEOUT the transaction is looked up
in the database is how had started my prototype.  

as I just stated to thomas we probably need to nail down this multiple
address decision with the working group and relook at the cache.  

as I did not want to search the whole database on a client request and I
timed out and whether it was a retransmit or request is irrelevant ergo
the idempotent issues with addresses.

once the XID_TIMEOUT expires for a transaction a request for the client
is not a retransmit even if it really is.  But the protocol for
retransmission will still work at the client end regardless of the
server behavior.  which was our design principle all along.  I think we
have kept that design priniciple in tact and one we borrowed from
DHCPv4 directly.

at the chicago meeting we discussed this.  this way we do not have to
keep transaction ids as they are essentially volatile as part of the key
to look up clients for requests in the database.

I personally don't believe the protocol will work as well without a
retransmit cache and so far from my prototyping I appear to be right.

but it could be an implementation advantage that we can take out of the
spec and my implementation will be the best and make the most money in
the market -----)..... 

/jim



From dhcp-v6@bucknell.edu  Fri Nov 19 13:56:30 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA23580
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 13:56:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA10797;
	Fri, 19 Nov 1999 13:52:10 -0500 (EST)
Date: Fri, 19 Nov 1999 13:52:10 -0500 (EST)
Message-Id: <199911191758.MAA0000012807@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>From Charlie Perkins:

>This again relates to the discussion about whether clients should
>be instructed when to use a protocol.  Since you see here how a Mobile
>IP client might use two different servers, it makes sense to you.  Yet,
>you do not allow for the operation to make sense _in the abstract_
>(or, perhaps, "theoretical") case.  I claim that it is better to get
>the protocol right in the abstract, knowing that there will be specific
>cases.  I do not wish to get involved with listing the special cases
>whereby clients might wish to interact with multiple DHCP servers.

Working Group members this above from Charlie I strongly agree with and
is the crux of our many disagreement with Thomas as a WG member inputing
to the spec.  

You need to help us here and respond to this abstract issue as it can
reduce the time to get this spec in shape and fix the real protocol bugs
we are finding from this discussion.

Charlie and I are not going to give into Thomas on this stuff unless we
hear from the WG we should and must go that so please advise.

Right now we have a stalemate here between us almost on the philosophy
of specs and to some degree what is necessary for protocol definition.

Can others please respond to this issue Thomas, Francis, Barr, Stig,
Tim, et al...

Ralph also as chair you view of whats happening here is very important
on the constant disagreement charlie and I have with thomas on this
subtle but important set of points.

Right now no one can say the authors including Mike now are not
listening what we are saying is we don't agree with 1 individual in the
IETF about what we have to do hear to get this spec to PS.  So we are
not the bad guys here.  Now if 5 or more persons on this list tell us
Thomas's philosophy is right and we don't change or give the spec to
someone else which we would not do then we will make the changes.

So help us get out of this stalemate mode.  

Or Thomas you could see the wisdom of what both charlie and I are
relaying to you and give us a break here we are trying to just get to PS
and get folks to implement the spec.  We agree to all the issues Mike
sent can we please just resolve them and not try to add second guessing
what implementations will do until we get some started.

thanks
/jim


From dhcp-v4@bucknell.edu  Fri Nov 19 15:41:36 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA10865
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 15:41:34 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA23134;
	Fri, 19 Nov 1999 15:39:17 -0500 (EST)
Date: Fri, 19 Nov 1999 15:39:17 -0500 (EST)
Message-Id: <3835B0BA.776A6537@ehsco.com>
Errors-To: droms@bucknell.edu
Reply-To: ehall@ehsco.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "Eric A. Hall" <ehall@ehsco.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.7 [en] (WinNT; I)
Content-Transfer-Encoding: 7bit


> reconfiguring DNS is not going to change application which have
> already initialised.

Do the apps do their own DNS lookups or do they call the system
resolver? If the resolver is able to recognize the changes then the apps
that call it won't care.

> Changing IP masks is going to breka existing connections..

How so? IP only cares about the mask for delivery purposes. TCP and UDP
don't care about the mask, and neither should any of the applications.
Changing the IP address would break the connections, but changing the
mask certainly shouldn't have any more of an effect on those layers than
changing the default route (ie, it should have no effect at all).

> And how does the DHCP client know which apps are interested
> in which parameters..

What parameters should application-layer protocols be interested in? I
can see why there'd be some interest in changing "default home page" or
"default mail server" but I don't know of many applications that use
those paramters.

> basically, I think it is useless for desk top computers to bother
> with changes to parameters unless they reboot.

I'd have to disagree with you 100% on this.

-- 
Eric A. Hall                                            ehall@ehsco.com
+1-650-685-0557                                    http://www.ehsco.com


From dhcp-v6@bucknell.edu  Fri Nov 19 16:29:38 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA01304
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 16:29:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA06785;
	Fri, 19 Nov 1999 16:26:27 -0500 (EST)
Date: Fri, 19 Nov 1999 16:26:27 -0500 (EST)
Message-Id: <3835BEDF.DED8E8E1@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2,
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Jim,

> the issue with the transation id is that if it wraps hence they are not
> part of the server database entry in section 6.  the reason we did this
> was because of the wrapping.

But the XID_TIMEOUT will occur long before transaction-ID wrapping
will make any difference.  And retransmissions will occur long before
XID_TIMEOUT.  If these statements are not true, then we may have to
put more bits into the transaction-ID field.  On the other hand,
I can't think of a client making 100 transactions per second for
600 seconds.  That sounds preposterous; thus, I don't think this is
a problem.  And, I can't think about why it would take 10 minutes
for a server to manage a DHCP Request either.  Thus, the default
values seem to work O.K. to me.

> once the retransmit is over via XID_TIMEOUT the transaction is looked up
> in the database is how had started my prototype.

The way I parse this means that transactions aren't finished for
10 minutes (600 seconds), and I'm sure that's not the way you
meant it.

> as I did not want to search the whole database on a client request and I
> timed out and whether it was a retransmit or request is irrelevant ergo
> the idempotent issues with addresses.

When a new DHCP Request comes in, the server should naturally look
at the client's binding.  This should happen no matter what extensions
are present, and no matter how often new DHCP Requests arrive.

> once the XID_TIMEOUT expires for a transaction a request for the client
> is not a retransmit even if it really is.  But the protocol for
> retransmission will still work at the client end regardless of the
> server behavior.

While I agree with your statement, I think the situation would
never arise, by the timing considerations I outlined previously
in this note.

> I personally don't believe the protocol will work as well without a
> retransmit cache and so far from my prototyping I appear to be right.

While I agree with this as stated, it doesn't imply that a correct
protocol is required to have such a retransmission cache.

> but it could be an implementation advantage that we can take out of the
> spec and my implementation will be the best and make the most money in
> the market -----).....

How about if it merely makes a few hundred million $, but not the most?

Regards,
Charlie P.


From dhcp-v4@bucknell.edu  Fri Nov 19 16:33:28 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA03060
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 16:33:28 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA29209;
	Fri, 19 Nov 1999 16:30:55 -0500 (EST)
Date: Fri, 19 Nov 1999 16:30:55 -0500 (EST)
Message-Id: <EA5107D3A9ADD211B7530000C041E4F4011B2AFF@honts318.wal-mart.com>
Errors-To: droms@bucknell.edu
Reply-To: ndlane@wal-mart.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Nathan Lane <ndlane@wal-mart.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP server/Client behavior when options change while
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2650.21)
Mime-Version: 1.0



> -----Original Message-----
> From:	Eric A. Hall [SMTP:ehall@ehsco.com]
> Sent:	Friday, November 19, 1999 2:42 PM
> To:	Multiple recipients of list
> Subject:	Re: DHCP server/Client behavior when options change while
> 
> 
> > reconfiguring DNS is not going to change application which have
> > already initialised.
> 
> Do the apps do their own DNS lookups or do they call the system
> resolver? If the resolver is able to recognize the changes then the apps
> that call it won't care.
> 
	[Nathan Lane]  Most apps I know of here use gethostbyname() whenever they need to resolve a name, so they will notice changes immediately.  Even IE4 and Netscape call gethostbyname about once every 15 minutes (they cache in the interim).

> > basically, I think it is useless for desk top computers to bother
> > with changes to parameters unless they reboot.
> 
> I'd have to disagree with you 100% on this.
> 
	[Nathan Lane]  I agree with the disagreement.  There are many times when a parameter would need to change!  We are currently removing many, many WINS servers from our network and workstations need to be told of the change.  Rebooting is not an option.  I also rolled new DNS servers to some 20K clients transparently because the clients picked up the change upon renew (NT SP3 and later).

	We've also designed in a short renew interval (but a relatively longer lease time) specifically so we can roll mass parameter changes under emergency server outage conditions.

	-Nathan Lane
	Wal-Mart Stores, Inc.
	(501) 277-5786
>  


From dhcp-v6@bucknell.edu  Fri Nov 19 17:12:33 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA21042
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 17:12:30 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA20940;
	Fri, 19 Nov 1999 17:09:27 -0500 (EST)
Date: Fri, 19 Nov 1999 17:09:27 -0500 (EST)
Message-Id: <3835C9FC.EAB72CAA@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #3, Reply to RECONFIGURE? (Final)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Erik Nordmark wrote:

> Multicast protocols tend to deal with this by having the sender
> of the multicast packet include a parameter to control over how long time
> the receivers should do a random delay. That way, since the DHCP server knows
> who many clients are listening on each multicast address, it can
> suggest a reasonable random delay to avoid overload.

I can live with this, but there are two other points worth noting.

1. We already spread out responses over the time interval
   sefined by RECONF_MMSG_MIN_RESP and RECONF_MMSG_MAX_RESP.

2. It is expected that DHCP servers would assign populations of DHCP
   clients to different multicast groups, in order to spread out the
   workload for reconfiguration to the different groups.  Being able to
   limit the size of the multicast group was a prime consideration
   for creating this feature.  Being able to manage renumbering
   separately for different departments was an excellent side benefit.

Therefore, I suggest that any such random delay parameter as you
suggest should be left for later, to see if its needed in real
deployments.  This is the kind of thing that comes better with
experience, I think; it is doable as a new DHCP extension at a later
time.

> Making the RECONFIGURE* message(s) idenpotent seems like a good idea.
> But even with a RECONFIGURE_INIT/REQUEST/REPLY exchange we need random
> delays to avoid a request implosion.

As noted, we already do have some degree of randomness.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Fri Nov 19 17:51:05 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA08928
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 17:51:05 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA28556;
	Fri, 19 Nov 1999 17:08:09 -0500 (EST)
Date: Fri, 19 Nov 1999 17:08:09 -0500 (EST)
Message-Id: <3835C7C1.2302F39F@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Erik Nordmark wrote:

> I've heard of interoperability problems in DHCPv4 since it isn't clear
> what options the client need to explicitly ask for and what the server will
> automatically provide. Thus a client might not need to ask for an option
> for some server implementations but other server implementations might insist
> on being asked to provide the option.
> 
> Isn't this something we should try to avoid in DHCPv6?

This is a tricky problem!  On the one hand, we might specify a
list of extensions that servers MUST provide when first they create
a client binding.  Alternatively, we might specify that clients
have to include specific requests if they expect specific extensions
to come back in a DHCP Reply message.  The problem with the former
approach is that it is possible we don't know what extensions are
always needed.  Surely the NTP server isn't one of them.  I'd
guess that even DNS isn't always needed.  The problem with the latter
approach is that it makes the DHCP Request message more bulky,
especially
unwlcome when dealing with slow media.

What if we don't pick either method, but we also require that
clients MUST NOT depend upon server behavior in this matter?
Since IPv6 nodes using stateless address autoconfiguration are
still allowed to use DHCPv6 (say, to find DNS), I don't think we
can legislate server behavior, and therefore we have to leave it
to be a matter of local policy.  This tends towards making clients
be specific with their needs, unless local administration is careful
in establishing configuration information for its DHCP clients.
But, wireless speeds are going up, and the extra bulk is only present
in relatively rare control messages, so I guess it's better to lean
towards solutions that put more burden on the clients, unfortunately.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Fri Nov 19 17:56:30 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA10896
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 17:56:29 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA17970;
	Fri, 19 Nov 1999 17:55:05 -0500 (EST)
Date: Fri, 19 Nov 1999 17:55:05 -0500 (EST)
Message-Id: <199911192210.OAA00647@feller.mentat.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Tim Hartrick <tim@mentat.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.





Erik,

> 
> Dealing with multi-homed machines, or to be more specific, multi-interfaced

I was talking about multi-homing in the provider sense not in the interface
sense.  Clearly, DHCPv6 is a protocol that from the client's perspective
operates on individual interfaces.

I was more concerned about cases where I have multiple providers and as a
result possibly multiple DHCPv6 servers providing me with resources.  The
most likely case that I can imagine for this to occur is for mobile IPv6
but I can also imagine cases where I am using a cable modem or DSL system
to provide me with packet transit but I am actually have multiple providers
which use DHCPv6 to provide me with resources.



Tim Hartrick
Mentat Inc.


From dhcp-v6@bucknell.edu  Fri Nov 19 17:58:27 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA11638
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 17:58:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA13361;
	Fri, 19 Nov 1999 17:56:24 -0500 (EST)
Date: Fri, 19 Nov 1999 17:56:24 -0500 (EST)
Message-Id: <3835D10C.B4E655FB@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2,
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Jim,

I would like to suggest that we do not need to add new
behavior for the DHCP server to be able to disambiguate
requests for new addresses from requests to renew lifetimes
for existing addresses.

The main point, as discussed earlier today, is that the DHCP
server has to keep track of what has been assigned to its
clients.  Thus, if a client asks to renew the lifetime of its
assigned address, the server will know that the client already
has the address assigned, so it can do the right thing.

Conversely, if a client sends a request in with an address
extension including an address that has not been assigned to it,
the server can (if desired) make that new address assignment.

If a client sends a request that does NOT have an address
extension, the server has to figure out whether or not to
put an address extension on the DHCP Reply.  I suggest that
this needs to be done by having administrative information
present at the DHCP server that will determine the relevant
address assignment policy for that subnet.  If the subnet
uses stateless address autoconfiguration, the server doesn't
have to send an address.  If the subnet doesn't use stateless,
and the server has no records for the client, then the server
probably should send an address out to the client.

If the client has gotten some addresses from one server, and
then sends a request to another server, then the second server
will not necessarily know about all of the addresses that the client
has received from the first server.  In that case, the client might
still wish to receive another IP address from the second server.
That would typically be the desired behavior when the second
server was administratively configured to provide addresses for
the subnet from which the client made the request.  If the client
has an IP address already assigned, then it would use that IP
address as the source IP address for its DHCP Request packet,
so alternate servers can always know when the client already
has an IP address assigned.

If you have more cases, then we can discuss them to make sure
that our current mechanism handles each case well.  And, if it
turns out that there is a case which isn't yet handled, we can
evaluate whether creating new bits in the DHCP Request message
will solve the problem.  I don't want to create the bit before
I can understand what problem it will be solving.


> If the client requests 1 address now continuing the discussion I would
> return the client one of the addresses I have already given it as a
> server implementor.  But this is not clear in the spec and if I don't do
> it (implement) this way the REQUEST would not be idempotent if I give
> the client a new address.

Idempotency has to be judged for messages that all have the same
transaction ID.  I think it would be a mistake to constrain messages
with different transaction IDs to necessarily have the same results.
If I ask for a bowl of ice cream, I don't want somebody telling me that
I already got one bowl.

> That is why we need to do one of the following things as I see it in the
> protocol:
> 
>   1.  Add a bit in the request that is a "pure" request for a brand
>       new address or addresses and addr extentions would follow.
> 
>   2.  The client must ask  for all its addresses back and an additional
>       extension if it wants a new one.  This would be done by the
>       client providing multiple addr extentions in the request.
> 
>   3.  Or we don't permit this behavior and all the client does
>       is have a bit set in the request header "give me my addresses"
>       an no addr extensions are provided by the client request msg.
> 
> I would like to see us adopt #2 above and is the compromise.

As I suggested earlier in this message, I don't yet see the need
for any new bits, nor a change to the specified behavior in the
protocol.

> >Thus, I assert that removal of the binding cache in DHCPv6 would
> >result in the server returning differnt answers than it would
> >otherwise. Hence, it is not just a performance optimization, it is
> >needed for correctness.
> 
> I don't agree see above.  Also the once the binding cache times out the
> problem happens anyway.  We don't keep the cache forever.

I think that the behavior with respect to the client is contolled
by the client binding, not by anything that depends strictly on
the transaction ID.  The transaction ID can be part of the client
binding, but if a server allocates an IP address to the client for
50 hours, it has to keep the client binding around for 50 hours.
It might be useful to specify a maximum holding time for
client bindings, but it has to be a lot larger than XID_TIMEOUT.
There isn't any such maximum time specified now in DHCPv6, and
it might just as well be implementation-dependent.



> If the cache times out and the client retransmits then the server must
> respond by whats in the database.  This is probably not likely but could
> happen.  Ergo why I say above there is bug here we need to fix and I
> provided my view of that fix.

Is there anything with using what's in the database?  Is there
ever a circumstance where the database contents could disagree
with the contents of the cache?  I hope not...


Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Fri Nov 19 23:40:22 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA26305
	for <dhc-archive@odin.ietf.org>; Fri, 19 Nov 1999 23:40:22 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA28483;
	Fri, 19 Nov 1999 23:38:12 -0500 (EST)
Date: Fri, 19 Nov 1999 23:38:12 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.943071602.2166.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DHCPv6 Input from Francis Dupont 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

> I think first it should be discussed what is wrong with the spec first
> before we say it needs changing.  I don't think Francis was saying there
> is an issue with the specs behavior but you are I think?
> 
> Can you articulate what is wrong with the spec as is presently on this
> topic?

*If* we agree that a client should be identified as the same client
by a server independent of the relay agent it is using on the link
then I think the spec needs a slight tweak.
It currently says that the client is identified by <link-local address,
agent-address> and I think this should be <link-local address, agent-prefix>.
(where agent-prefix could also be called link-prefix since it is the prefix of
the client's link that matters).

In addition to clarifying the scope (site-local vs. global) for
the agent-prefix as you point out we also need to make sure that
two agents on a link pick the same prefix as agent-prefix,
since there can be multiple global prefixes assigned to the link.
I think some simple rules (pick the numerically largest prefix)
would suffice.

  Erik
 




From dhcp-v4@bucknell.edu  Sat Nov 20 09:04:35 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA19221
	for <dhc-archive@odin.ietf.org>; Sat, 20 Nov 1999 09:04:35 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id IAA27479;
	Sat, 20 Nov 1999 08:59:05 -0500 (EST)
Date: Sat, 20 Nov 1999 08:59:05 -0500 (EST)
Message-Id: <19991120135007.20284.rocketmail@web1903.mail.yahoo.com>
Errors-To: droms@bucknell.edu
Reply-To: anassar2000@yahoo.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ayman Nassar <anassar2000@yahoo.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Relay agent - IP information
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0


Does draft-ietf-dhc-agent-options-07.txt "DHCP Relay
agent Information Option" specify how a relay agent
with an interface like ATM, with no IP address, can be
set to dynamically get assigned an IP address
to one of its subinterfaces that are connected to a
LAN (DHCP clients) via a PVC.

Under section 3.1 "Agent circuit ID suboption",
(option 82) the client could receive IP configuration
information from the DHCP server via the relay agent
using an identifier, which in this case could be
either the router sub-interface number or the ATM VC
number. 

Ayman


__________________________________________________
Do You Yahoo!?
Bid and sell for free at http://auctions.yahoo.com


From dhcp-v4@bucknell.edu  Sat Nov 20 10:42:50 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA06447
	for <dhc-archive@odin.ietf.org>; Sat, 20 Nov 1999 10:42:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA28640;
	Sat, 20 Nov 1999 10:42:34 -0500 (EST)
Date: Sat, 20 Nov 1999 10:42:34 -0500 (EST)
Message-Id: <Pine.GSO.3.95.991120101327.24377A-100000@codex.cis.upenn.edu>
Errors-To: droms@bucknell.edu
Reply-To: waa@dsl.cis.upenn.edu
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: "William A. Arbaugh" <waa@dsl.cis.upenn.edu>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Send me your authentication draft comments/edits please
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; charset=US-ASCII
Mime-Version: 1.0


Now that everyone has returned and plowed through their inboxes- you
should have time to send me your edits/comments right? :-)

Several of you mentioned that you had edits- I won't call you out by name
(but you know who you are). Please send me your edits/comments as soon as
possible and I'll generate a new draft so we can review it prior to final
call.

Thanks, Bill

-----------------------------------------------------------------------
Bill Arbaugh			   
email:  waa@dsl.cis.upenn.edu 
web:    http://www.cis.upenn.edu/~waa
-----------------------------------------------------------------------  



From dhcp-v4@bucknell.edu  Sat Nov 20 19:36:43 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA09933
	for <dhc-archive@odin.ietf.org>; Sat, 20 Nov 1999 19:36:42 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA30315;
	Sat, 20 Nov 1999 19:35:45 -0500 (EST)
Date: Sat, 20 Nov 1999 19:35:45 -0500 (EST)
Message-Id: <4.2.0.58.19991120160653.009f1690@mail.join.com>
Errors-To: droms@bucknell.edu
Reply-To: robs@join.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Rob Stevens <robs@join.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: DHCP server/Client behavior when options change while
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"; format=flowed
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 

At 04:31 PM 11/19/99 -0500, Nathan Lane wrote:



> > Do the apps do their own DNS lookups or do they call the system
> > resolver? If the resolver is able to recognize the changes then the apps
> > that call it won't care.
> >
>         [Nathan Lane]  Most apps I know of here use gethostbyname() 
> whenever they need to resolve a name, so they will notice changes 
> immediately.  Even IE4 and Netscape call gethostbyname about once every 
> 15 minutes (they cache in the interim).

Calling gethostbyname() is not going to change
the DNS server to which the packet is sent (on most platforms): that
was established implicitly by the first call or
an explicitly via res_init().


> > > basically, I think it is useless for desk top computers to bother
> > > with changes to parameters unless they reboot.
> >
> > I'd have to disagree with you 100% on this.
> >
>         [Nathan Lane]  I agree with the disagreement.  There are many 
> times when a parameter would need to change!  We are currently removing 
> many, many WINS servers from our network and workstations need to be told 
> of the change.  Rebooting is not an option.  I also rolled new DNS 
> servers to some 20K clients transparently because the clients picked up 
> the change upon renew (NT SP3 and later).
>
>         We've also designed in a short renew interval (but a relatively 
> longer lease time) specifically so we can roll mass parameter changes 
> under emergency server outage conditions.
>

And I disagree with the agreement with the...whoops stack overflow.

I don't doubt that there are may be times when a parameter would need
to change. Nor that Walmart is in the happy position of having all the
clients and their apps in their stores respond (nee, instantaneously)
to the changes which just came down the pike from DHCP.


But other platforms and applications will not.
There is a whole universe of applications out there, and
the vast bulk don't know about the wonderful world
that DHCP made possible, in which any parameter
known to DHCP and used by these apps can change
asynchronously.

Therefore, I think to dismiss all these apps as being,
somehow, broken, is extreme. And  what is more
to the point, that it is misleading to convey the
impression that ISPs, sysadmins, whomever,
can get their all the apps throughout their domain
to respond to option changes in a time predictable
manner by DHCP alone. Some applications are just
going to go on using whatever they started up with,
and nothing DHCP says is going to change that.

I am quite happy for servers to send these parameters
whenever they want; even willing to accept that they
be sent when not requested (hasn't anyone run out of space
yet?). That is not my issue (at least, not now).

Rob





From dhcp-v6@bucknell.edu  Sat Nov 20 22:05:14 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA23260
	for <dhc-archive@odin.ietf.org>; Sat, 20 Nov 1999 22:05:12 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA24317;
	Sat, 20 Nov 1999 22:03:04 -0500 (EST)
Date: Sat, 20 Nov 1999 22:03:04 -0500 (EST)
Message-Id: <199911210258.VAA0000012496@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


Hi Charlie,

>> the issue with the transation id is that if it wraps hence they are not
>> part of the server database entry in section 6.  the reason we did this
>> was because of the wrapping.

>But the XID_TIMEOUT will occur long before transaction-ID wrapping
>will make any difference.  And retransmissions will occur long before
>XID_TIMEOUT.  If these statements are not true, then we may have to
>put more bits into the transaction-ID field.  On the other hand,
>I can't think of a client making 100 transactions per second for
>600 seconds.  That sounds preposterous; thus, I don't think this is
>a problem.  And, I can't think about why it would take 10 minutes
>for a server to manage a DHCP Request either.  Thus, the default
>values seem to work O.K. to me.

I seem to have lost track of the issue what your saying I agree with.
My statements above are on a completely different track and response to
another matter so I am confused ???? sorry...

My response above was a reason that was given to the WG why we took the
transaction-id out of the database key binding which is mandatory.
The database I was talking about and section 6. is the server database
for the clients.  For the BINDING cache I agree. 
But it is not listed as a key for the binding as required.
I for one would use it for the binding cache.

Section 6:

 A server maintains a collection of client records, called
   ``bindings''.  Each binding is uniquely identifiable by the ordered
   pair <link-local address, agent-address>, since the link-local
   address is guaranteed to be unique [20] on the link identified by the
  agent-address and prefix.  An implementation MUST support bindings
   consisting of at least a client's link-local address, agent-address,
  preferred lifetime and valid lifetime [20] for each client address.

AND THEN LATER ON: 

   In addition to the client binding a server must maintain an
   XID_TIMEOUT binding cache to determine if a previous transaction-ID
   is being retransmitted by a client.  An implementation of an
   XID_TIMEOUT binding cache MUST support at least a tuple consisting of
  the client's link-local address, agent-address prefix, IPv6 address,
   and XID_TIMEOUT value when the cache entry can be deleted (see
   Section 8).

I would not use it for the servers collection records though.  

>> once the retransmit is over via XID_TIMEOUT the transaction is looked up
>> in the database is how had started my prototype.

>The way I parse this means that transactions aren't finished for
>10 minutes (600 seconds), and I'm sure that's not the way you
>meant it.

I am saying if the client does not retransmit until 20 minutes later 
the Request will be looked up in the server client database if
XID_TIMEOUT is at default.  

The point for me wearing my implementor hat is that if a customers
client does this and I am the server I have to write code for that case
in the protocol spec.

This is where the entire issue of idempotency appears from Thomas.  His
question is valid, but we are fine IMO as best I can tell.  But the
multiple address decisions on that thread affects what I will do for and
to keep idempotency in my software as a server.  If the spec don't do it
I have to do it as a product engineer.  So IMO we have to resolve this
issue one way or the other.  I can't guess at it which is what I am
doing now by driving the policy at the set up or something.  

>> as I did not want to search the whole database on a client request and I
>> timed out and whether it was a retransmit or request is irrelevant ergo
>> the idempotent issues with addresses.

>When a new DHCP Request comes in, the server should naturally look
>at the client's binding.  This should happen no matter what extensions
>are present, and no matter how often new DHCP Requests arrive.

My statement implied a "retransmit client request" from previous client
requests, sorry for the terseness, I was quite busy when I wrote the
above.  For a retransmit I would not have to look at the cleint database
but only for the REPLY I must resend.  This is why we have the binding
cache in the spec so I don't have to look up the server database for
retransmits and keep the cache entries small.

>> once the XID_TIMEOUT expires for a transaction a request for the client
>> is not a retransmit even if it really is.  But the protocol for
>>retransmission will still work at the client end regardless of the
>> server behavior.

>While I agree with your statement, I think the situation would
>never arise, by the timing considerations I outlined previously
>in this note.

Well I disagree and I think I clearly articulate why and how above. I am
interested where you see a flaw in my above response.  Unless you don't
believe a client will respond later than 600 seconds?  Which I would
disagree with if told it can't happen.

> I personally don't believe the protocol will work as well without a
> retransmit cache and so far from my prototyping I appear to be right.

>While I agree with this as stated, it doesn't imply that a correct
>protocol is required to have such a retransmission cache.

I agree. We could actually put this all in an appendix is your
implication of such a response?

>> but it could be an implementation advantage that we can take out of the
>> spec and my implementation will be the best and make the most money in
>> the market -----).....
>
>How about if it merely makes a few hundred million $, but not the most?

That would be fine too.

thanks
/jim


From dhcp-v6@bucknell.edu  Sat Nov 20 23:11:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA26450
	for <dhc-archive@odin.ietf.org>; Sat, 20 Nov 1999 23:11:57 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA19167;
	Sat, 20 Nov 1999 23:09:38 -0500 (EST)
Date: Sat, 20 Nov 1999 23:09:38 -0500 (EST)
Message-Id: <199911210343.WAA0000019344@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Charlie,

>I would like to suggest that we do not need to add new
>behavior for the DHCP server to be able to disambiguate
>requests for new addresses from requests to renew lifetimes
>for existing addresses.

I would suggest we need to add text to the spec though it could be
much more clear and its not now as we go thru this excercise.

>The main point, as discussed earlier today, is that the DHCP
>server has to keep track of what has been assigned to its
>clients.  Thus, if a client asks to renew the lifetime of its
>assigned address, the server will know that the client already
>has the address assigned, so it can do the right thing.

I don't agree that was the main point of the discussion.  The main point
of the discussion is twofold:

  1.  did we specify clearly what the server must do with the
      combinations of getting addresses.  I think we uncovered a 
      case per the WG mail list discusion which is good.  If I am
      wrong that is great less spec changes.  But I think there 
      is a bug.

  2.  Does the WG support changing the model of DHCPv4 addresses to
      clinent.  I hope so when this discussion ends.

But to get to the other part of your sentence above.  I agree but would
add this means the client also provides the extension in the msg to
renew the lifetimes.

But we should add this set of statements to the spec.

>>Conversely, if a client sends a request in with an address
>>extension including an address that has not been assigned to it,
>>the server can (if desired) make that new address assignment.

This is the case where a client now has sent in an address suggesting 
to the client I want to use this address, and the client is telling the
server to add this address to their database.  

This is exactly why we need to specify in the specs what the
combinations of extensisons are and their syntax and semantics so its
clear.  It not now.

I would change the word desire to MAY though as I don't find it to
desirable in the norm to tell servers to get addresses from clients.
But I would not disallow folks from building servers to do this in the
future.

So now we have two cases of interpreting extensions.  

>If a client sends a request that does NOT have an address
>extension, the server has to figure out whether or not to
>put an address extension on the DHCP Reply.  I suggest that
>this needs to be done by having administrative information
>present at the DHCP server that will determine the relevant
>address assignment policy for that subnet.  If the subnet
>uses stateless address autoconfiguration, the server doesn't
>have to send an address.  If the subnet doesn't use stateless,
>and the server has no records for the client, then the server
>probably should send an address out to the client.

Your statement above on stateless is not necessarily true.  The
stateless addrconf can be running and when the router sets the M bit it
says go get your addresses or additional addresses from dhcpv6.
So I strongly don't agree with your suggestion.  DHCPv6 and Stateless
will operate together on the same link potentially as Addrconf is realy
just a subset of Neighbor Discovery, so if the M bit is on the dhcpv6
server will be contacted if not then dhcpv6 won't be contacted.

But what if the server has already given the client addresses?

In this case the server should not give the client an address if there
is no address extension.

If the client has no addresses then I agree with you a good citizen
server would do this and no policy is needed as you suggest we just
define the defaults maybe?

But there is another case.

What if the server gets an empty address extension which is permitted
today.

 1.  If the client has no addresses then I think an address 
     can be returneded for each client address extension if the 
     client is permitted to have X addresses.  This is where 
     policy is needed and has nothing to do with stateless addrconf.

 2.  If the client has existing addresses do you give them another 
     one or not.  I would say yes up to the number of addresses
     allowed per client by policy.

But then we are back to that a client must provide an extension for an
address request unless its the clients first request to get an address
per server.  

>If the client has gotten some addresses from one server, and
>then sends a request to another server, then the second server
>will not necessarily know about all of the addresses that the client
>has received from the first server.  In that case, the client might
>still wish to receive another IP address from the second server.
>That would typically be the desired behavior when the second
>server was administratively configured to provide addresses for
>the subnet from which the client made the request.  If the client
>has an IP address already assigned, then it would use that IP
>address as the source IP address for its DHCP Request packet,
>so alternate servers can always know when the client already
>has an IP address assigned.

I think this is of benefit for when multiple servers are used.  

>If you have more cases, then we can discuss them to make sure
>that our current mechanism handles each case well.  And, if it
>turns out that there is a case which isn't yet handled, we can
>evaluate whether creating new bits in the DHCP Request message
>will solve the problem.  I don't want to create the bit before
>I can understand what problem it will be solving.

I suggested two ways to solve the problem one with bits and one with
extensions.  I would rather not add bits to the protocol here either but
I am not going to not suggest solutions at this point in the WG because
we have to changed the spec in this particular space.  We need to hear
any ideas at this point IMO.

/jim


> If the client requests 1 address now continuing the discussion I would
> return the client one of the addresses I have already given it as a
> server implementor.  But this is not clear in the spec and if I don't do
> it (implement) this way the REQUEST would not be idempotent if I give
> the client a new address.

Idempotency has to be judged for messages that all have the same
transaction ID.  I think it would be a mistake to constrain messages
with different transaction IDs to necessarily have the same results.
If I ask for a bowl of ice cream, I don't want somebody telling me that
I already got one bowl.

> That is why we need to do one of the following things as I see it in the
> protocol:
> 
>   1.  Add a bit in the request that is a "pure" request for a brand
>       new address or addresses and addr extentions would follow.
> 
>   2.  The client must ask  for all its addresses back and an additional
>       extension if it wants a new one.  This would be done by the
>       client providing multiple addr extentions in the request.
> 
>   3.  Or we don't permit this behavior and all the client does
>       is have a bit set in the request header "give me my addresses"
>       an no addr extensions are provided by the client request msg.
> 
> I would like to see us adopt #2 above and is the compromise.

As I suggested earlier in this message, I don't yet see the need
for any new bits, nor a change to the specified behavior in the
protocol.

> >Thus, I assert that removal of the binding cache in DHCPv6 would
> >result in the server returning differnt answers than it would
> >otherwise. Hence, it is not just a performance optimization, it is
> >needed for correctness.
> 
> I don't agree see above.  Also the once the binding cache times out the
> problem happens anyway.  We don't keep the cache forever.

I think that the behavior with respect to the client is contolled
by the client binding, not by anything that depends strictly on
the transaction ID.  The transaction ID can be part of the client
binding, but if a server allocates an IP address to the client for
50 hours, it has to keep the client binding around for 50 hours.
It might be useful to specify a maximum holding time for
client bindings, but it has to be a lot larger than XID_TIMEOUT.
There isn't any such maximum time specified now in DHCPv6, and
it might just as well be implementation-dependent.



> If the cache times out and the client retransmits then the server must
> respond by whats in the database.  This is probably not likely but could
> happen.  Ergo why I say above there is bug here we need to fix and I
> provided my view of that fix.

Is there anything with using what's in the database?  Is there
ever a circumstance where the database contents could disagree
with the contents of the cache?  I hope not...


Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Sat Nov 20 23:12:08 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA26545
	for <dhc-archive@odin.ietf.org>; Sat, 20 Nov 1999 23:12:08 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA29494;
	Sat, 20 Nov 1999 23:10:48 -0500 (EST)
Date: Sat, 20 Nov 1999 23:10:48 -0500 (EST)
Message-Id: <199911210409.XAA0000022507@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DHCPv6 Input from Francis Dupont 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Hi Erik,

>> I think first it should be discussed what is wrong with the spec first
>> before we say it needs changing.  I don't think Francis was saying there
>> is an issue with the specs behavior but you are I think?
>> 
>> Can you articulate what is wrong with the spec as is presently on this
>> topic?

>*If* we agree that a client should be identified as the same client
>by a server independent of the relay agent it is using on the link
>then I think the spec needs a slight tweak.

I think we all agreed long ago that a client needs to be identified as
you very nicely stated, if anyone does not agree I would be surprised.

>It currently says that the client is identified by <link-local address,
>agent-address> and I think this should be <link-local address, agent-prefix>.
>(where agent-prefix could also be called link-prefix since it is the prefix of
>the client's link that matters).

It should say agent-prefix that was a typo in draft-14 for some reason,
the authors agree with you, and there are other places where we speak of
prefix for sure, but this is an edit error in the spec presently.

>In addition to clarifying the scope (site-local vs. global) for
>the agent-prefix as you point out we also need to make sure that
>two agents on a link pick the same prefix as agent-prefix,
>since there can be multiple global prefixes assigned to the link.
>I think some simple rules (pick the numerically largest prefix)
>would suffice.

I agree.  We need to add that and the scope issue.

thanks
/jim



From dhcp-v6@bucknell.edu  Sun Nov 21 00:14:50 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id AAA26403
	for <dhc-archive@odin.ietf.org>; Sun, 21 Nov 1999 00:14:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id AAA10322;
	Sun, 21 Nov 1999 00:12:47 -0500 (EST)
Date: Sun, 21 Nov 1999 00:12:47 -0500 (EST)
Message-Id: <199911210505.AAA0000014759@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Charlie,

Now I will respond to the rest of your mail.  I wanted to split it up.

>> If the client requests 1 address now continuing the discussion I would
>> return the client one of the addresses I have already given it as a
>> server implementor.  But this is not clear in the spec and if I don't do
>> it (implement) this way the REQUEST would not be idempotent if I give
>> the client a new address.

>Idempotency has to be judged for messages that all have the same
>transaction ID.  I think it would be a mistake to constrain messages
>with different transaction IDs to necessarily have the same results.
>If I ask for a bowl of ice cream, I don't want somebody telling me that
>I already got one bowl.

Then you are saying you want transaction-ids to be mandatory in the
binding for the server client database?

I think in that case it changes the landscape.  But it don't address the
problem documented above.  This must be specified one way or the other.

If we are to add transaction-ids I would claim they are necessary if we
move the binding cache to the appendix.  Otherwise they are not.  Just
treat any request as a request and then all requests are idempotent.
Which I think is a worthwhile and attainable for dhcpv6 with few words
in the specs.  But first we need to address and define the address
logistics which I think we are doing now  and in my previous response to
this mail on the first part.

>> That is why we need to do one of the following things as I see it in the
>> protocol:
>> 
>>   1.  Add a bit in the request that is a "pure" request for a brand
>>       new address or addresses and addr extentions would follow.
>> 
>>   2.  The client must ask  for all its addresses back and an additional
>>       extension if it wants a new one.  This would be done by the
>>       client providing multiple addr extentions in the request.
>> 
>>   3.  Or we don't permit this behavior and all the client does
>>       is have a bit set in the request header "give me my addresses"
>>       an no addr extensions are provided by the client request msg.
>> 
>> I would like to see us adopt #2 above and is the compromise.

>As I suggested earlier in this message, I don't yet see the need
>for any new bits, nor a change to the specified behavior in the
>protocol.

Well I hope you getting there now after my responses.  It don't have to
be new bits and I said do #2 whihc has no new bits.

A client has to use extensions to request addresses.  If it has 3 and
wants another one it has to provide 4 extensions.  

But if you don't like this idea we still have the problem  as stated in
the previous mail.

>> >Thus, I assert that removal of the binding cache in DHCPv6 would
>> >result in the server returning differnt answers than it would
>> >otherwise. Hence, it is not just a performance optimization, it is
>> >needed for correctness.
>> 
>> I don't agree see above.  Also the once the binding cache times out the
>> problem happens anyway.  We don't keep the cache forever.
>
>I think that the behavior with respect to the client is contolled
>by the client binding, not by anything that depends strictly on
>the transaction ID.  The transaction ID can be part of the client
>binding, but if a server allocates an IP address to the client for
>50 hours, it has to keep the client binding around for 50 hours.
>It might be useful to specify a maximum holding time for
>client bindings, but it has to be a lot larger than XID_TIMEOUT.
>There isn't any such maximum time specified now in DHCPv6, and
>it might just as well be implementation-dependent.

I agree the holding time should be implementation-dependent.

>> If the cache times out and the client retransmits then the server must
>> respond by whats in the database.  This is probably not likely but could
>> happen.  Ergo why I say above there is bug here we need to fix and I
>> provided my view of that fix.

>Is there anything with using what's in the database?  Is there
>ever a circumstance where the database contents could disagree
>with the contents of the cache?  I hope not...

Yes.  Of course.  The cache is for the binding for retransmits the
server database is completely different.  If the binding cache times out
the entry is gone.  The binding cache is only there to look for
retransmits so the entire database is not searched.

binding cache entry:

  link-local-address+agent-prefix /*search key*/
  request-transid
  *reply-data-buffer-sent 

server  database:

   link-local-address+agent-prefix /*search key*/
   *client-records
   link-local-address+agent-prefix /*search key*/
   *client-records
   etc. etc...

The binding cached and server database (which could be a cache) are
different.

WHEN_CLIENT_REQUEST_RECEIVED
  If (search_retransmit_binding_cache(*CLIENT_REQUEST))
     return; /* all work for retransmit done in function above */
  else 
     locate_client_in_database(*CLIENT_REQUEST)
     ....process the request and build reply.....
     ....add entry to binding cache for this request....

If the client has been taken out of the binding_cache above then the
request is treated as a non-retransmitted request.  Of if we put the
binding cache in the appendix.  We don't really then differentiate btw
retransmits vs normal requests at the server.

But regardless we need to say what happens when a client requests an
address when the server has already provided them an address and under
what circmustances is that permitted and what are the semantics in the
protocol.  This is a bug in the spec IMO.

The entire binding cache discussion I clain is not relevant nor is
required for correct behavior and does nothing to help with correct
behavior if there is a bug after XID_TIMEOUT and a client retransmits 20
minutes later.  

I don't think its a "big" bug.  It is matter first agreeing if we want
to permit the client to request as many addresses as it wants as long as
the server wants to give them to the client.  Once the client reaches
the policy line where its not allowed more addresses by the
administrator the server tells the client no more addresses are
available.

I think we handle all the cases except for what tells a server what to
do with an empty extension and the server has given the client all but
one more address it allows.  And the case you just defined in other mail
where there are no extensions specified we don't even discuss that case
in the spec and we should.  

We can't resolve the issue if we don't get to the details.  Which I
think we are now?  But there is a problem with the spec in this space
and I agree with Thomas it needs to be resolved.  What the answer is I
don't think Thomas and I agree.  But I think we are on the same page of
the problems?

/jim


From dhcp-v4@bucknell.edu  Sun Nov 21 13:20:46 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA18386
	for <dhc-archive@odin.ietf.org>; Sun, 21 Nov 1999 13:20:46 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA25388;
	Sun, 21 Nov 1999 13:20:15 -0500 (EST)
Date: Sun, 21 Nov 1999 13:20:15 -0500 (EST)
Message-Id: <199911211744.MAA00478@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> Therefore, I think to dismiss all these apps as being,
> somehow, broken, is extreme.

I would say "out of date," not broken.  And what will encourage them
to be brought up to date?  DHCP Clients that are not (and here I think
the word is used correctly) broken, in the sense of ignoring updates
from the servers.

			       _MelloN_



From dhcp-v4@bucknell.edu  Mon Nov 22 13:16:18 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA04667
	for <dhc-archive@odin.ietf.org>; Mon, 22 Nov 1999 13:16:18 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA20226;
	Mon, 22 Nov 1999 13:15:45 -0500 (EST)
Date: Mon, 22 Nov 1999 13:15:45 -0500 (EST)
Message-Id: <4.2.0.58.19991122094710.009579b0@mail.join.com>
Errors-To: droms@bucknell.edu
Reply-To: robs@join.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Rob Stevens <robs@join.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"; format=flowed
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 


>
>I would say "out of date," not broken.  And what will encourage them
>to be brought up to date?  DHCP Clients that are not (and here I think
>the word is used correctly) broken, in the sense of ignoring updates
>from the servers.
>
>

Fair enough.
I was interested in the comments you made about routing tables.
I am wondering what all the ramifications about changing masks
are on all the diverse platforms out there.

It's true that I can't immediately see that any problems would
occur, but my gut feeling is that there must be some subtleties
with this. Has anyone any deep insights on other platforms?

Rob.




From dhcp-v4@bucknell.edu  Mon Nov 22 16:18:54 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA25155
	for <dhc-archive@odin.ietf.org>; Mon, 22 Nov 1999 16:18:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA20399;
	Mon, 22 Nov 1999 16:17:25 -0500 (EST)
Date: Mon, 22 Nov 1999 16:17:25 -0500 (EST)
Message-Id: <199911221904.LAA28785@ha1mpk-mail.eng.sun.com>
Errors-To: droms@bucknell.edu
Reply-To: Vipul.Gupta@Eng.Sun.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Vipul Gupta <Vipul.Gupta@Eng.Sun.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: Send me your authentication draft comments/edits please
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4_28 SunOS 5.8 sun4u sparc 
Content-Type: TEXT/plain; charset=us-ascii
Mime-Version: 1.0


  Bill,
  
    We need to ensure that the base DHCP authentication draft and
  the public-key auth draft (draft-gupta-dhcp-auth-01.txt)
  are back in synch. We seem to have made incompatible changes to
  our drafts based on different interpretations of the WG feedback
  from Oslo :-).
  
    In particular, I no longer borrow two bits from the Algorithm
  field whereas the base draft does that now. Please review Section
  2.0 in draft-gupta-dhcp-auth-01.txt to see if that format makes
  sense for both drafts or should we choose something else.
  
  thanks,
  
  vipul
  
  
> Now that everyone has returned and plowed through their inboxes- you
> should have time to send me your edits/comments right? :-)
> 
> Several of you mentioned that you had edits- I won't call you out by name
> (but you know who you are). Please send me your edits/comments as soon as
> possible and I'll generate a new draft so we can review it prior to final
> call.
> 
> Thanks, Bill
> 
> -----------------------------------------------------------------------
> Bill Arbaugh			   
> email:  waa@dsl.cis.upenn.edu 
> web:    http://www.cis.upenn.edu/~waa
> -----------------------------------------------------------------------  
> 



From dhcp-v4@bucknell.edu  Mon Nov 22 21:38:40 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA08337
	for <dhc-archive@odin.ietf.org>; Mon, 22 Nov 1999 21:38:40 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA18358;
	Mon, 22 Nov 1999 21:38:14 -0500 (EST)
Date: Mon, 22 Nov 1999 21:38:14 -0500 (EST)
Message-Id: <14393.63922.53759.298172@spg-east-006>
Errors-To: droms@bucknell.edu
Reply-To: Peter.Memishian@East.Sun.COM
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Peter Memishian <Peter.Memishian@East.Sun.COM>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
X-Mailer: VM 6.43 under 20.4 "Emerald" XEmacs  Lucid
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit


 > It's true that I can't immediately see that any problems would
 > occur, but my gut feeling is that there must be some subtleties
 > with this. Has anyone any deep insights on other platforms?

well, i don't have any deep insights, but it seems to me that changing
the netmask on an interface is going to cause the same sorts of
problems that changing the ip address will; any applications that have
previously used the netmask to figure out the network address or
broadcast address are going to be confused at best.

of course, the real "problem" here is that the network topology has
changed; the fact that the dhcp client is providing the updated
netmask is just a side effect of this topology change.  so having the
client not provide the updated mask is worse; at least by providing
the updated mask, dhcp-aware applications have a chance of recovering.

-- 
meem


From dhcp-v6@bucknell.edu  Mon Nov 22 22:11:46 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA25876
	for <dhc-archive@odin.ietf.org>; Mon, 22 Nov 1999 22:11:45 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA09077;
	Mon, 22 Nov 1999 22:11:01 -0500 (EST)
Date: Mon, 22 Nov 1999 22:11:01 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.943325374.7215.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

> This is a tricky problem!  On the one hand, we might specify a
> list of extensions that servers MUST provide when first they create
> a client binding.  Alternatively, we might specify that clients
> have to include specific requests if they expect specific extensions
> to come back in a DHCP Reply message.  The problem with the former
> approach is that it is possible we don't know what extensions are
> always needed.  Surely the NTP server isn't one of them.  I'd
> guess that even DNS isn't always needed.  The problem with the latter
> approach is that it makes the DHCP Request message more bulky,
> especially
> unwlcome when dealing with slow media.
> 
> What if we don't pick either method, but we also require that
> clients MUST NOT depend upon server behavior in this matter?

We could do that but in practise this means picking "the client
must explicitly request all extensions it needs" since otherwise
some server might not automatically return something the client
has come to expect from some other server implementation/configuration.

The point is we will end up having to pick in any case. So why not
make the choice explicit and documented instead of something each
implementor has to figure out the hard way when there are interoperability
problems?

   Erik



From dhcp-v6@bucknell.edu  Mon Nov 22 22:12:54 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA26451
	for <dhc-archive@odin.ietf.org>; Mon, 22 Nov 1999 22:12:53 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA03552;
	Mon, 22 Nov 1999 22:11:17 -0500 (EST)
Date: Mon, 22 Nov 1999 22:11:17 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.943325577.24875.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

> I was talking about multi-homing in the provider sense not in the interface
> sense.  Clearly, DHCPv6 is a protocol that from the client's perspective
> operates on individual interfaces.
> 
> I was more concerned about cases where I have multiple providers and as a
> result possibly multiple DHCPv6 servers providing me with resources.  The
> most likely case that I can imagine for this to occur is for mobile IPv6
> but I can also imagine cases where I am using a cable modem or DSL system
> to provide me with packet transit but I am actually have multiple providers
> which use DHCPv6 to provide me with resources.

I understand how multi-interfaced will work; client sends DHCP messages
out all interfaces.

I don't understand how multi-ISPed will work. The client will
send a single DHCP message to a single relay. Will that relay be
configured (or automatically using multicast) to ask two DHCP servers
one in each ISP?
Even if that is the case how will the client be told about the (potentially
conflicting) information from the two ISP's DHCP servers?

   Erik



From dhcp-v6@bucknell.edu  Mon Nov 22 22:51:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA16009
	for <dhc-archive@odin.ietf.org>; Mon, 22 Nov 1999 22:51:04 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA24607;
	Mon, 22 Nov 1999 22:49:21 -0500 (EST)
Date: Mon, 22 Nov 1999 22:49:21 -0500 (EST)
Message-Id: <199911230346.TAA01709@feller.mentat.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Tim Hartrick <tim@mentat.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.





Erik,

> 
> I don't understand how multi-ISPed will work. The client will
> send a single DHCP message to a single relay. Will that relay be
> configured (or automatically using multicast) to ask two DHCP servers
> one in each ISP?

Right.

> Even if that is the case how will the client be told about the (potentially
> conflicting) information from the two ISP's DHCP servers?
> 

Nasty problem.  It would depend on the extension as to how one handles
conflicting parameterizations received from multiple servers.  A client
implementation that wanted to do this would need to be pretty smart.




Tim Hartrick
Mentat Inc.


From dhcp-v4@bucknell.edu  Mon Nov 22 22:52:26 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA16696
	for <dhc-archive@odin.ietf.org>; Mon, 22 Nov 1999 22:52:25 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA11099;
	Mon, 22 Nov 1999 22:52:15 -0500 (EST)
Date: Mon, 22 Nov 1999 22:52:15 -0500 (EST)
Message-Id: <199911230327.WAA00796@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: mellon@isc.org
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4


> any applications that have
> previously used the netmask to figure out the network address or
> broadcast address are going to be confused at best.

Other than DHCP servers, I can't think of any applications that would
need to do this...

			       _MelloN_


From dhcp-v4@bucknell.edu  Tue Nov 23 01:16:54 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id BAA01558
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 01:16:54 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id BAA19849;
	Tue, 23 Nov 1999 01:16:29 -0500 (EST)
Date: Tue, 23 Nov 1999 01:16:29 -0500 (EST)
Message-Id: <Pine.LNX.4.10.9911231127350.14749-100000@rnd.cts-corp.com>
Errors-To: droms@bucknell.edu
Reply-To: venkat@rnd.cts-corp.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: venkat@rnd.cts-corp.com
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: A clarification on load balancing (Failover)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: TEXT/PLAIN; charset=US-ASCII
Mime-Version: 1.0

Can anyone explain me as to why 32 bytes Hash Bucket Assignment option has
been used for load balancing. Instead we could have very well given the
starting and ending limits for each failover server as Pearson's hash
algorithm provides a distributed value between 0 & 255 & then we could
have checked if the hashed value falls between the defined limits.

Thanks & Regards,
Venkat.




From dhcp-v4@bucknell.edu  Tue Nov 23 09:25:13 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA16747
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 09:25:13 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id JAA01953;
	Tue, 23 Nov 1999 09:24:18 -0500 (EST)
Date: Tue, 23 Nov 1999 09:24:18 -0500 (EST)
Message-Id: <009E1919.DC966B2D.70@process.com>
Errors-To: droms@bucknell.edu
Reply-To: volz@process.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Bernie Volz <volz@process.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: RE: A clarification on load balancing (Failover)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4

>From: venkat@rnd.cts-corp.com
>Subject: A clarification on load balancing (Failover)
>
>Can anyone explain me as to why 32 bytes Hash Bucket Assignment option has
>been used for load balancing. Instead we could have very well given the
>starting and ending limits for each failover server as Pearson's hash
>algorithm provides a distributed value between 0 & 255 & then we could
>have checked if the hashed value falls between the defined limits.
>
>Thanks & Regards,
>Venkat.

What's wrong in allowing a discontinous set of hash buckets? We felt
that this was a much more generic way to handle it.

As far as any user interface/configuration issues, there is no reason
why a user could not enter a percentage, a begin/end range, individual
bucket numbers, whatever you want.

As far as the protocol goes, what are we taking about? Sending 32 bytes
vs 2 (one for start/one for end)? Who cares - this is only sent once
at connection initiation time and the extra 30 bytes are soon lost in
the noise.

Also, any server will likely turn this into a 32-byte (256-bit) mask
anyway to easily determine whether to respond or not. The code in
either case is likely the same in terms of "cost":

	if ((bucket# >= low-bucket) && (bucket# <= high-bucket))
	{ respond }
	else
	{ ignore }
vs:
	if ((hsa[bucket# >> 3] & (1 >> (bucket# & 7)) != 0)
	{ respond }
	else
	{ ignore }

And, consider how easy it is to turn one HSA into a second, just
complement the 32-bytes. Try doing that as easily with a low/high
range (note that the other server potentially has TWO ranges since
it has 0 to low, high to 255). So the above checking code works
only for one server.

If you wanted to simply things, you could send a single byte which
says the point at which to split the range.

When you consider the greater flexibility of the 32-byte buffer,
why not use it.

- Bernie Volz
  Process Software


From dhcp-v6@bucknell.edu  Tue Nov 23 10:57:22 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA05148
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 10:57:22 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA12530;
	Tue, 23 Nov 1999 10:50:43 -0500 (EST)
Date: Tue, 23 Nov 1999 10:50:43 -0500 (EST)
Message-Id: <383AB63C.BCD42BF7@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Erik and Tim,

Erik Nordmark == >>
Tim Hartrick  == >

>> I don't understand how multi-ISPed will work. The client will
>> send a single DHCP message to a single relay. Will that relay be
>> configured (or automatically using multicast) to ask two DHCP servers
>> one in each ISP?
> 
> Right.

I think we should distinguish the operation of the relay(s) from
the operation of the local (default) router(s).  Assume that there
is one router.  The router is likely to be advertising prefixes from
both ISPs.  Each relay is likely to have an address for the ISP that
it cares about.  If there is only one relay, it will have multiple
addresses.

If a client hears router advertisements with two managed-link
prefixes, the client should respond to both such router
advertisements with two distinct DHCP Solicitations and so on.
The protocol should proceed naturally from this point.

>> Even if that is the case how will the client be told about the (potentially
>> conflicting) information from the two ISP's DHCP servers?
 
> Nasty problem.  It would depend on the extension as to how one handles
> conflicting parameterizations received from multiple servers.  A client
> implementation that wanted to do this would need to be pretty smart.

It should not be the responsibility of the client to resolve
conflicting parameterizations.  In fact, a client DHCP host computer
should notify the user that there are conflicts, and then the user
should call the local administration and credit $200/hour to
the user's department away from the administrative budget for making
such mistakes and wasting peoples' time.  Part of the burden of having
two ISPs should be to make sure that the local DHCP servers are
compatible, or else that the ISPs have compatible DHCP service if
the DHCP servers are off-site.

It is probably impossible to make clients robust against all
administrative
errors.  They certainly are not so today.  I can hardly imagine them
being
so anytime in the foreseeable future.  I reckon there is a theorem that
says for any administrative domain, there are administrative errors that
can cause clients to become isolated from the network.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 23 10:57:51 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA05369
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 10:57:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA14740;
	Tue, 23 Nov 1999 10:49:45 -0500 (EST)
Date: Tue, 23 Nov 1999 10:49:45 -0500 (EST)
Message-Id: <383AB230.87A52A2F@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Hello Erik,

> We could do that but in practise this means picking "the client
> must explicitly request all extensions it needs" since otherwise
> some server might not automatically return something the client
> has come to expect from some other server implementation/configuration.

Not necessarily.  The downside of mandating that the client make
all requirements explicit, is a corresponding heavier load on
the DHCP Request, which might be sent over slow links.

Here is an alternative sequence:

- At bootup, client finds server address(es).
- Client sends DHCP Request.
- Server sends back all the extensions that are configured in the
  server database for that (class of) client
- Client wants some more extensions, so it asks for the ones
  it didn't get.

How does the client know which extensions it needs?  That depends
on the extension.  For instance, if the client needs DNS, it should
check during a later part of the bootup sequence whether DNS was
configured or not.  If the client needs a timeserver, the same answer
applies.  For many clients, these two items of information would
come automatically because the local administration has configured
their DHCP server that way.  Clients that perform no checking take
a certain amount of risk.  By way of analogy, I have seen programs
that do not check for errors after system calls.  The risk is that
these programs do not work in every circumstance.  We would like to
make it obvious how DHCP clients can work perfectly in every
circumstance.
Some clients might cut corners.  The risk is that those clients would
only work under certain conditions, and not in every circumstance.

Bottom line: we can make the protocol automatically fit every
circumstance by making it bulkier, or we can leave it slimmer by
requiring some design decisions to be made by the systems engineers
who build programs to use the protocol.

I am in favor of the latter in this case, since we can clearly
point out scenarios where the all-purpose cure is not needed and
adds substantially to the bandwidth burden for using the protocol.

> The point is we will end up having to pick in any case.

I hope that I have shown that this statement is not necessarily
true.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 23 11:57:51 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA04455
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 11:57:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA22779;
	Tue, 23 Nov 1999 11:53:15 -0500 (EST)
Date: Tue, 23 Nov 1999 11:53:15 -0500 (EST)
Message-Id: <199911231609.LAA03816@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> It should not be the responsibility of the client to resolve
> conflicting parameterizations.  In fact, a client DHCP host computer
> should notify the user that there are conflicts, and then the user
> should call the local administration and credit $200/hour to
> the user's department away from the administrative budget for making
> such mistakes and wasting peoples' time.  Part of the burden of having
> two ISPs should be to make sure that the local DHCP servers are
> compatible, or else that the ISPs have compatible DHCP service if
> the DHCP servers are off-site.

Maybe it's because I haven't been following this protocol very
carefully in the past, and don't know much about IPv6, but what you
are suggesting seems extremely unlikely to me.  What if the two ISPs
have two different sets of default routes, or two different sets of
domain name prefixes?  What if a client is supporting more than one
network interface?  I think the thing that resolves the conflicts
_has_ to be the client, because there's no other entity in a position
to do so.  The client may get different parameters from the _same_
DHCP server on two different interfaces, and this does not indicate a
configuration error on the part of the server administrator.

I also don't think it makes sense to specify in the draft what the
client should do in this case, but I think it's important to keep this
problem in mind when implementing clients, because this problem will
come up, and finger-pointing won't solve it.   It may or may not be
worthwhile to mention this problem in the draft and suggest solutions,
although that in itself is a path that is fraught with danger, as we
have seen in the DHCPv4 specification.

			       _MelloN_


From dhcp-v6@bucknell.edu  Tue Nov 23 11:58:46 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA04829
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 11:58:44 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA12904;
	Tue, 23 Nov 1999 11:54:16 -0500 (EST)
Date: Tue, 23 Nov 1999 11:54:16 -0500 (EST)
Message-Id: <199911231618.LAA03864@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> The downside of mandating that the client make
> all requirements explicit, is a corresponding heavier load on
> the DHCP Request, which might be sent over slow links.

How much heavier of a load can this create?  I think this is a
non-problem.  Furthermore, if you have the server send all the options
it has, that's almost certainly going to create a larger load than
having the client request all the options it needs.

> - Client wants some more extensions, so it asks for the ones
>   it didn't get.

This seems like a reasonable solution to the problem of getting
information to applications.

> I hope that I have shown that this statement is not necessarily
> true.

In fact, in the case of DHCPv4, failing to specify exactly what the
server and client should do in this very situation _has_ resulted in
differing and mutually incompatible implementations.  We now have
deeply-entrenched (i.e., in boot PROM) implementations from some
vendors that don't behave the way other vendors assumed they would,
and we have interoperability problems as a result.  Specifying exactly
what should be done here is a really good idea, IMHO.

OTOH, I think it's fine to say "the client should either request
all the extensions it needs, in which case the server should send only
those extensions, or it should not request any extensions, and the
server will send those extensions it has been configured to send."

			       _MelloN_


From dhcp-v6@bucknell.edu  Tue Nov 23 13:16:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA13947
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 13:16:56 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA28002;
	Tue, 23 Nov 1999 13:11:47 -0500 (EST)
Date: Tue, 23 Nov 1999 13:11:47 -0500 (EST)
Message-Id: <199911231708.MAA04049@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> What if the two ISPs
> have two different sets of default routes, or two different sets of
> domain name prefixes?

Sorry, I meant "domain name *suffixes*".   :'}

			       _MelloN_


From dhcp-v6@bucknell.edu  Tue Nov 23 13:17:02 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA14008
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 13:17:01 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA23528;
	Tue, 23 Nov 1999 13:12:53 -0500 (EST)
Date: Tue, 23 Nov 1999 13:12:53 -0500 (EST)
Message-Id: <383ACE59.4D3DADAF@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Hello again Ted,

>> == me
>  == Ted

>> The downside of mandating that the client make
>> all requirements explicit, is a corresponding heavier load on
>> the DHCP Request, which might be sent over slow links.
 
> How much heavier of a load can this create?  I think this is a
> non-problem.  Furthermore, if you have the server send all the options
> it has, that's almost certainly going to create a larger load than
> having the client request all the options it needs.

The server should be configured to do the right thing.  I don't
think it should automatically send all of the options that it
possibly could send.  But it might well be configured to automatically
send a few obvious ones like DNS, and default router.

The question about how much additional load on DHCP requests
is an interesting problem, and I lean heavily towards minimizing
it.  I expect DHCP to be used for getting care-of addresses for
Mobile IP, for instance, on heavily bandwidth-constrained
cellular radio links, and I'd rather see such requests be
allowed to be as lean as possible.  Other applications for
embedded computers (with minimal PROMs, as you suggest elsewhere)
also indicate the need for caution when laying down new mandatory
features.

> In fact, in the case of DHCPv4, failing to specify exactly what the
> server and client should do in this very situation _has_ resulted in
> differing and mutually incompatible implementations.  We now have
> deeply-entrenched (i.e., in boot PROM) implementations from some
> vendors that don't behave the way other vendors assumed they would,
> and we have interoperability problems as a result.  Specifying exactly
> what should be done here is a really good idea, IMHO.

I'm very much in agreement.  We should try to explicitly prohibit
any behavior that has led to problems in DHCPv4 implementations.
I'd rather do this by proscribing bad behavior than by mandating
new behaviors.

> OTOH, I think it's fine to say "the client should either request
> all the extensions it needs, in which case the server should send only
> those extensions, or it should not request any extensions, and the
> server will send those extensions it has been configured to send."

Can't we have it so that the client can request a particular IP address,
and the server send down the DNS server name without being asked?

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 23 13:17:10 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA14081
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 13:17:09 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA07804;
	Tue, 23 Nov 1999 13:12:29 -0500 (EST)
Date: Tue, 23 Nov 1999 13:12:29 -0500 (EST)
Message-Id: <383ACB6E.DC012647@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Ted,

Thanks for your note.

Ted Lemon wrote:

> > It should not be the responsibility of the client to resolve
> > conflicting parameterizations.  In fact, a client DHCP host computer
> > should notify the user that there are conflicts, and then the user
> > should call the local administration and credit $200/hour to
> > the user's department away from the administrative budget for making
> > such mistakes and wasting peoples' time.  Part of the burden of having
> > two ISPs should be to make sure that the local DHCP servers are
> > compatible, or else that the ISPs have compatible DHCP service if
> > the DHCP servers are off-site.
> 
> Maybe it's because I haven't been following this protocol very
> carefully in the past, and don't know much about IPv6, but what you
> are suggesting seems extremely unlikely to me.

I think that the protocol details we are discussing are not deeply
tied into anything that is IPv6 specific, except for the behavior
of stateful clients in response to router advertisements.

>                                               What if the two ISPs
> have two different sets of default routes, or two different sets of
> domain name prefixes?

These are good questions, but they are not questions that can
be answered within the DHCPv6 specification.

>                      What if a client is supporting more than one
> network interface?

I agree with the sentiment that says that each network interface should
be treated somewhat independently as far as the default router from
that interface.  Other considerations about choice of network interface
are not easily solved within DHCPv6.  There is a server preference field
that could be adapted for use for preference of network interface, but
I hope we don't have to solve that problem before going to Proposed
Standard.  These are big policy issues that could take years to decide.


>                  I think the thing that resolves the conflicts
> _has_ to be the client, because there's no other entity in a position
> to do so.

In a way I agree with you.  But I don't agree that the conflict
resolution
mechanism has to be specified in the DHCPv6 protocol document.  It's
just too far afield from where DHCP needs to be.  It's a problem that
affects the entire network programming paradigm in a way that is not
well understood today.  I suggest that we make DHCP work as best we
can and hope that multihoming and multi-network conflict resolution
be investigated more thoroughly as time goes on.


>              The client may get different parameters from the _same_
> DHCP server on two different interfaces, and this does not indicate a
> configuration error on the part of the server administrator.

As noted above, I don't think that different parameters on different
network interfaces _necessarily_ represents any configuration error.
So, yes, I agree that such a thing is possible, and in fact I believe
represents possibly correct behavior.

> I also don't think it makes sense to specify in the draft what the
> client should do in this case, but I think it's important to keep this
> problem in mind when implementing clients, because this problem will
> come up, and finger-pointing won't solve it.   It may or may not be
> worthwhile to mention this problem in the draft and suggest solutions,
> although that in itself is a path that is fraught with danger, as we
> have seen in the DHCPv4 specification.

Agreed, and I would support the inclusion of any relevant text to
lessen the dangers.  Perhaps you can offer some paragraphs from
your experience...?

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 23 13:17:17 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA14160
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 13:17:16 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA26191;
	Tue, 23 Nov 1999 13:13:07 -0500 (EST)
Date: Tue, 23 Nov 1999 13:13:07 -0500 (EST)
Message-Id: <199911231756.MAA04328@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> The server should be configured to do the right thing.  I don't
> think it should automatically send all of the options that it
> possibly could send.

But how can you guarantee that a server with a heterogenous client
base can send the right options to all the clients?   This is a hard
problem, and the only way to really get it right is to use vendor
classing and require the person setting up the server (or the server
vendor!) to track what clients want what, and have different defaults
for each.   OTOH, the client vendor _knows_ what options the client
wants and/or can use.   From the perspective of efficient distribution
of labor, putting the onus on the client to say what it wants seems
much better than putting it on the server.

What the ISC server currently does for DHCPv4 is to specify a very
small list of options that will be returned to any client, regardless
of what the client asks for - for example, the message type, the lease
time, the server identifier, the offered IP address and subnet mask,
the default route, and the DNS server list.  If the client sends a
parameter request list, the server sends those options in addition to
the "standard" options.   If the client sends no parameter request list,
the server sends first the options that I mentioned above, and then
all other options it has available, up to the available space in the
packet.

I can see having a configuration parameter that enumerates the set of
options the server will always send, but in practice I think requiring
the server administrator to configure this correctly is just going to
create problems.   Many DHCP server administrators are not
particularly knowledgable, and while deploying solutions that require
them to be configure stuff like this is just going to create a lot of
support work for DHCP server vendors (i.e., I'm going to have to
answer a lot of "stupid" questions on the dhcp-server mailing list).

> I expect DHCP to be used for getting care-of addresses for
> Mobile IP, for instance, on heavily bandwidth-constrained
> cellular radio links, and I'd rather see such requests be
> allowed to be as lean as possible.

Yes, but this argues for having the client say what parameters it
wants, because the client's request list is going to be way shorter
than even one option the server sends that's not useful to the
client.   How does the server know what clients are mobile and what
aren't, and further, how does it know what options any particular
mobile client will find useful?

> Other applications for embedded computers (with minimal PROMs, as
> you suggest elsewhere) also indicate the need for caution when
> laying down new mandatory features.

The days of 16k eproms are pretty much behind us at this point.  I
think the extra 16 or so bytes this would add to the PROM isn't likely
to be an issue for any new device.  Seriously, the latest Nokia phone
has enough extra space in the ROM to put in Tetris!  I don't think you
can legitimately use this as a reason for making a decision about how
to specify what the client should send.

> I'm very much in agreement.  We should try to explicitly prohibit
> any behavior that has led to problems in DHCPv4 implementations.
> I'd rather do this by proscribing bad behavior than by mandating
> new behaviors.

The problem is that the behaviour is unspecified, not that any
particular vendor is doing the wrong thing.  So people had to use
common sense to figure out what to do.  Unfortunately, different
people who I know personally and who are very sensible came up with
very different answers for what was common sense in this case.  I
would agree that we should not unreasonably limit flexibility, but we
must mandate specific behaviour to some degree, or the result will be
the same sorts of problems we've had with DHCPv4.

> Can't we have it so that the client can request a particular IP address,
> and the server send down the DNS server name without being asked?

Sure, but what if the client doesn't want the DNS name?  A mobile
client doesn't need to know the DNS name of its temporary IP address,
does it?  Maybe it doesn't care!  Was the parameter request list
longer or shorter than the DNS name the server sent?

What about specifically enumerating some small set of options that
must always be sent from the server to the client (e.g., IP address,
lease duration, message type, and so on) and then allowing the client
to either send a parameter request list, in which case it will get
only those additional parameters (possibly none!) that it specified,
or to send no parameter request list, in which case it will get
whatever the server is configured to send by default?

			       _MelloN_


From dhcp-v6@bucknell.edu  Tue Nov 23 13:33:41 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA13948
	for <dhc-archive@odin.ietf.org>; Tue, 23 Nov 1999 13:16:56 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA05256;
	Tue, 23 Nov 1999 13:12:40 -0500 (EST)
Date: Tue, 23 Nov 1999 13:12:40 -0500 (EST)
Message-Id: <199911231729.MAA04233@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> Agreed, and I would support the inclusion of any relevant text to
> lessen the dangers.  Perhaps you can offer some paragraphs from
> your experience...?

I'll have to think about this - I can write something from a DHCPv4
perspective, but I don't have any DHCPv6 experience, so it'd be tough
for me to write language that you could just drop in without some
serious cogitation.   I think it's worth it for me to do the
cogitating, but right now I'm cooking the failover protocol in my
brain, so I can't do this immediately, which likely means it'll get
dropped on the floor if it remains my responsibility.

Is there anybody else reading this who agrees that this is a problem,
understands DHCPv6, and would be willing to work with me to come up
with the correct verbiage?   I'd be happy to write something up from a
strictly DHCPv4 perspective and provide hints as to how I think it
might be adapted.

			       _MelloN_


From dhcp-v6@bucknell.edu  Wed Nov 24 18:38:56 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA12390
	for <dhc-archive@odin.ietf.org>; Wed, 24 Nov 1999 18:38:55 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA07108;
	Wed, 24 Nov 1999 18:37:06 -0500 (EST)
Date: Wed, 24 Nov 1999 18:37:06 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.943485182.18572.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0


> > Even if that is the case how will the client be told about the (potentially
> > conflicting) information from the two ISP's DHCP servers? > > 
> 
> Nasty problem.  It would depend on the extension as to how one handles
> conflicting parameterizations received from multiple servers.  A client
> implementation that wanted to do this would need to be pretty smart.

Hmm - does DHCPv6 specify what happens when the client sends one 
request message through the relay but gets 2 or more reply messages
from the relay?
My guess would be that it would ignore all but the 1st reply.

This would mean that it is the DHCP relay which would need to 
both "fork" the request to two different servers and then "merge"
the replies. I guess that could be done by a smart relay.

But I don't see how this would make sense business wise for
ISPs serving a home or a small office.
Wouldn't e.g. the cable TV ISP own the cable modem (or set-top box)
and the DSL ISP own the DSL modem?
Those boxes are likely to be the DHCP relays.
Thus you'd find two relays each willing to serve the hosts - and
handing out addresses from each ISP. So I can't see very the above
postulated "smart" relay would exist in practise.

Thus I think this can only be done if all the hosts in the home are
smart enough to either pick multiple DHCP relays/server and send
requests to all of them and merging the results.
Alternatively, the simply hosts would somehow just pick one relay/server.
Today such a simple host can't tell which ISP (or administration) is
running a server. If we think this is an important case perhaps the
DHCP advertise should contain something naming or identifying the
admin domain.

  Erik



From dhcp-v6@bucknell.edu  Wed Nov 24 18:38:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA12401
	for <dhc-archive@odin.ietf.org>; Wed, 24 Nov 1999 18:38:56 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA12342;
	Wed, 24 Nov 1999 18:36:17 -0500 (EST)
Date: Wed, 24 Nov 1999 18:36:17 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.943485395.21991.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #2, multiple IP addresses
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

> Not necessarily.  The downside of mandating that the client make
> all requirements explicit, is a corresponding heavier load on
> the DHCP Request, which might be sent over slow links.
> 
> Here is an alternative sequence:
> 
> - At bootup, client finds server address(es).
> - Client sends DHCP Request.
> - Server sends back all the extensions that are configured in the
>   server database for that (class of) client
> - Client wants some more extensions, so it asks for the ones
>   it didn't get.

I stand corrected - the above would work although it is a bit more complex than
the other two mechanisms.

Is this scheme something that everybody can live with?
If so we should document this issue and the above 
recommended solution in the draft.

   Erik



From dhcp-v6@bucknell.edu  Wed Nov 24 18:39:01 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA12454
	for <dhc-archive@odin.ietf.org>; Wed, 24 Nov 1999 18:39:01 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA02971;
	Wed, 24 Nov 1999 18:36:57 -0500 (EST)
Date: Wed, 24 Nov 1999 18:36:57 -0500 (EST)
Message-Id: <Roam.SIMC.2.0.6.943485582.17557.nordmark@jurassic>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Erik Nordmark <Erik.Nordmark@eng.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Mime-Version: 1.0

> If a client hears router advertisements with two managed-link
> prefixes, the client should respond to both such router
> advertisements with two distinct DHCP Solicitations and so on.
> The protocol should proceed naturally from this point.

BTW: The prefix in the router advertisement doesn't have a Managed bit.
The managed bit is part of the router advertisement.
Thus the above suggestion doesn't work as far as I can tell.
Perhaps there is some other way that the client can tell that there
are two different DHCP domains out there?

  Erik



From dhcp-v6@bucknell.edu  Thu Nov 25 07:39:12 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id HAA11518
	for <dhc-archive@odin.ietf.org>; Thu, 25 Nov 1999 07:39:11 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id HAA00425;
	Thu, 25 Nov 1999 07:36:19 -0500 (EST)
Date: Thu, 25 Nov 1999 07:36:19 -0500 (EST)
Message-Id: <01BF376D.657829C0.purushn@future.futsoft.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Purushothaman N <purushn@future.futsoft.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: BOOTP Relay agent
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
X-Mailer: Microsoft Internet E-mail/MAPI - 8.0.0.4211

Hi,
  pls. point out some web sites where i can get more info on BOOTP relay agent over WAN links.
Thanx & Regards,
N.Purushothaman



From dhcp-v6@bucknell.edu  Fri Nov 26 11:19:34 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA18074
	for <dhc-archive@odin.ietf.org>; Fri, 26 Nov 1999 11:19:33 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA31326;
	Fri, 26 Nov 1999 11:13:34 -0500 (EST)
Date: Fri, 26 Nov 1999 11:13:34 -0500 (EST)
Message-Id: <199911261543.KAA0000022946@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Hi Ted,

>Is there anybody else reading this who agrees that this is a problem,
>understands DHCPv6, and would be willing to work with me to come up
>with the correct verbiage?   I'd be happy to write something up from a
>strictly DHCPv4 perspective and provide hints as to how I think it
>might be adapted.

FYI I agree with you 100%.  Worst case just write it up with dhcpv4
lingo and we can add it to the next draft which has to reviewed anyway
before we do the follow on draft for PS.

thanks
/jim


From dhcp-v6@bucknell.edu  Fri Nov 26 12:55:28 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA03457
	for <dhc-archive@odin.ietf.org>; Fri, 26 Nov 1999 12:55:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA32340;
	Fri, 26 Nov 1999 12:51:57 -0500 (EST)
Date: Fri, 26 Nov 1999 12:51:57 -0500 (EST)
Message-Id: <199911261722.MAA0000005482@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: DISCUSSION: Use of Address Extensions with DHCPv6
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

folks,

I think we need to discuss some details here.  what do folks think about
the attached:

ONLY DISCUSSING ADDRESS EXTENSIONS other parameters and what clients or
servers should do is being discussed globally on another mail thread.

Semantics of messages:

1.  Client sends DHCPv6 Request with no address extension. 

1a. Server responds with all addresses available to the client 
    if the sys admin has defined the server can do this.

1b. If the server gets a second message from the client with no
    address extension it resends the client addresses it gave 
    them previously cumultatively (see below).

This would be the default behavior for DHCPv6 and how clients request
addresses.  The reason there is a sys admin decision is that the network
admin folks may have decided to use stateless addrconf and dhcpv6 is
only being used to provide non-address parameters to the clients or to
update DNS, etc.. 

2.  Client sends Request with 1 or more "empty" address extensions 
    
2a. Server responds, if the client is permitted a new address, with a new 
    IPv6 address in addition to the ones given to the client previously.
    If the client is not permitted a new address then an error is 
    returned to the client.
    The previous addresses are not resent to the client in the
    Server Reply.

2b. The client MUST NOT send this through a RELAY-AGENT and the 
    client is assumed to have an address with enough scope to 
    reach the server directly.  This permits the use of DHCPv6
    authentication or IPsec to be used to access the server.
    [In fact it probably should be a MUST have security for 
     this type of transaction.]

This would be a configured behavior permitted by sys admins.
To support this behavior we must require that the transaction-id be part
of the clients binding for the server/client database kept by the
server, when the client uses this feature.  The binding would be as follows:

    client-link-local-address
    dhcpv6-agent-address-prefix
    ipv6-address given to client
    valid and preferred lifetimes
    transaction-id from the request by the client

* we can add in the appendix that a server can structure the database
  so that the main key is link-local + agent-prefix and sub keys are
  the remaining parts.  We should also point out that when dhcpv6 
  operates in this mode there is more cost for memory and for 
  performance as the sever must now locate previous transactions 
  with this behavior or use a binding-cache (see discussion below).

3.  Client sends Request with address extension or extensions
    which are not empty.

3a. Server adds these addresses to the sever database for this 
    client.  Server sends Reply back to the client if it is 
    permitted to do this with the addresses the client sent
    back to the client in the Reply.

3b. The client MUST NOT send this through a RELAY-AGENT and the 
    client is assumed to have an address with enough scope to 
    reach the server directly.  This permits the use of DHCPv6
    authentication or IPsec to be used to access the server.
    [In fact it probably should be a MUST have security for 
     this type of transaction.]

We really need to discuss this scenario.  Do we want to permit this type
of behavior in DHCPv6.  This breaks the model of DHCPv4 yet again and is
really new stuff???????????  But we could make it work.

What is the affect of #3 above to any future server to server protocol,
could clients "updating" dhcpv6 server databases throw a synched server
to server protocol database out of synch?

BINDING-CACHE:  We would move the binding cache discussion to the
appendix as the above I think fixes idempotent problems in the protocol
standard.  The binding-cache would also lesson the cost/performance hit
I speak of above in implementations.  The server would only need to
store the Reply that has to be resent and an not the entire client
database.

Comments and Shoot Holes in the above please...if there are any...

Or just say you only want the default and why?

Whatever but if we can wrap this one up we will have made great progress
for the next rev of the spec.

thanks
/jim


From dhcp-v6@bucknell.edu  Sat Nov 27 14:57:50 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA06533
	for <dhc-archive@odin.ietf.org>; Sat, 27 Nov 1999 14:57:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA23744;
	Sat, 27 Nov 1999 14:55:35 -0500 (EST)
Date: Sat, 27 Nov 1999 14:55:35 -0500 (EST)
Message-Id: <384031DF.590035D7@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Erik Nordmark wrote:


> Hmm - does DHCPv6 specify what happens when the client sends one
> request message through the relay but gets 2 or more reply messages
> from the relay?
> My guess would be that it would ignore all but the 1st reply.

This should never happen.  DHCP Requests are addressed to a specific
DHCP Server.  I don't see now how it is that a relay would duplicate
or fork such a request message (or a Reply).

There has been thought given to the possibility of using multicast
messages for requests, but it has not been specified.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Sat Nov 27 15:34:29 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA15541
	for <dhc-archive@odin.ietf.org>; Sat, 27 Nov 1999 15:34:29 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA27059;
	Sat, 27 Nov 1999 15:32:55 -0500 (EST)
Date: Sat, 27 Nov 1999 15:32:55 -0500 (EST)
Message-Id: <384037A1.55CEFBAF@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Charles Perkins <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Erik Nordmark wrote:

> > If a client hears router advertisements with two managed-link
> > prefixes, the client should respond to both such router
> > advertisements with two distinct DHCP Solicitations and so on.
> > The protocol should proceed naturally from this point.
> 
> BTW: The prefix in the router advertisement doesn't have a Managed bit.
> The managed bit is part of the router advertisement.
> Thus the above suggestion doesn't work as far as I can tell.
> Perhaps there is some other way that the client can tell that there
> are two different DHCP domains out there?

How about if we say that a router advertisement with the 'M' bit
set means that all prefixes advertised in the extensions are for
managed links?

If a router has some managed links and some unmanaged links,
then it needs to send out two different router advertisements,
one for the managed links, and one for the unmanaged links.

Is this reasonable?

I note that such separation of advertisements would similarly
be required whenever the different prefixes required other parameters
to be different (besides just whether the links were managed).
For instance, if a router wanted to be default router for some
but not all of the prefixes, separate advertisements would be
needed.

If more ambitious suggestions were in order, I might suggest
the creation of an 'M' flag from the reserved flags of the Prefix
Information option.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Mon Nov 29 04:51:10 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id EAA08209
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 04:51:10 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id EAA24797;
	Mon, 29 Nov 1999 04:45:50 -0500 (EST)
Date: Mon, 29 Nov 1999 04:45:50 -0500 (EST)
Message-Id: <199911290936.KAA23112@givry.inria.fr>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Francis Dupont <Francis.Dupont@inria.fr>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

 In your previous mail you wrote:

   Can others please respond to this issue Thomas, Francis, Barr, Stig,
   Tim, et al...
   
=> I believe this problem can be solved if ways to use DHCPv6 are described
in a separate document:
 - we keep mechanisms in current I-Ds which should become PS ASAP
 - we explain how to implement/use DHCPv6 in an application document
   which doesn't need to become a PS today
We have this kind of things for routing protocols and I think this is a
good way to split theory and pratice.
Another point, I believe we have to update/improve autoconfig specs
because interactions between stateless (which is described in these specs)
and statefull (aka DHCPv6) autoconfigs were/are not fully understood.
An application document can help us to see what are the problems and
will be easier for implementor feedback.

Is this a good solution for you (Thomas and I-D authors)?

Regards

Francis.Dupont@inria.fr


From dhcp-v6@bucknell.edu  Mon Nov 29 04:52:00 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id EAA08412
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 04:51:57 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id EAA03058;
	Mon, 29 Nov 1999 04:48:25 -0500 (EST)
Date: Mon, 29 Nov 1999 04:48:25 -0500 (EST)
Message-Id: <199911290944.KAA13866@givry.inria.fr>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Francis Dupont <Francis.Dupont@inria.fr>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

 In your previous mail you wrote:

   How about if we say that a router advertisement with the 'M' bit
   set means that all prefixes advertised in the extensions are for
   managed links?
   
=> the 'M' bit meaning is "this link is a managed link".
Of course it applies to all prefixes of the managed link.

   If a router has some managed links and some unmanaged links,
   then it needs to send out two different router advertisements,
   one for the managed links, and one for the unmanaged links.
   
=> I can't see your problem here, router advertisements on different
links are simply router advertisements on different links...

   I note that such separation of advertisements would similarly
   be required whenever the different prefixes required other parameters
   to be different (besides just whether the links were managed).
   For instance, if a router wanted to be default router for some
   but not all of the prefixes, separate advertisements would be
   needed.
   
=> to be a default router is at most a link property (usually
it is a node property, ie. there is one default router for the whole node).

   If more ambitious suggestions were in order, I might suggest
   the creation of an 'M' flag from the reserved flags of the Prefix
   Information option.
   
=> there is an 'A' bit for per prefix control of autoconfiguration
(its meaning is specified for stateless autoconfiguration and still
unknown for statefull autoconfiguration).

Regards

Francis.Dupont@inria.fr

PS: please don't discuss about multiple interface links (:-)...


From dhcp-v4@bucknell.edu  Mon Nov 29 11:57:33 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA26938
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 11:57:31 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id LAA27715;
	Mon, 29 Nov 1999 11:55:56 -0500 (EST)
Date: Mon, 29 Nov 1999 11:55:56 -0500 (EST)
Message-Id: <4.2.0.58.19991129105541.0167e4a0@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: kkinnear@cisco.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Kim Kinnear <kkinnear@cisco.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Failover conference call(s)
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 

  
	
*****
** Conference Call Announcement to discuss the
** Failover draft -- you *must* reply to participate!
*****

We are planning to revise the failover draft again, both to
tighten up the technical details, as well as to tighten up the
text itself in an effort to shorten the draft.  The plans are to
do this well *before* the next deadline, and possibly produce two
revisions before the next IETF.

Prior to this next revision, we would like to gather what input
people wish to give on the current failover draft:

http://www.ietf.org/internet-drafts/draft-ietf-dhc-failover-05.txt

In order to gather input on the latest version  of the draft
prior to the next revision, I'm proposing one (or possibly two,
if necessary) conference calls next week, the week of December
6-10, 1999.  The 12pm-3pm EST slot worked pretty well for the
folks that attended last time, although we will organize these
calls around the folks that want to attend this time.

Proposal:

     Call 1:      Tuesday, December 7, 1999 12pm-3pm EST

	and if necessary:

     Call 2:      Thursday, December 9, 1999 12pm-3pm EST


**************************************************
         Conference Call Announcement
        YOU MUST REPLY TO PARTICIPATE
**************************************************

If you want to participate, I need email from you with answers to
the following questions:

1.  Your name(s), affiliation, and TIME ZONE.

2.  The topics in the current failover draft
(http://www.ietf.org/internet-drafts/draft-ietf-dhc-failover-05.txt)
which you would like to discuss in a conference call.  Please
order a list by priority.  Examples might be: integration with
the load balancing draft, security, how to better organize the
draft, etc.  

3.  Will Tuesday December 7, 12-3pm EST work for you, and will
Thursday December 9, 12-3pm EST work for you?  If not, please 
make some proposals.

4.  These will be toll free calls inside of the continental US.
If you are outside of the toll free area, and would need
financial assistance as a condition of participation in such a
call, please mention that and we'll see what we can work out.

Send your reply *only* to kkinnear@cisco.com (i.e., don't reply
to the whole dhcp-v4 list!).

************************************************

Along about Thursday, December 2, I'll let all of the respondents
know if the proposed dates worked, schedule the actual phone
calls if they did, and distribute the phone numbers.

Cheers -- Kim

Kim Kinnear
Cisco Systems
Chelmsford, MA 01824

(978) 244-8376
kkinnear@cisco.com




From dhcp-v6@bucknell.edu  Mon Nov 29 14:29:51 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA03003
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 14:29:51 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA04332;
	Mon, 29 Nov 1999 13:53:17 -0500 (EST)
Date: Mon, 29 Nov 1999 13:53:17 -0500 (EST)
Message-Id: <4D0A23B3F74DD111ACCD00805F31D8101CA21752@RED-MSG-50>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Richard Draves <richdr@microsoft.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
X-Mailer: Internet Mail Service (5.5.2650.21)

> From: Charles Perkins [mailto:charliep@iprg.nokia.com]
> How about if we say that a router advertisement with the 'M' bit
> set means that all prefixes advertised in the extensions are for
> managed links?
> 
> If a router has some managed links and some unmanaged links,
> then it needs to send out two different router advertisements,
> one for the managed links, and one for the unmanaged links.
> 
> Is this reasonable?

I've just joined this discussion so I'm not up to speed... Are you saying,
send different RAs on different links (of course this is OK) or send two
different RAs on the same link (maybe OK, but most implementations today
don't support it I would guess).

> I note that such separation of advertisements would similarly
> be required whenever the different prefixes required other parameters
> to be different (besides just whether the links were managed).
> For instance, if a router wanted to be default router for some
> but not all of the prefixes, separate advertisements would be
> needed.

This is confused - a router either advertises itself as a default router or
not, there's no concept of being "default router for a prefix".

Rich


From dhcp-v6@bucknell.edu  Mon Nov 29 14:45:45 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA09999
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 14:45:43 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA25002;
	Mon, 29 Nov 1999 14:42:45 -0500 (EST)
Date: Mon, 29 Nov 1999 14:42:45 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE8E@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


..boy!  take a few vacation days and see what you miss??!!!??!!!

My apologies in advance to the group for my replies which may seem a bit out
of date after a week.....

--Barr


[Tim Hartrick wrote on 17 November]

> If we are going to have releaseable resources and we aren't going to
> specify a server-to-server protocol to coordinate resource allocation then
> I don't see how the default behavior can be anything but to release the
> resources to the server which allocated them.  The alternative of
> releasing
> a resource to any old server could result in servers believing resources
> are in use when they are not and resources being released to servers which
> aren't managing the resource being released.  This is bad.
> 
[Barr Hibbs]

--I completely agree:  I see no other way that the clients (and servers)
could operate.

[...stuff deleted...]

> If we believe that this is going to be a common configuration then it
> might
> be best to just bite the bullet and deal with the complexity that will
> allow
> clients to handle getting resources from multiple, possibly unrelated,
> servers.
> 
[Barr Hibbs]

..while I agree that addressing this now is probably best for the long-term
development of the protocol, I have some reservations about attempting to
completely specify this mode of operation -- we want to complete work on the
protocol sometime soon or it will be irrelevant.

[...stuff deleted...]

> It is hard to see what good could come from releasing resources to a
> server that didn't allocate them in the first place.  I can imagine
> circumstances in which it would work but they all require thinking about
> server-to-server protocols which is a rat-hole that we desperately want to
> avoid I think.
> 
[Barr Hibbs]

Yes.  Especially since "multiple" servers seems to imply more than 2, which
is the only DHCPv4 case currently being considered.  An "n-way" interserver
protocol is probably much more complex than the 2-way failover protocol, so
we should not tie DHCPv6 to the existence of an inter-server protocol.




From dhcp-v6@bucknell.edu  Mon Nov 29 14:46:15 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA10239
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 14:46:13 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA06555;
	Mon, 29 Nov 1999 14:43:42 -0500 (EST)
Date: Mon, 29 Nov 1999 14:43:42 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE92@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


..boy!  this backlog is quite something....

--Barr


> Charles Perkins <charliep@iprg.nokia.com> writes:
> 
> > I don't think that multi-homing _requires_ multiple DHCP servers,
> > but it certainly suggests the possibility of it.  This would be
> > particularly true if the multi-homing were for unrelated ISPs.
> 
[Thomas Narten]

> "Possible" doesn't imply likely or desirable. What would be the
> benefit of such a configuration? At the least, it would seem to
> complicate client implementations (see below).
> 
[Barr Hibbs]

suppose we have one administrative domain for each network interface, then
multiple interfaces imply multiple servers because the likelihood of
different administrative domains having the same server is, I think,
vanishingly small....


[...stuff deleted...]

> Note: I am assuming that for the purposes of this dicussion, multiple
> DHCP servers means one server has one prefix, another server has a
> different prefix. I.e., the two servers do not have the same
> information about what prefixes (or resources/parameters in general)
> to give out and a client must contact both of them to get all of the
> addresses/parameters it needs. That is what I understood Tim to be
> getting at.
> 
[Barr Hibbs]

Okay, but what about the case where there are multiple servers per
administrative domain?

[Thomas Narten]

> Where in the current spec does it call out for a client to ask for
> addresses/parameters from more than one dhcp server?  My read of the
> spec is that it says you pick one server and get all your resources
> from it.
> 
[Barr Hibbs]

..see Erick's comments about "per-interface" servers...


[...stuff deleted...]

> Unless the draft clearly spells out when and under what conditions
> simultaneous transactions with multiple servers are to take place,
> clients won't implement the feature. 
> 
[Barr Hibbs]

..true, identifying these would be very useful, but in their absence,
implementors may do it anyway....


[...stuff deleted...]

> I suspect (but am not entirely sure) that nodes implementing mobile IP
> may have special requirements with regard to DHCP (e.g., a need to
> interact with servers at both the local and home site
> simultaneously). In such cases, it would probably make sense to have
> mobile nodes (as part of the mobile IP spec) specify those
> requirements as part of what implementing mobile IP means. It is not
> clear to me that all clients would need to implement this. In any
> case, we need to talk specifics here in order to decide whether
> somethign is a general requirement or one appropriate for mobile
> nodes.
> 
[Barr Hibbs]

..roaming (but not mobile-IP) hosts will do this...




From dhcp-v6@bucknell.edu  Mon Nov 29 14:47:27 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA10746
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 14:47:26 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA07130;
	Mon, 29 Nov 1999 14:43:09 -0500 (EST)
Date: Mon, 29 Nov 1999 14:43:09 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE8F@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0



> ...again, my apologies to the group for responding after taking several
> vacation days...
> 
> --Barr
> 
> 
> (From Thomas Narten)
> 
> Let me clarify. I agree that releasing resources via the server one
> got the resources from makes sense.  Having that be the default seems
> fine.  However, the current draft goes further and essentially says:
> you can't release via another server, period. The middle ground that
> has been suggested is to first try releasing via the server one got
> the resource from in the first place.  If that fails, however, then
> fall back by trying other servers. Having a fall back might make
> things work in the case that another server was available. This could
> happen via the server-server protocol, but could also happen if a DHCP
> server was retired and replaced by another with configuation
> information migrated from one server to the other manually.  Thus,
> there is no requirement that a server-server protocol be in place, but
> we'd be leaving in hooks to make things work better in such an
> environment.
> 
> [Barr Hibbs]
> 
> Thomas, I disagree:  if a server is being retired gracefully by an
> administrator, that administrator probably will not really care if every
> last resource is released because the retirement process will most likely
> first entail stopping the soon-to-be-retired server from allocating any
> resources, then simply waiting the anticipated lifetime of all assigned
> resources, leaving any that fall out to simply be handled as exceptions.
> 
> I also agree that we shouldn't assume a server-server protocol is
> present (since it doesn't quite exist yet), but we should also keep in
> mind that changing clients later when/if we do have a server-server
> protocol will be very difficult (i.e., we may not have the opporunity
> of DHCPv6 clients are already deployed). Thus, anything we can do now
> that helps the server-server case should be considered.
> 
> [Barr Hibbs]
> 
> while the DHCPv4 failover protocol is proceeding, I don't see an
> interserver protocol work item for DHCPv6 on our agenda:  that strongly
> suggests that one won't be forthcoming for at least a year, so unless
> Charlie or Jim or Kim or Bernie or someone else takes that as a project,
> spending any great amount of time on DHCPv6 "hooks" or "features" to
> support or that depend on an interserver protocol is, I think, a serious
> tactical error....
> 
> [...stuff deleted...]
> 
> If you think that the normal situation would be for a client to need
> to talk to multiple DHCP servers to get its configuration, the current
> protocol may well be inadequate for that. It has the clear notion that
> you locate one DHCP server and then get all your configuration
> information from it. There is no notion in the protocol that one 
> should be conversing with multiple servers simultaneously.  Do you
> think this is a requirement?
> 
> [Barr Hibbs]
> 
> What about "gateway-like" devices with NICs on different network segments?
> 
> 


From dhcp-v6@bucknell.edu  Mon Nov 29 14:47:40 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA10860
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 14:47:40 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA01091;
	Mon, 29 Nov 1999 14:43:18 -0500 (EST)
Date: Mon, 29 Nov 1999 14:43:18 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE90@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


..more catching up on the discussion thread...

--Barr


[Tim Hartrick]

> As an example, lets assume that I have a "standard" cable modem in my
> house.  This modem and my cable company allow me to choose from any number
> of providers or multiple providers.  If I did indeed have multiple
> providers then I could be interacting with two different DHCPv6 servers
> managed by those two providers.  They don't know anything about each
> other and it makes no sense for me to attempt to release an address
> that was allocated by provider A to the server administered by provider
> B.  This is a unlikely to be common but given appropriate competition in
> the market it won't be impossible.
> 
[Barr Hibbs]

while I think I disagree as to the likelihood of this occurring, I
definitely do agree that your point is well-taken....


[...stuff deleted...]

> I agree that we probably don't want to say that the client MUST release
> the resources to the original server.  That precludes the possibility of
> designing server-to-server and server redundancy protocols.  All I am
> saying is that in order to get things to work correctly in the more common
> case we need to say that the client SHOULD release resources to the
> original server and if that fails then they MAY release them to some other
> known server with the understanding that absent some adminstrative co-
> ordination, the description of which is out of scope for the draft, the
> release may not work.  That's it.  Nothing more complicated than that.
> 
[Barr Hibbs]

How about something like:  "Clients must release resources to the
[administrative] domain which assigned them."

-- obvious problems are how to identify domains, and
-- what is a domain has multiple servers and no interserver protocol?




From dhcp-v6@bucknell.edu  Mon Nov 29 14:48:21 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA11139
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 14:48:19 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA03778;
	Mon, 29 Nov 1999 14:43:30 -0500 (EST)
Date: Mon, 29 Nov 1999 14:43:30 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE91@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


..still catching up...

--Barr


[Erick Nordmark]

> Dealing with multi-homed machines, or to be more specific,
> multi-interfaced
> or multi-linked machines doesn't need anything more than stating that
> DHCPv6 operates on a per-interface basis plus some way to resolve
> differences for single-valued parameters.
> 
> 
[Barr Hibbs]

Yes.  That seems to simplify several issues.



From dhcp-v6@bucknell.edu  Mon Nov 29 14:48:39 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA11263
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 14:48:37 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id OAA07430;
	Mon, 29 Nov 1999 14:43:55 -0500 (EST)
Date: Mon, 29 Nov 1999 14:43:55 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE93@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


..still catching up...

--Barr


> > As an example, lets assume that I have a "standard" cable modem in my
> > house.  This modem and my cable company allow me to choose from any
> number
> > of providers or multiple providers.  If I did indeed have multiple
> > providers then I could be interacting with two different DHCPv6 servers
> > managed by those two providers.
> 
> My expectation would be that in this case the connectivity provider
> would arrange to have your DHCP packets delivered to the DHCP server
> of your ISP and NOT to the other DHCP servers. Or, the various DHCP
> servers would get the client requests, but only the client's selected
> ISP would respond. In both cases, the client would appear to be
> interacting with only 1 DHCP server, not multiple ones, and I think
> think this should work fine today.
> 
[Barr Hibbs]

is this process for ISP selection specified anywhere?


[...stuff deleted...]

> As a data point, I have an ADSL connection to a telco (GTE). GTE also
> provides ISP services. However, my ISP is not GTE, it is someone
> else. I use DHCP to get addresses and configuration info. It just
> works without any special configuration on my part. GTE arranges to
> tunnel all of my non-local traffic via bridging to my ISP.
> 
[Barr Hibbs]

..I'm missing something here:  you have one transport provider and a
different ISP.  Why is this at all unusual?




From dhcp-v6@bucknell.edu  Mon Nov 29 15:43:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA05512
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 15:43:04 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA00131;
	Mon, 29 Nov 1999 15:40:11 -0500 (EST)
Date: Mon, 29 Nov 1999 15:40:11 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE94@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


..continuing to catch up...

--Barr


[Charlie Perkins]

> DHCP(v6) is at the service of network computers and applications.
> DHCP is not the application.  It is constructed to offer benefits
> to applications.  If an application needs to get configuration
> information from each of two DHCP servers at different ISPs, then
> it is the job of DHCP to make that happen.  The benefit of having
> DHCP work to serve that possibility is to simplify application
> programming.  Insofar as applications care about multi-homing and
> multiple ISPs, I think it is quite clear that they will need such
> features from DHCP.  If you want to argue that applications should
> not ever care about multi-homing and ISPs, then I can't add anything
> more to that discussion except my intuition that it would be a
> terrible mistake to remove the relevant features from DHCP.
> 
[Barr Hibbs]

AGREED!!!


> > Where in the current spec does it call out for a client to ask for
> > addresses/parameters from more than one dhcp server?  My read of the
> > spec is that it says you pick one server and get all your resources
> > from it.
> 
> Then you are reading something into the spec that isn't anywhere in
> there.  The job of a specification is to make protocol operations
> possible, not to tell clients when to use the protocol.  The specification
> also
> does not say what kind of routers to use, nor how many bits per second
> of bandwidth is required, nor whether mobile computers can use the
> messages, nor whether a collision-detection scheme is needed at
> layer-2, nor whether the server must store its data using LDAP, nor
> any of a thousand other details.  All of these thousand other details
> are ABSOLUTELY CRUCIAL for lots of implementations, and yet the
> specification does not discuss them.
> 
[Barr Hibbs]

..how about "one server selected per [administrative] domain?"


[...stuff deleted...]

> It is a mistake to follow the road of prescribing every possible
> application interaction with a protocol.  Not to mention time-consuming,
> tedious, and self-limiting.  Such an authoritarian view of protocol
> design is almost always self-defeating, I think.
> 
[Barr Hibbs]

yes, Yes, YES!!!




From dhcp-v6@bucknell.edu  Mon Nov 29 15:43:30 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA05727
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 15:43:29 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA01094;
	Mon, 29 Nov 1999 15:41:04 -0500 (EST)
Date: Mon, 29 Nov 1999 15:41:04 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE96@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0



..still catching up with the discussion thread...

--Barr


[Jim Bound wrote...]

> You need to help us here and respond to this abstract issue as it can
> reduce the time to get this spec in shape and fix the real protocol bugs
> we are finding from this discussion.
> 
[Barr Hibbs]

..if I may generalize....

The Europeans and Asians desperately want an entire suite of protocols for
IPv6 -- this discussion is ultimately productive if it results in a more
robust protocol, but if we are simply bogged down in the minutiae of a
theoretical discussion of protocol specifications, then we waste everyone's
time. 

Seems to me that we have the basics already:

1. multiple servers in multiple domains are both possible and likely

2. absent interserver AND interdomain protocols we must address releasable
resources in the DHCPv6 draft

3. while it places a small burden on a client, requiring resources to be
released to the assigning entity is simply the best, most workable starting
point -- we are now including versioning information which permits us the
means to modify the protocol in the future if implementations can't make
this work





From dhcp-v6@bucknell.edu  Mon Nov 29 15:43:41 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA05829
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 15:43:40 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA14593;
	Mon, 29 Nov 1999 15:41:19 -0500 (EST)
Date: Mon, 29 Nov 1999 15:41:19 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE97@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0



..almost caught up!

--Barr


[Erik Nordmark]

> I don't understand how multi-ISPed will work. The client will send a
> single DHCP message to a single relay. Will that relay be configured (or
> automatically using multicast) to ask two DHCP servers one in each ISP?
> Even if that is the case how will the client be told about the
> (potentially conflicting) information from the two ISP's DHCP servers?
> 
[Barr Hibbs]

This is NOT impossible!

--either client request contains some discriminator allowing server to
ignore,

--or--

--mechanism for specifying an ISP exists,

--or--

--clients/servers must deal with "stray" requests/responses





From dhcp-v6@bucknell.edu  Mon Nov 29 15:43:55 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA05956
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 15:43:54 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA25539;
	Mon, 29 Nov 1999 15:41:31 -0500 (EST)
Date: Mon, 29 Nov 1999 15:41:31 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE98@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


..still catching up!

--Barr


[Charlie Perkins]

> I think we should distinguish the operation of the relay(s) from
> the operation of the local (default) router(s).  Assume that there
> is one router.  The router is likely to be advertising prefixes from
> both ISPs.  Each relay is likely to have an address for the ISP that
> it cares about.  If there is only one relay, it will have multiple
> addresses.
> 
[Barr Hibbs]

Charlie, one router is NOT a good assumption.  In our case we NEVER
provision a network segment with fewer than two routers, and for many busy
or critical segments we often use three or four, each connected to different
paths.

{Charlie Perkins]

> If a client hears router advertisements with two managed-link
> prefixes, the client should respond to both such router
> advertisements with two distinct DHCP Solicitations and so on.
> The protocol should proceed naturally from this point.
> 
[Barr Hibbs]

Good!!

> >> Even if that is the case how will the client be told about the
> (potentially
> >> conflicting) information from the two ISP's DHCP servers?
>  
> > Nasty problem.  It would depend on the extension as to how one handles
> > conflicting parameterizations received from multiple servers.  A client
> > implementation that wanted to do this would need to be pretty smart.
> 
> It should not be the responsibility of the client to resolve
> conflicting parameterizations.  In fact, a client DHCP host computer
> should notify the user that there are conflicts, and then the user
> should call the local administration and credit $200/hour to
> the user's department away from the administrative budget for making
> such mistakes and wasting peoples' time.  Part of the burden of having
> two ISPs should be to make sure that the local DHCP servers are
> compatible, or else that the ISPs have compatible DHCP service if
> the DHCP servers are off-site.
> 
[Barr Hibbs]

..uh... isn't this the client/user's responsibility?





From dhcp-v6@bucknell.edu  Mon Nov 29 15:44:27 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA06229
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 15:44:26 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA25551;
	Mon, 29 Nov 1999 15:40:52 -0500 (EST)
Date: Mon, 29 Nov 1999 15:40:52 -0500 (EST)
Message-Id: <21839C593E75D211BF7800805FCCD026EAAE95@msgsrv03.srv.PacBell.COM>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "HIBBS, BARR (PB)" <RBHIBBS@msg.pacbell.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain
X-Mailer: Internet Mail Service (5.5.2448.0)
Mime-Version: 1.0


> ...still catching up!
> 
> --Barr
> 
> 
> [Jim Bound]
> 
> Your read of the spec is an interpretation of it what it can do  but not
> what it can't do.  There is nothing in the protocol that prohibits
> multiple servers. How the client does this is the implementors business.
> If folks want an appendix to describe an implementation scenario where:
> 
>   1. a new interface comes up on the same link (same sever though)
>      this is easy.
>   2. a new interface comes up on a link that did not exist and 
>      may be a new server after the solicit goes out.
> 
> [Barr Hibbs]
> 
> ...mmmmm....  shouldn't a solicit always be sent in this case?  at least,
> if it is necessary to make the protocol work gracefully....
> 
> 
> [...stuff deleted...]
> 
> It could be when I implement this completely that we will find bugs and
> I think we all will.  What I don't think is possible is try to find
> every bug in this spec before it goes to PS and that is impossible IMO.
> To find the bugs left (other than Mikes issues he is raising) we need to
> go implement.
> 
> [Barr Hibbs]
> 
> ...while I don't agree that well-considered analysis CAN'T find all
> protocol bugs (at least, in theory), in practice test implementations
> really are needed to get to this point...
> 
> 
> [...stuff deleted...]
> 
> If we are to attempt to find every bug in these specs without going to
> PS we will not be successful be here for another 4 years discussing
> cause none of us are implememnting this spec until it goes to PS.
> 
> [Barr Hibbs]
> 
> Agreed!!
> 
> 
> [...stuff deleted...]
> 
> I also would like to say that this is not needed if it is needed to move
> this spec to PS.  And I would like to hear from others on this besides
> the Area Directors and us authors?
> 
> [Barr Hibbs]
> 
> Jim, I definitely agree with you on this point!
> 
> 


From dhcp-v6@bucknell.edu  Mon Nov 29 16:33:08 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA00243
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 16:33:08 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA27396;
	Mon, 29 Nov 1999 16:28:50 -0500 (EST)
Date: Mon, 29 Nov 1999 16:28:50 -0500 (EST)
Message-Id: <3842E9E8.30F6D3F6@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Barr,

Thanks for reading over all the mail and making comments!

>> I think we should distinguish the operation of the relay(s) from
>> the operation of the local (default) router(s).  Assume that there
>> is one router.
 
> Charlie, one router is NOT a good assumption.  In our case we NEVER
> provision a network segment with fewer than two routers, and for many busy
> or critical segments we often use three or four, each connected to different
> paths.

Sorry for being unclear.  I only made the assumption so that
the discussion would be easier to manage.  Having multiple
routers is fine, and doesn't change anything except for making
the sentence structure of the explanation more complicated.


>> == me

>> It should not be the responsibility of the client to resolve
>> conflicting parameterizations.  In fact, a client DHCP host computer
>> should notify the user that there are conflicts, and then the user
>> should call the local administration and credit $200/hour to
>> the user's department away from the administrative budget for making
>> such mistakes and wasting peoples' time.  Part of the burden of having
>> two ISPs should be to make sure that the local DHCP servers are
>> compatible, or else that the ISPs have compatible DHCP service if
>> the DHCP servers are off-site.

> = Barr Hibbs
> 
> ..uh... isn't this the client/user's responsibility?

I'm not sure what "this" is in your sentence.  My answer here is
when it means the burden of having compatible DHCP servers when
the servers are located at the ISPs' sites.

First, I am assuming the existence of a local network administration
that has negotiated the service agreements with the ISPs.  If there
is not such a local network administration, then it falls on the
shoulders
of the users to do this work.

But if there is such an administration, I should think that it is their
job to make sure that the user population can be served well in this
manner.  Most of the users would have to study for some while before
understanding all of the ways that multiple DHCP servers might be
incompatibly administered.

What about home users...?   Right now, I can't think of the right
answer for making dual-ISP home users able to rely on compatible
DHCP service.  On the other hand, I don't think this is sufficiently
critical that we have to solve the problem just now.  I think that
the solution, in any case, will depend upon answers related to the
general multi-homing problem.  I don't think we know those answers yet.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Mon Nov 29 16:33:12 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA00283
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 16:33:11 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id QAA12486;
	Mon, 29 Nov 1999 16:30:03 -0500 (EST)
Date: Mon, 29 Nov 1999 16:30:03 -0500 (EST)
Message-Id: <384524D0.C54F4D2D@east.sun.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Michael W. Carney" <michael.carney@east.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Summary: DC meeting question #1
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.6 [en] (X11; I; Linux 2.2.5-22 i686)
Content-Transfer-Encoding: 7bit



Hi folks,

I've sifted through the WG alias traffic for messages addressing this
question,
and I'll attempt to summarize. The question that was asked was:

Seeking examples of releasable resources. An IP address is one.
Others? When would they be released?

The consensus was while there are not ready examples of releasable
resources
over and above IP addresses, it was an useful exercise to consider
generic
releasable resources and what their semantics are. I've attempted to
summarize
what's been discussed, and take a stab at filling in holes:

Semantics of releasable resources:

1) General Form

    *	Releasable resource extensions need to have a general form so that
	clients/servers can manage them in a generic fashion.

    *	There is a generic bit in the extensions header which is set if
the
	extension is carrying a releasable resource. When this bit is set,
	there are two 32 bit fields present in the extension which represent
	the relative times in seconds when the client should start attempting
	to extend the lease on the resource, and when the client must stop
	using the resource if it is unable to extend it.

    *	Since extensions carrying releasable resources have a general
	form/semantics, clients/servers do not need to change when new
	extensions are defined which are releasable.

2) How are they allocated/requested, and how and to whom are they
released?

    *	Servers advertise releasable resources. We (WG) need to consider
how
	releasable resources are advertised.

    *	Clients MUST request releasable resources, otherwise the server
	will not allocate them.

    *	Servers allocate releasable resources for a period of time set by
	administrative policy

    *	A client can renew releasable resources w/ allocating server,
after
	the first timer expires. If unable to contact the server, the client
	must stop using resource when second timer expires.

    *	A client SHOULD release releasable resources to the allocating
server.
	If it fails to contact the allocating server, it MAY locate another
DHCP
	server and attempt to release them. In any case, the server will
recover
	the resource when its lease has expired. Thus, like DHCPv4, releasable
	resources (addresses) will ultimately be reclaimed as administrative
	policy dictates, and it isn't critical for clients to contact the
server.

3) Multiple Server Environments

    *	Servers advertise available IP address(es) and other releasable
resources,
	thus clients know what resources are available from which server.

    *	We should avoid over specifying how DHCP works within multi-server
	environments. However, an introductory/scenario section w/ a protocol
timeline
	would help validate the protocol model, and give some guidance to
implementors
	where needed.

Miscellaneous:

*	I think we do need an "extension request extension" - having such an
extension
	would simplify the advertisement of informational extensions, as well
as simply
	a client's method for requesting extensions. Note: We need to specify
that a
	policy switch should be available on the server to determine when an
extension
	is returned (because of administrative policy, or because client
requested it?).
	This point has been a bone of contention in DHCPv4, and we should
endeavor to
	clarify the semantics in DHCPv6.

*	Erik raised an interesting point about application layer extensions on
machines
	multiple interfaces: If you have multiple NTP servers available from
DHCP information
	from different interfaces, which one do you select? In our (Sun's)
DHCPv4 client
	implementation, we have a notion of a "primary" interface, from which
application
	layer parameters take precedence over those set in other interfaces. It
seems to
	me that a similar DHCPv6 implementation strategy could resolve this
issue. Charlie
	suggested that the server preference value be used to determine
preference. His
	suggestion has the advantage that preference is set by the
administrator. I prefer
	the former solution ("primary" interface).

--

Mike Carney


From dhcp-v6@bucknell.edu  Mon Nov 29 19:56:45 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA03352
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 19:56:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA11567;
	Mon, 29 Nov 1999 19:47:17 -0500 (EST)
Date: Mon, 29 Nov 1999 19:47:17 -0500 (EST)
Message-Id: <38455481.44912C5A@east.sun.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Michael W. Carney" <michael.carney@east.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DISCUSSION: Use of Address Extensions with DHCPv6
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.6 [en] (X11; I; Linux 2.2.5-22 i686)
Content-Transfer-Encoding: 7bit


Hi Jim,

Seems to me (after no small amount of consideration) that a server needs to
provide
more information in its advertisements then we have been proposing (note that
providing
this information doesn't change the protocol as defined at all). Consider my
summary of
question #1 - I've used it in responding to this message.

Jim Bound wrote:

> folks,
>
> I think we need to discuss some details here.  what do folks think about
> the attached:
>
> ONLY DISCUSSING ADDRESS EXTENSIONS other parameters and what clients or
> servers should do is being discussed globally on another mail thread.
>
> Semantics of messages:
>
> 1.  Client sends DHCPv6 Request with no address extension.
>
> 1a. Server responds with all addresses available to the client
>     if the sys admin has defined the server can do this.
>
> 1b. If the server gets a second message from the client with no
>     address extension it resends the client addresses it gave
>     them previously cumultatively (see below).
>
> This would be the default behavior for DHCPv6 and how clients request
> addresses.  The reason there is a sys admin decision is that the network
> admin folks may have decided to use stateless addrconf and dhcpv6 is
> only being used to provide non-address parameters to the clients or to
> update DNS, etc..

#1 goes against the constraint I listed in the Summary: DC meeting question #1
message: Clients must explicitly request releaseable resources. If the server
had explicit (through policy) or implicit (dynamic addresses) to offer to the
client,
it would do so by adding IP address extensions carrying these addresses to the
advertisement. The client would request one or all of these IP addreses in its
request. If the site is using stateless autoconf, then the server would not
have any
IP address extensions in its advertisement, and thus the client could not
request
an IP address.

>
>
> 2.  Client sends Request with 1 or more "empty" address extensions
>
> 2a. Server responds, if the client is permitted a new address, with a new
>     IPv6 address in addition to the ones given to the client previously.
>     If the client is not permitted a new address then an error is
>     returned to the client.
>     The previous addresses are not resent to the client in the
>     Server Reply.
>
> 2b. The client MUST NOT send this through a RELAY-AGENT and the
>     client is assumed to have an address with enough scope to
>     reach the server directly.  This permits the use of DHCPv6
>     authentication or IPsec to be used to access the server.
>     [In fact it probably should be a MUST have security for
>      this type of transaction.]

I'm assuming that this request is in addition to the IP address(es) offered in
the advertisement. The client will have knowledge of what networks are
available
from the advertisement, and thus would need to tell the server which network it

would like an IP address on. Thus, an empty IP address extension isn't enough -
the
client would need to communicate which network. At DC, I proposed a change to
the
IP address extension which would permit the client to identify which network it
wants
an additional IP address on. Whether or not it would receive an address is up
to
administrator policy. Note that a server could use this modification for
advertising
network prefixes, without advertising a specific IP address on that network.

>
>
> This would be a configured behavior permitted by sys admins.
> To support this behavior we must require that the transaction-id be part
> of the clients binding for the server/client database kept by the
> server, when the client uses this feature.  The binding would be as follows:

I don't see that the transaction id needs to be part of the binding. The XID is
cached
to enable a server to distinguish retransmissions from new requests. The
link-local
address, agent prefix, and IP address should be enough. Prehaps I'm missing
something.

>
>
>     client-link-local-address
>     dhcpv6-agent-address-prefix
>     ipv6-address given to client
>     valid and preferred lifetimes
>     transaction-id from the request by the client
>
> * we can add in the appendix that a server can structure the database
>   so that the main key is link-local + agent-prefix and sub keys are
>   the remaining parts.  We should also point out that when dhcpv6
>   operates in this mode there is more cost for memory and for
>   performance as the sever must now locate previous transactions
>   with this behavior or use a binding-cache (see discussion below).

Do we really want/need to specify this?

>
>
> 3.  Client sends Request with address extension or extensions
>     which are not empty.
>
> 3a. Server adds these addresses to the sever database for this
>     client.  Server sends Reply back to the client if it is
>     permitted to do this with the addresses the client sent
>     back to the client in the Reply.
>
> 3b. The client MUST NOT send this through a RELAY-AGENT and the
>     client is assumed to have an address with enough scope to
>     reach the server directly.  This permits the use of DHCPv6
>     authentication or IPsec to be used to access the server.
>     [In fact it probably should be a MUST have security for
>      this type of transaction.]
>
> We really need to discuss this scenario.  Do we want to permit this type
> of behavior in DHCPv6.  This breaks the model of DHCPv4 yet again and is
> really new stuff???????????  But we could make it work.
>
> What is the affect of #3 above to any future server to server protocol,
> could clients "updating" dhcpv6 server databases throw a synched server
> to server protocol database out of synch?

Jim, I don't think this is a good scenario. A Client should never be involved
in
telling a server about IP addresses (or other releasable resources) it may have

that the server has no knowledge of. Sure, a client MAY request that a certain
address be assigned to it, and if the request is valid from a network topology
point of view and a administrator policy view, then fine. I just don't think we

should be configuring DHCP servers with DHCP clients, IMHO.

>
>
> BINDING-CACHE:  We would move the binding cache discussion to the
> appendix as the above I think fixes idempotent problems in the protocol
> standard.  The binding-cache would also lesson the cost/performance hit
> I speak of above in implementations.  The server would only need to
> store the Reply that has to be resent and an not the entire client
> database.

Binding cache is very useful for preventing unnecessary database lookups. Ah,
now I see where you're coming from wrt saving the transaction ID - you're
attempting to record the assignment transaction to avoid confusion between
retransmissions and new requests.

Is it really necessary to be this heavy-weight? Sure, as specified, there is
a window where the server's XID timeout cache may timeout cache entries
while a client is retransmitting (server reboots, loses cache information), but

that window is very small. In most cases, the XID cache timeout will be large
enough and the server stable enough to prevent the accidental assignment of
additional IP address(es) (or other releasable resources). Alternately, a
server could write its XID cache to stable storage from time to time, and
reload it upon restart, thus satisfying the idempotent problem w/o having the
cost of the extra space in the database and extra initial database lookup.

>
>
> Comments and Shoot Holes in the above please...if there are any...
>
> Or just say you only want the default and why?
>
> Whatever but if we can wrap this one up we will have made great progress
> for the next rev of the spec.
>
> thanks
> /jim

--

Mike Carney





From dhcp-v6@bucknell.edu  Mon Nov 29 20:58:50 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29299
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 20:58:50 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA00174;
	Mon, 29 Nov 1999 20:56:07 -0500 (EST)
Date: Mon, 29 Nov 1999 20:56:07 -0500 (EST)
Message-Id: <Version.32.19991129150913.00e55ea0@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 01:53 PM 11/19/99 -0500, you wrote:
>
>>From Charlie Perkins:
>
>>This again relates to the discussion about whether clients should
>>be instructed when to use a protocol.  Since you see here how a Mobile
>>IP client might use two different servers, it makes sense to you.  Yet,
>>you do not allow for the operation to make sense _in the abstract_
>>(or, perhaps, "theoretical") case.  I claim that it is better to get
>>the protocol right in the abstract, knowing that there will be specific
>>cases.  I do not wish to get involved with listing the special cases
>>whereby clients might wish to interact with multiple DHCP servers.
>
>Working Group members this above from Charlie I strongly agree with and
>is the crux of our many disagreement with Thomas as a WG member inputing
>to the spec.  
>
>You need to help us here and respond to this abstract issue as it can
>reduce the time to get this spec in shape and fix the real protocol bugs
>we are finding from this discussion.

Put succinctly, I know of no abstract reason to preclude the concurrent
interoperation of a DHCP client with multiple servers.  In my opinion, the
current spec does not adequately describe the details of concurrent use of
multiple servers to ensure interoperable DHCPv6 clients that will provide
predictable, manageable, useful behavior.

Potentially open issues in the use of multiple servers:

* Text that implies interaction with a single server through use of
singular reference to "server" and through omission of any specific
reference to multiple servers
* Semantics of Solicit message in the presence of multiple DHCP services:
how does a client know when it should interact with multiple servers; how
does a client differentiate between redundant servers (from the same
service) and servers with different resources?
* Text in section 5.3 describing use of server preference field that
implies use of single server with highest preference

- Ralph



From dhcp-v6@bucknell.edu  Mon Nov 29 20:58:58 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29339
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 20:58:56 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA32318;
	Mon, 29 Nov 1999 20:55:36 -0500 (EST)
Date: Mon, 29 Nov 1999 20:55:36 -0500 (EST)
Message-Id: <Version.32.19991129142841.00e53a90@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Based on re-reading this discussion thread, and thinking about two
scenarios in more detail, here are some additional thoughts on how to
manage and release "releasable resources".

The two scenarios are based on whether a client "MUST", "SHOULD" or "MAY"
release releasable resources back to the server from which the the resource
was originally obtained.  If the spec says "MUST", then the client may be
unduly constrained from returning the resource to some other server that
might be able to process the release.  If the spec says "SHOULD" or "MAY",
then there is the risk of releasing the resource to a server that cannot
process the release.  

I think the problem arises because there may be more than one server (but
not all servers) that can process a particular resource.  That is, there
may be some redundant or backup server that can process a particular
resource, but there may be some other servers that cannot process the
resource.  In this situation, choosing a backup server at random may not
work; the client may choose the wrong server.

Rather than associating a resource with a particular *server*, it might be
possible to associate a resource with a DHCP *service* (which may be
provided by more than one server).  If servers and resources were
identified by service, a client could choose among multiple, equivalent
servers within a service.  In this way, a client could take advantage of
redundant *servers* within a *service*, and avoid the risk of returning the
resource to some other server that cannot process the resource.

- Ralph




From dhcp-v6@bucknell.edu  Mon Nov 29 20:59:01 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29351
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 20:58:57 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA30253;
	Mon, 29 Nov 1999 20:55:58 -0500 (EST)
Date: Mon, 29 Nov 1999 20:55:58 -0500 (EST)
Message-Id: <Version.32.19991129145624.00e56b90@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 10:56 AM 11/19/99 -0500, you wrote:
>>                                                    But because there
>> is no text calling for clients to actually do this (and the current
>> text certainly implies that one get all resources from one server), I
>> believe it is fair to say that DHCP as currently written does not
>> support getting resources from multiple servers in the case Tim
>> mentions.

I disagree with Thomas, here; perhaps we disagree on the meaning of the
word "support".  The protocol, as documented in the current draft, *can* be
used to obtain resources from multiple servers, although much of the text
seems (to my reading) to imply otherwise.  There may be some technical
issues about server selection in section 5.3 (see my previous note); it may
be that we need to clarify that a client needs to send multiple Solicit
messages, one to each potential DHCP service, to find the highest
preference server from each DHCP service.  

>By now, I hope my reasons for disagreeing totally with this paragraph
>are very clear.  It is not fair to say that our specification does
>not support Cisco routers just because we don't say how to use them.
>It is not fair to say that our spec does not support mobile computers
>just because we don't say how to use them.  I think your contention here
>is, I wish I could say it more tactfully but I don't know how, totally
>wrong.

I disagree with Charlie, here.  The examples he cites are irrelevant to
this situation.  The ability to use DHCPv6, as documented in the current
draft, with multiple servers does *not* imply that it will be implemented
in a useful, interoperable and useful way.

Let me take this a step further and say that using the dictum of "let the
market decide" has *not* worked for DHCPv4 *clients*.  There is one
predominant client, which has some well-documented interesting behaviors
that have been fixed in later releases.  The cost of deployment has
precluded the deployment of new releases of that client in many cases.  The
cost of deploying a different vendor's client would be prohibitive and
stand in the way of any market selection mechanisms.

We must be very careful to get the client specification right from the
beginning...

- Ralph




From dhcp-v6@bucknell.edu  Mon Nov 29 20:59:32 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29550
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 20:59:30 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA11290;
	Mon, 29 Nov 1999 20:55:47 -0500 (EST)
Date: Mon, 29 Nov 1999 20:55:47 -0500 (EST)
Message-Id: <Version.32.19991129143932.00e57390@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 10:56 AM 11/19/99 -0500, you wrote:
>> Where in the current spec does it call out for a client to ask for
>> addresses/parameters from more than one dhcp server?  My read of the
>> spec is that it says you pick one server and get all your resources
>> from it.
>
>Then you are reading something into the spec that isn't anywhere in
>there.  The job of a specification is to make protocol operations
>possible,
>not to tell clients when to use the protocol.  The specification also
>does not say what kind of routers to use, nor how many bits per second
>of bandwidth is required, nor whether mobile computers can use the
>messages, nor whether a collision-detection scheme is needed at
>layer-2, nor whether the server must store its data using LDAP, nor
>any of a thousand other details.  All of these thousand other details
>are ABSOLUTELY CRUCIAL for lots of implementations, and yet the
>specification does not discuss them.

While the spec does not specifically require the use of one server at a
time, much of the language seems to me to imply the use of a single server.
 And, as far as I can recall, this is the first time we've formally
discussed the idea of multiple simultaneous transactions with different
servers.  I think the claim that the spec currently allows the client to
ask for parameters from more than one server is not entirely supported by
the text in draft-ietf-dhc-dhcpv6-14.txt; e.g.:

3.3. Request/Response Processing Model

   The request/response processing for DHCPv6 is transaction based and
   uses a set of best-effort messages to complete the transaction.

   To find a server, a client sends a DHCP Solicit from the interface

   which it wishes to configure.  The client then awaits a DHCP
   Advertise message containing an IP address of a DHCP server.
   Transactions are started by a client with a DHCP Request, which may
   be issued after the client knows the server's address.  The server
   then unicasts a DHCP Reply, possibly via a relay.  At this point in
   the flow all data has been transmitted and is presumed to have been
   received.

4.3. DHCP Request Message Format

   In order to request configuration parameters from a server, a client
   sends a DHCP Request message, and MAY append extensions [15].  If
   the client does not know any server address, it MUST first obtain
   one by multicasting a DHCP Solicit message (see Section 4.1).

5.3. Receiving DHCP Advertise Messages

   After a client has received a DHCP Advertise message, it has the
   address of a server for subsequent DHCP Request messages.

   [...]

   Unless a DHCP Advertisement is received with a preference equal
   to 255 (see Section 4.2), the client MUST wait CLIENT_ADV_WAIT
   seconds after issuing the DHCP Solicit message in order to receive
   the Advertisement with the highest preference.  After waiting for
   that period of time, a client MUST select the highest preference
   server as the target of its DHCP request.  If a client receives an
   advertisement with a preference of 255, it does not have to wait for
   any more advertisements.

   If a client sends a DHCP Request to a highly preferred server but
   fails to receive a DHCP reply from that server after following the
   retransmission algorithm in section 8, the client MUST then try to
   send a DHCP Request to a less preferred server.

(I think this section, at least, needs to be rewritten to allow interaction
with multiple servers.  Under what conditions may a client disregard the
server preference and begin a transaction with a second server with a
[presumably] lower preference?)





From dhcp-v6@bucknell.edu  Mon Nov 29 20:59:34 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29577
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 20:59:34 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA24477;
	Mon, 29 Nov 1999 20:54:55 -0500 (EST)
Date: Mon, 29 Nov 1999 20:54:55 -0500 (EST)
Message-Id: <Version.32.19991129142513.00e54e00@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 10:56 AM 11/19/99 -0500, Charles Perkins wrote:
>We implemented Mobile IP with DHCPv4 more than once.  The problem we
>had was with the state machine implementation of DHCPv4, and the
>drastic inconvenience or impossibility of starting multiple state
>machines.

>That problem has been avoided in DHCPv6, to the best of my current
>understanding.

Charlie, to improve my understanding of DHCPv4 and to be able to tell
students about when state machines may not be an appropriate specficiation
or implementation tool, can you give me an example of how the DHCPv4 state
machine was a problem in your implementation of DHCPv4.  How would the
implementation have been made easier without the state machine, and how can
the behavior of the DHCPv4 client be specified clearly and concisely
without a state machine?

- Ralph




From dhcp-v4@bucknell.edu  Mon Nov 29 21:31:16 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA12407
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 21:31:12 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA11807;
	Mon, 29 Nov 1999 21:30:44 -0500 (EST)
Date: Mon, 29 Nov 1999 21:30:44 -0500 (EST)
Message-Id: <199911300212.SAA08026@scv1.apple.com>
Errors-To: droms@bucknell.edu
Reply-To: cheshire@apple.com
Originator: dhcp-v4@bucknell.edu
Sender: dhcp-v4@bucknell.edu
Precedence: bulk
From: Stuart Cheshire <cheshire@apple.com>
To: Multiple recipients of list <dhcp-v4@bucknell.edu>
Subject: Re: DHCP server/Client behavior when options change while lease is valid
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion of DHCP for IPv4
Content-Type: text/plain; charset="US-ASCII"
Mime-Version: 1.0
X-Mailer: Claris Emailer 2.0v3, January 22, 1998

>Should the client check the set of options received
>each time a lease is renewed and take appropriate action
>if the some of the options have changed?

I believe that if the server administrator changes some parameter like 
the DNS server address, then it would be desirable for all client hosts, 
eventually, to start using that new DNS server. The change doesn't need 
to happen immediately, but it would be good to know that it will happen 
eventually, in some bounded about of time. If the DHCP address renewal 
time is set to 24 hours, then it would be good to know that any changed 
options will also take effect within at most 24 hours.

Given that it might be quite reasonable for DHCP client hosts to run for 
months at a time without reboots, renewing their address leases hundreds 
or thousands of times, this implies that they need to be checking the 
options returned in their renewal packets to see if anything has changed.

The Mac OS DHCP client currently does not do this. It is one of the 
things on our list of work to do.

Stuart Cheshire <cheshire@apple.com>
 * <A HREF="http://ResComp.Stanford.EDU/~cheshire/">Web Page</A>
 * Wizard Without Portfolio, Apple Computer



From dhcp-v6@bucknell.edu  Mon Nov 29 22:43:49 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA17044
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 22:43:48 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id WAA19351;
	Mon, 29 Nov 1999 22:41:36 -0500 (EST)
Date: Mon, 29 Nov 1999 22:41:36 -0500 (EST)
Message-Id: <199911300315.WAA0000011634@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Hi Francis,

>=> I believe this problem can be solved if ways to use DHCPv6 are described
>in a separate document:
> - we keep mechanisms in current I-Ds which should become PS ASAP
> - we explain how to implement/use DHCPv6 in an application document
>   which doesn't need to become a PS today
>We have this kind of things for routing protocols and I think this is a
>good way to split theory and pratice.
>Another point, I believe we have to update/improve autoconfig specs
>because interactions between stateless (which is described in these specs)
>and statefull (aka DHCPv6) autoconfigs were/are not fully understood.
>An application document can help us to see what are the problems and
>will be easier for implementor feedback.
>
>Is this a good solution for you (Thomas and I-D authors)?

This sounds good to me but do want to hear the input on the issues Mike
sent out, or ones like yours.  If it is really broken or NOT CLEAR we
should fix it.

I don't think we should mess with stateless autoconfig because of DHCPv6
until we have seen there is a real problem from implementation and
interoperability testing DHCPv6 in the presence of stateless addrconf.
But I don't think we can get there till DHCPv6 is PS.

/jim
Regards

Francis.Dupont@inria.fr


From dhcp-v6@bucknell.edu  Mon Nov 29 23:22:18 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA02436
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 23:22:17 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA08380;
	Mon, 29 Nov 1999 23:20:42 -0500 (EST)
Date: Mon, 29 Nov 1999 23:20:42 -0500 (EST)
Message-Id: <Version.32.19991129211408.00e5b6a0@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 02:43 PM 11/29/99 -0500, you wrote:
>> (From Thomas Narten)
>> 
>> Let me clarify. I agree that releasing resources via the server one
>> got the resources from makes sense.  Having that be the default seems
>> fine.  However, the current draft goes further and essentially says:
>> you can't release via another server, period. The middle ground that
>> has been suggested is to first try releasing via the server one got
>> the resource from in the first place.  If that fails, however, then
>> fall back by trying other servers. Having a fall back might make
>> things work in the case that another server was available. This could
>> happen via the server-server protocol, but could also happen if a DHCP
>> server was retired and replaced by another with configuation
>> information migrated from one server to the other manually.  Thus,
>> there is no requirement that a server-server protocol be in place, but
>> we'd be leaving in hooks to make things work better in such an
>> environment.
>> 
>> [Barr Hibbs]
>> 
>> Thomas, I disagree:  if a server is being retired gracefully by an
>> administrator, that administrator probably will not really care if every
>> last resource is released because the retirement process will most likely
>> first entail stopping the soon-to-be-retired server from allocating any
>> resources, then simply waiting the anticipated lifetime of all assigned
>> resources, leaving any that fall out to simply be handled as exceptions.

Perhaps part of our problem is in not having a clear definition of the
semantics of a releasable resource.  In particular, what happens when a
releasable resource is not correctly released?  In the case of an IPv6
address (the only concrete example we have of a releasable resource), I
believe we can tolerate the failure of a release.  That is, if a release
operation fails for any reason, the lease on the address will eventually
expire and the address will become unassigned.

This model seems to work OK for DHCPv4.  Administrators have the freedom to
choose lease lengths appropriate to the IPv4 address usage patterns, so
that lease expiration 

However, because we're allowing for other (as yet undefined) releasable
resources, it may be the case that these other releasable resources may
require explicit, reliable releases; simply waiting for a lease to expire
isn't sufficient.

- Ralph




From dhcp-v6@bucknell.edu  Mon Nov 29 23:22:44 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA02663
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 23:22:44 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA32355;
	Mon, 29 Nov 1999 23:20:59 -0500 (EST)
Date: Mon, 29 Nov 1999 23:20:59 -0500 (EST)
Message-Id: <Version.32.19991129213821.00df2ba0@mail.bucknell.edu>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 02:43 PM 11/29/99 -0500, HIBBS, BARR (PB) wrote:
>> I agree that we probably don't want to say that the client MUST release
>> the resources to the original server.  That precludes the possibility of
>> designing server-to-server and server redundancy protocols.  All I am
>> saying is that in order to get things to work correctly in the more common
>> case we need to say that the client SHOULD release resources to the
>> original server and if that fails then they MAY release them to some other
>> known server with the understanding that absent some adminstrative co-
>> ordination, the description of which is out of scope for the draft, the
>> release may not work.  That's it.  Nothing more complicated than that.
>> 
>[Barr Hibbs]
>
>How about something like:  "Clients must release resources to the
>[administrative] domain which assigned them."
>
>-- obvious problems are how to identify domains, and
>-- what is a domain has multiple servers and no interserver protocol?
> 

Barr - sounds like we developed similar ideas in parallel.  In a note I
posted earlier today, I suggested associating an address with a "DHCP
service", which I read to be the same idea as your "administrative domain"...

- Ralph




From dhcp-v6@bucknell.edu  Mon Nov 29 23:22:49 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA02718
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 23:22:49 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA00823;
	Mon, 29 Nov 1999 23:21:23 -0500 (EST)
Date: Mon, 29 Nov 1999 23:21:23 -0500 (EST)
Message-Id: <199911300359.WAA00559@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> We must be very careful to get the client specification right from the
> beginning...

Hear, hear.   The number of broken DHCP clients for which there are
subtle workarounds in the ISC DHCP server is kind of depressing.
Vendors who aren't particularly interested in DHCPv6 can't be expected
to think these issues through to the level of detail that we would
think them through.   That's why they pay us the big bucks, if you
will pardon my saying so.

I _still_ hear RFC1541 referred to as gospel WRT DHCPv4, and encounter
new clients implemented to it, even though it is actually not
interoperable with the current RFC.

			       _MelloN_


From dhcp-v6@bucknell.edu  Mon Nov 29 23:23:07 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA02865
	for <dhc-archive@odin.ietf.org>; Mon, 29 Nov 1999 23:23:06 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id XAA18995;
	Mon, 29 Nov 1999 23:21:12 -0500 (EST)
Date: Mon, 29 Nov 1999 23:21:12 -0500 (EST)
Message-Id: <199911300354.WAA00544@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> >We implemented Mobile IP with DHCPv4 more than once.  The problem we
> >had was with the state machine implementation of DHCPv4, and the
> >drastic inconvenience or impossibility of starting multiple state
> >machines.

I didn't see this go by the first time, but it _is_ a rather
surprising statement to me, because the ISC DHCP client in fact
does implement arbitrary numbers of state machines!

			       _MelloN_



From dhcp-v6@bucknell.edu  Tue Nov 30 00:31:01 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id AAA03426
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 00:30:59 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id AAA07590;
	Tue, 30 Nov 1999 00:28:13 -0500 (EST)
Date: Tue, 30 Nov 1999 00:28:13 -0500 (EST)
Message-Id: <199911300507.AAA0000027918@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

hi barr,

>> ...still catching up!

we appreciate that alot...

>> Your read of the spec is an interpretation of it what it can do  but not
>> what it can't do.  There is nothing in the protocol that prohibits
>> multiple servers. How the client does this is the implementors business.
>> If folks want an appendix to describe an implementation scenario where:
>> 
>>   1. a new interface comes up on the same link (same sever though)
>>      this is easy.
>>   2. a new interface comes up on a link that did not exist and 
>>      may be a new server after the solicit goes out.
>> 
>> [Barr Hibbs]
>> 
> ...mmmmm....  shouldn't a solicit always be sent in this case?  at least,
> if it is necessary to make the protocol work gracefully....

yes my sentence is to terse.  the solicit would learn if this is a new
sever is what I should have said in #2.

/jim


From dhcp-v6@bucknell.edu  Tue Nov 30 00:31:40 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id AAA03804
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 00:31:39 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id AAA22135;
	Tue, 30 Nov 1999 00:29:50 -0500 (EST)
Date: Tue, 30 Nov 1999 00:29:50 -0500 (EST)
Message-Id: <199911300509.AAA0000029225@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>..if I may generalize....
>
>The Europeans and Asians desperately want an entire suite of protocols for
>IPv6 -- this discussion is ultimately productive if it results in a more
>robust protocol, but if we are simply bogged down in the minutiae of a
>theoretical discussion of protocol specifications, then we waste everyone's
>time. 
>
>Seems to me that we have the basics already:
>
>1. multiple servers in multiple domains are both possible and likely
>
>2. absent interserver AND interdomain protocols we must address releasable
>resources in the DHCPv6 draft
>
>3. while it places a small burden on a client, requiring resources to be
>released to the assigning entity is simply the best, most workable starting
>point -- we are now including versioning information which permits us the
>means to modify the protocol in the future if implementations can't make
>this work

I agree and a good point.  With the version # we are taking a lesson
from DHCPv4 for sure.

/jim





From dhcp-v6@bucknell.edu  Tue Nov 30 02:15:12 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id CAA03450
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 02:15:12 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id CAA00815;
	Tue, 30 Nov 1999 02:12:47 -0500 (EST)
Date: Tue, 30 Nov 1999 02:12:47 -0500 (EST)
Message-Id: <199911300653.BAA0000017592@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DISCUSSION: Use of Address Extensions with DHCPv6 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Hi Mike,

>Seems to me (after no small amount of consideration) that a server needs to
>provide
>more information in its advertisements then we have been proposing (note that
>providing
>this information doesn't change the protocol as defined at all). Consider my
>summary of
>question #1 - I've used it in responding to this message.

OK.  I think I agree with you on that count.

>Jim Bound wrote:
>
>> folks,
>>
>> I think we need to discuss some details here.  what do folks think about
>> the attached:
>>
>> ONLY DISCUSSING ADDRESS EXTENSIONS other parameters and what clients or
>> servers should do is being discussed globally on another mail thread.
>>
>> Semantics of messages:
>>
>> 1.  Client sends DHCPv6 Request with no address extension.
>>
>> 1a. Server responds with all addresses available to the client
>>     if the sys admin has defined the server can do this.
>>
>> 1b. If the server gets a second message from the client with no
>>     address extension it resends the client addresses it gave
>>     them previously cumultatively (see below).
>>
>> This would be the default behavior for DHCPv6 and how clients request
>> addresses.  The reason there is a sys admin decision is that the network
>> admin folks may have decided to use stateless addrconf and dhcpv6 is
>> only being used to provide non-address parameters to the clients or to
>> update DNS, etc..

>#1 goes against the constraint I listed in the Summary: DC meeting question #1
>message: Clients must explicitly request releaseable resources. If the server
>had explicit (through policy) or implicit (dynamic addresses) to offer to the
>client,

>it would do so by adding IP address extensions carrying these addresses to the
>advertisement. The client would request one or all of these IP addreses in its
>request. If the site is using stateless autoconf, then the server would not
>have any
>IP address extensions in its advertisement, and thus the client could not
>request
>an IP address.

that would work yes.  same affect.  I can live with this if others can?

>> 2.  Client sends Request with 1 or more "empty" address extensions
>>
>> 2a. Server responds, if the client is permitted a new address, with a new
>>     IPv6 address in addition to the ones given to the client previously.
>>     If the client is not permitted a new address then an error is
>>     returned to the client.
>>     The previous addresses are not resent to the client in the
>>     Server Reply.
>>
>> 2b. The client MUST NOT send this through a RELAY-AGENT and the
>>     client is assumed to have an address with enough scope to
>>     reach the server directly.  This permits the use of DHCPv6
>>     authentication or IPsec to be used to access the server.
>>     [In fact it probably should be a MUST have security for
>>      this type of transaction.]

>I'm assuming that this request is in addition to the IP address(es) offered in
>the advertisement. The client will have knowledge of what networks are
>available
>from the advertisement, and thus would need to tell the server which network it

>would like an IP address on. Thus, an empty IP address extension isn't enough -
>the
>client would need to communicate which network. At DC, I proposed a change to
>the
>IP address extension which would permit the client to identify which network it
>wants
>an additional IP address on. Whether or not it would receive an address is up
>to
>administrator policy. Note that a server could use this modification for
>advertising
>network prefixes, without advertising a specific IP address on that network.

Again this is an optimization I can live with.

But for this case a client would have to provide an extension for each
address it wanted.

Do you agree with this semantic for the protocol?

And with the behavior?  This is one of Thomas's issues as a note and I
would like to resolve that meta issue very clearly this time?

>> This would be a configured behavior permitted by sys admins.
>> To support this behavior we must require that the transaction-id be part
>> of the clients binding for the server/client database kept by the
>> server, when the client uses this feature.  The binding would be as follows:
>
>I don't see that the transaction id needs to be part of the binding. The XID is
>cached
>to enable a server to distinguish retransmissions from new requests. The
>link-local
>address, agent prefix, and IP address should be enough. Prehaps I'm missing
>something.

will respond further down.

>>
>>     client-link-local-address
>>     dhcpv6-agent-address-prefix
>>     ipv6-address given to client
>>     valid and preferred lifetimes
>>     transaction-id from the request by the client
>>
>> * we can add in the appendix that a server can structure the database
>>   so that the main key is link-local + agent-prefix and sub keys are
>>   the remaining parts.  We should also point out that when dhcpv6
>>   operates in this mode there is more cost for memory and for
>>   performance as the sever must now locate previous transactions
>>   with this behavior or use a binding-cache (see discussion below).
>
>Do we really want/need to specify this?

We should not have to I was being nice.  

>>
>>
>> 3.  Client sends Request with address extension or extensions
>>     which are not empty.
>>
>> 3a. Server adds these addresses to the sever database for this
>>     client.  Server sends Reply back to the client if it is
>>     permitted to do this with the addresses the client sent
>>     back to the client in the Reply.
>>
>> 3b. The client MUST NOT send this through a RELAY-AGENT and the
>>     client is assumed to have an address with enough scope to
>>     reach the server directly.  This permits the use of DHCPv6
>>     authentication or IPsec to be used to access the server.
>>     [In fact it probably should be a MUST have security for
>>      this type of transaction.]
>>
>> We really need to discuss this scenario.  Do we want to permit this type
>> of behavior in DHCPv6.  This breaks the model of DHCPv4 yet again and is
>> really new stuff???????????  But we could make it work.
>>
>> What is the affect of #3 above to any future server to server protocol,
>> could clients "updating" dhcpv6 server databases throw a synched server
>> to server protocol database out of synch?
>
>Jim, I don't think this is a good scenario. A Client should never be involved
>in
>telling a server about IP addresses (or other releasable resources) it may have
>
>that the server has no knowledge of. Sure, a client MAY request that a certain
>address be assigned to it, and if the request is valid from a network topology
>point of view and a administrator policy view, then fine. I just don't think we
>
>should be configuring DHCP servers with DHCP clients, IMHO.

Same difference.  If the address is not topologically correct then the
server would not permit it?

>
>
>> BINDING-CACHE:  We would move the binding cache discussion to the
>> appendix as the above I think fixes idempotent problems in the protocol
>> standard.  The binding-cache would also lesson the cost/performance hit
>> I speak of above in implementations.  The server would only need to
>> store the Reply that has to be resent and an not the entire client
>> database.
>
>Binding cache is very useful for preventing unnecessary database lookups. Ah,
>now I see where you're coming from wrt saving the transaction ID - you're
>attempting to record the assignment transaction to avoid confusion between
>retransmissions and new requests.

No if the XID-TIMEOUT happens and the retransmission gets pass the
binding cache is the problem.  

>Is it really necessary to be this heavy-weight? Sure, as specified, there is
>a window where the server's XID timeout cache may timeout cache entries
>while a client is retransmitting (server reboots, loses cache information), but
>
>that window is very small. In most cases, the XID cache timeout will be large
>enough and the server stable enough to prevent the accidental assignment of
>additional IP address(es) (or other releasable resources). Alternately, a
>server could write its XID cache to stable storage from time to time, and
>reload it upon restart, thus satisfying the idempotent problem w/o having the
>cost of the extra space in the database and extra initial database lookup.

but it could happen Mike.  if it can happen don't we have to cover it?

Whats the worst thing that could happen.  What I come up with is:

We don't give the client another address because we think we already
gave them their last one because this retransmission is treated as a new
request (got past the binding cache) and when we look at the database we
the server believe we have given them all the addresses they can get.
The client would get an error returned.

Seems that putting the transaction id in the sever database is a small
price to pay to respond to clients when the XID_TIMEOUT was not big
enough?

/jim 


From dhcp-v6@bucknell.edu  Tue Nov 30 10:04:33 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA10331
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 10:04:31 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA18444;
	Tue, 30 Nov 1999 10:01:58 -0500 (EST)
Date: Tue, 30 Nov 1999 10:01:58 -0500 (EST)
Message-Id: <199911301442.JAA0000017575@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>>You need to help us here and respond to this abstract issue as it can
>>reduce the time to get this spec in shape and fix the real protocol bugs
>>we are finding from this discussion.
>
>Put succinctly, I know of no abstract reason to preclude the concurrent
>interoperation of a DHCP client with multiple servers.  In my opinion, the
>current spec does not adequately describe the details of concurrent use of
>multiple servers to ensure interoperable DHCPv6 clients that will provide
>predictable, manageable, useful behavior.

>Potentially open issues in the use of multiple servers:

>* Text that implies interaction with a single server through use of
>singular reference to "server" and through omission of any specific
>reference to multiple servers

This is pretty clear and pretty easy to fix IMO.

>* Semantics of Solicit message in the presence of multiple DHCP services:
>how does a client know when it should interact with multiple servers; how
>does a client differentiate between redundant servers (from the same
>service) and servers with different resources?

Just sent suggestion to fix that which will support getting this done
rather expediently in the spec.  

I am hearing from the working group very loud we need to make choices
that can move this along expediently.

>* Text in section 5.3 describing use of server preference field that
>implies use of single server with highest preference

Agreed.

/jim


From dhcp-v6@bucknell.edu  Tue Nov 30 10:05:43 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA10998
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 10:05:42 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA11787;
	Tue, 30 Nov 1999 10:02:08 -0500 (EST)
Date: Tue, 30 Nov 1999 10:02:08 -0500 (EST)
Message-Id: <199911301448.JAA0000019335@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Hi Ralph,

>>>                                                    But because there
>>> is no text calling for clients to actually do this (and the current
>>> text certainly implies that one get all resources from one server), I
>>> believe it is fair to say that DHCP as currently written does not
>>> support getting resources from multiple servers in the case Tim
>>> mentions.
>
>I disagree with Thomas, here; perhaps we disagree on the meaning of the
>word "support".  The protocol, as documented in the current draft, *can* be
>used to obtain resources from multiple servers, although much of the text
>seems (to my reading) to imply otherwise.  There may be some technical
>issues about server selection in section 5.3 (see my previous note); it may
>be that we need to clarify that a client needs to send multiple Solicit
>messages, one to each potential DHCP service, to find the highest
>preference server from each DHCP service.  

The solicit is a multicast sendmsg() for a link (subnet) then the client
will hear from multiple servers, which as you pointed out in other mail
we need to clarify for the multiple server case.

Right now I don't think it wise to have different multicast solicit
address types on a link, until we get a lot more knowledge about IPv6
let alone DHCPv6?  I would like to see it work in practice a bit before
going that route.  Also this would presume the client has some knowledge
to differentiate solicit types at boot time and is that a good idea?

thanks
/jim
>By now, I hope my reasons for disagreeing totally with this paragraph
>are very clear.  It is not fair to say that our specification does
>not support Cisco routers just because we don't say how to use them.
>It is not fair to say that our spec does not support mobile computers
>just because we don't say how to use them.  I think your contention here
>is, I wish I could say it more tactfully but I don't know how, totally
>wrong.

I disagree with Charlie, here.  The examples he cites are irrelevant to
this situation.  The ability to use DHCPv6, as documented in the current
draft, with multiple servers does *not* imply that it will be implemented
in a useful, interoperable and useful way.

Let me take this a step further and say that using the dictum of "let the
market decide" has *not* worked for DHCPv4 *clients*.  There is one
predominant client, which has some well-documented interesting behaviors
that have been fixed in later releases.  The cost of deployment has
precluded the deployment of new releases of that client in many cases.  The
cost of deploying a different vendor's client would be prohibitive and
stand in the way of any market selection mechanisms.

We must be very careful to get the client specification right from the
beginning...

- Ralph




From dhcp-v6@bucknell.edu  Tue Nov 30 10:07:01 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA11747
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 10:07:00 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA08207;
	Tue, 30 Nov 1999 10:01:48 -0500 (EST)
Date: Tue, 30 Nov 1999 10:01:48 -0500 (EST)
Message-Id: <199911301433.JAA0000015539@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


>Rather than associating a resource with a particular *server*, it might be
>possible to associate a resource with a DHCP *service* (which may be
>provided by more than one server).  If servers and resources were
>identified by service, a client could choose among multiple, equivalent
>servers within a service.  In this way, a client could take advantage of
>redundant *servers* within a *service*, and avoid the risk of returning the
>resource to some other server that cannot process the resource.

Mike has proposed to add an address prefix of topological significance
to the address extension.  The only realeasable resource we have right
now.  For other reasons but close to the same reason.
This could be added to also support knowing if a server is within the
administrative domain as Barr suggested.

The way this would work is on the Advertisement we would add a field for
the server to put its prefix topological significance.  When the client
goes to release a resource to a server the address would be checked
against that stored value for servers and provide the client knowledge
if in fact that server can release the address.

This is more doable than using a service bit mask for now and if we in
the future define other than addresses we can use the server prefix as
part of that releasable resources associations.  

Also if this in the future needs to be expanded we can do so with a new
version of the protocol.

I think this solves the problem in an expedient manner for the spec and
keeps us on track but also addresses Ralph's suggestion I think?

Comments please?

thanks
/jim


From dhcp-v6@bucknell.edu  Tue Nov 30 10:18:24 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id KAA10330
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 10:04:31 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id KAA28306;
	Tue, 30 Nov 1999 10:01:02 -0500 (EST)
Date: Tue, 30 Nov 1999 10:01:02 -0500 (EST)
Message-Id: <199911301425.JAA0000014792@quarry.zk3.dec.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Jim Bound <bound@zk3.dec.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Hi Ralph,

>>> Where in the current spec does it call out for a client to ask for
>>> addresses/parameters from more than one dhcp server?  My read of the
>>> spec is that it says you pick one server and get all your resources
>>> from it.
>>
>>Then you are reading something into the spec that isn't anywhere in
>>there.  The job of a specification is to make protocol operations
>>possible,
>>not to tell clients when to use the protocol.  The specification also
>>does not say what kind of routers to use, nor how many bits per second
>>of bandwidth is required, nor whether mobile computers can use the
>>messages, nor whether a collision-detection scheme is needed at
>>layer-2, nor whether the server must store its data using LDAP, nor
>>any of a thousand other details.  All of these thousand other details
>>are ABSOLUTELY CRUCIAL for lots of implementations, and yet the
>>specification does not discuss them.

>While the spec does not specifically require the use of one server at a
>time, much of the language seems to me to imply the use of a single server.
> And, as far as I can recall, this is the first time we've formally
>discussed the idea of multiple simultaneous transactions with different
>servers.  I think the claim that the spec currently allows the client to
>ask for parameters from more than one server is not entirely supported by
>the text in draft-ietf-dhc-dhcpv6-14.txt; e.g.:

I think the issue here is it would make folks more comfortable if we had
some sections that actually mention "multiple servers" in some sense
where it might be appropriate?

But as I read the text below it is idempotent to the multiple server
question, except when we "say pick a server"?

>3.3. Request/Response Processing Model
>
>   The request/response processing for DHCPv6 is transaction based and
>   uses a set of best-effort messages to complete the transaction.

>   To find a server, a client sends a DHCP Solicit from the interface
>
>   which it wishes to configure.  The client then awaits a DHCP
>   Advertise message containing an IP address of a DHCP server.
>   Transactions are started by a client with a DHCP Request, which may
>   be issued after the client knows the server's address.  The server
>   then unicasts a DHCP Reply, possibly via a relay.  At this point in
>   the flow all data has been transmitted and is presumed to have been
>   received.

This is idempotent and I can see no where we break the multiple server
model.  The protocol keeps on ticking whether there is 1...N links and
1...N servers.  

4.3. DHCP Request Message Format

   In order to request configuration parameters from a server, a client
   sends a DHCP Request message, and MAY append extensions [15].  If
   the client does not know any server address, it MUST first obtain
   one by multicasting a DHCP Solicit message (see Section 4.1).

Here is where we might add that this solicit in fact will support
multiple instants of solicit on different links (subnets in IPv4 terms)
or in section 4.1.  

5.3. Receiving DHCP Advertise Messages

   After a client has received a DHCP Advertise message, it has the
   address of a server for subsequent DHCP Request messages.

   [...]

   Unless a DHCP Advertisement is received with a preference equal
   to 255 (see Section 4.2), the client MUST wait CLIENT_ADV_WAIT
   seconds after issuing the DHCP Solicit message in order to receive
   the Advertisement with the highest preference.  After waiting for
   that period of time, a client MUST select the highest preference
   server as the target of its DHCP request.  If a client receives an
   advertisement with a preference of 255, it does not have to wait for
   any more advertisements.

Here what we are saying was an addition from the Chicago meeting.

We assume the client on the same link could hear from multiple servers
as we were told by the working group in Chicago.  Per the working group
we added the preference field to the Advertisement message.  The way the
code and implementation could work is as follows:

  boolean head_cheese_server = FALSE;
  int *head_cheese_server_data
  pthread_start_client_solicit;  /* this is not how it looks but */
  time_interrupt_set (CLIENT_ADV_WAIT);
  while (pthread_CLIENT_ADV_WAIT_not_back) {
    get_advertisement_message(server_msg_block);
    If (server_msg_block->preference_value = 255) {
       head_cheese_server = TRUE;
       head_cheese_server_data = server_msg_block;
       break;
    } else 
        store_this_server_for_later_sort (server_msg_block);
  }
  /* not putting in check here for discussion if no advertisements
     came back */
  If (head_cheese_server)
   use_the_head_cheese_server_for_request(server_msg_block);
  else
   sort_and_pick_highest_preference_server(server_msg_block);

But the point is that we listen to multiple servers and select among
them based on the preference field from the server unless a server says
I am the boss or head cheese use me.  The first server to be heard that
does this wins.

The client may store all server advs for later use too.

   If a client sends a DHCP Request to a highly preferred server but
   fails to receive a DHCP reply from that server after following the
   retransmission algorithm in section 8, the client MUST then try to
   send a DHCP Request to a less preferred server.

>(I think this section, at least, needs to be rewritten to allow interaction
>with multiple servers.  Under what conditions may a client disregard the
>server preference and begin a transaction with a second server with a
>[presumably] lower preference?)

So do you believe we should say the client SHOULD store servers that
respond to the solicit and then discuss the case when the client will
use a server of lower preference?

This I believe resolves your concern and is doable?

thanks
/jim



From dhcp-v6@bucknell.edu  Tue Nov 30 12:28:36 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA24310
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 12:28:35 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id MAA12204;
	Tue, 30 Nov 1999 12:23:37 -0500 (EST)
Date: Tue, 30 Nov 1999 12:23:37 -0500 (EST)
Message-Id: <199911301647.LAA02252@rotala.raleigh.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Ralph Droms <droms@bucknell.edu> writes:

> At 10:56 AM 11/19/99 -0500, you wrote:
> >>                                                    But because there
> >> is no text calling for clients to actually do this (and the current
> >> text certainly implies that one get all resources from one server), I
> >> believe it is fair to say that DHCP as currently written does not
> >> support getting resources from multiple servers in the case Tim
> >> mentions.

> I disagree with Thomas, here; perhaps we disagree on the meaning of the
> word "support".  The protocol, as documented in the current draft, *can* be
> used to obtain resources from multiple servers, although much of the text
> seems (to my reading) to imply otherwise.  There may be some technical
> issues about server selection in section 5.3 (see my previous note); it may
> be that we need to clarify that a client needs to send multiple Solicit
> messages, one to each potential DHCP service, to find the highest
> preference server from each DHCP service.

I think we are actually in agreement. My use of "does not support"
appears to have been confusing and misleading. The underlying message
formats certainly could support a client having multiple
interactactions with different servers simultaneously.  However, the
spec itself has little text on this and my expectation of what folks
would actually implement (based on my read of what the current text
says should be implemented) is that clients will in practice pick one
server and interact only with it.

Thomas


From dhcp-v6@bucknell.edu  Tue Nov 30 13:41:57 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA01572
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 13:41:56 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA14966;
	Tue, 30 Nov 1999 13:39:22 -0500 (EST)
Date: Tue, 30 Nov 1999 13:39:22 -0500 (EST)
Message-Id: <199911301824.NAA02527@rotala.raleigh.ibm.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Thomas Narten <narten@raleigh.ibm.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: Summary: DC meeting question #1 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.

Mike,

I've got some clarification questions.

> Semantics of releasable resources:

> 1) General Form

>     *	Releasable resource extensions need to have a general form so that
> 	clients/servers can manage them in a generic fashion.

>     *	There is a generic bit in the extensions header which is set if the
> 	extension is carrying a releasable resource. When this bit is set,
> 	there are two 32 bit fields present in the extension which represent
> 	the relative times in seconds when the client should start attempting
> 	to extend the lease on the resource, and when the client must stop
> 	using the resource if it is unable to extend it.

Clarification: are you saying the list discussion is arguing for such
a bit? Or that one will go into the draft? Or? (I don't recall the
current draft having such a bit.)

>     *	Since extensions carrying releasable resources have a general
> 	form/semantics, clients/servers do not need to change when new
> 	extensions are defined which are releasable.

I believe the above is a goal. Are you saying that this goal has
already been achieved?

> 2) How are they allocated/requested, and how and to whom are they
> released?

>     *	Servers advertise releasable resources. We (WG) need to consider how
> 	releasable resources are advertised.

>     *	Clients MUST request releasable resources, otherwise the server
> 	will not allocate them.

>     *	Servers allocate releasable resources for a period of time set by
> 	administrative policy

>     *	A client can renew releasable resources w/ allocating server, after
> 	the first timer expires. If unable to contact the server, the client
> 	must stop using resource when second timer expires.

This points out another question. The IP address extension has valid
and preferred lifetimes. Presumably all releasable resources would
need to have an associated lifetime. Shouldn't all releasable resource
extensions then have a somewhat standard format so that clients can
find the lifetime information in a standard place? (I don't recall the
current draft doesn't mentioning this.)

>     * A client SHOULD release releasable resources to the allocating
> server.  If it fails to contact the allocating server, it MAY locate
> another DHCP server and attempt to release them. In any case, the
> server will recover the resource when its lease has expired. Thus,
> like DHCPv4, releasable resources (addresses) will ultimately be
> reclaimed as administrative policy dictates, and it isn't critical
> for clients to contact the server.

> 3) Multiple Server Environments

>     * Servers advertise available IP address(es) and other
> releasable resources, thus clients know what resources are available
> from which server.

>     * We should avoid over specifying how DHCP works within
> multi-server environments. However, an introductory/scenario section
> w/ a protocol timeline would help validate the protocol model, and
> give some guidance to implementors where needed.

I still don't understand what multi-server aspects we need to
address. I think this is an important question that needs to be
resolved. In some earlier messages, it was pointed out that having a
client get conflicting information from different servers and then
requiring the client to somehow prioritize which information to use is
a hard problem. I agree. However, not addressing the problem makes it
pretty hard to claim that the protocol works in a multi-server
environment. Before deciding whether to solve or not solve this
problem, we need to decide whether the problem needs to be
addressed. I.e., do we think this problem is important?

For clarity, I'm talking about the multi-server case where there are
two different servers with different parameters/resources and the
client is required to communicate with both in order to get all the
information it needs.

> Miscellaneous:

> 	* I think we do need an "extension request extension" - having
> such an extension would simplify the advertisement of informational
> extensions, as well as simply a client's method for requesting
> extensions. Note: We need to specify that a policy switch should be
> available on the server to determine when an extension is returned
> (because of administrative policy, or because client requested it?).
> This point has been a bone of contention in DHCPv4, and we should
> endeavor to clarify the semantics in DHCPv6.

>      * Erik raised an interesting point about application layer
> extensions on machines multiple interfaces: If you have multiple NTP
> servers available from DHCP information from different interfaces,
> which one do you select? In our (Sun's) DHCPv4 client
> implementation, we have a notion of a "primary" interface, from
> which application layer parameters take precedence over those set in
> other interfaces. It seems to me that a similar DHCPv6
> implementation strategy could resolve this issue. Charlie suggested
> that the server preference value be used to determine
> preference. His suggestion has the advantage that preference is set
> by the administrator. I prefer the former solution ("primary"
> interface).

I'm not sure the server preferance field should be used for this. My
understanding of the preference field was that it was to be used to
select the "highest priority" server a collection of servers, where
all of the servers in the set are equivalent (i.e., any of them
returns correct information). These servers presumably have the same
information and are administered by the same organization. But the
multi-server problem I mention above also occurs when the multiple
servers are not under the same administrative domain. i.e., it
wouldn't make sense to compare server preferences because they are
assigned idipendently by the different domains.

Thomas


From dhcp-v6@bucknell.edu  Tue Nov 30 13:43:29 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA02381
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 13:43:27 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id NAA20416;
	Tue, 30 Nov 1999 13:38:23 -0500 (EST)
Date: Tue, 30 Nov 1999 13:38:23 -0500 (EST)
Message-Id: <199911301737.MAA16660@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 10:01 AM 11/30/99 -0500, Jim Bound wrote:
>>While the spec does not specifically require the use of one server at a
>>time, much of the language seems to me to imply the use of a single server.
>> And, as far as I can recall, this is the first time we've formally
>>discussed the idea of multiple simultaneous transactions with different
>>servers.  I think the claim that the spec currently allows the client to
>>ask for parameters from more than one server is not entirely supported by
>>the text in draft-ietf-dhc-dhcpv6-14.txt; e.g.:
>
>I think the issue here is it would make folks more comfortable if we had
>some sections that actually mention "multiple servers" in some sense
>where it might be appropriate?

Yes, that's right.  At least a few words pointing out that a
client *may* choose to interact with multiple servers simultaneously.
There may be other text required to ensure correct interoperation with
multiple servers, as you 


>But as I read the text below it is idempotent to the multiple server
>question, except when we "say pick a server"?
>
>>3.3. Request/Response Processing Model
>>
>>   The request/response processing for DHCPv6 is transaction based and
>>   uses a set of best-effort messages to complete the transaction.
>
>>   To find a server, a client sends a DHCP Solicit from the interface
>>   which it wishes to configure.  The client then awaits a DHCP
>>   Advertise message containing an IP address of a DHCP server.
>>   Transactions are started by a client with a DHCP Request, which may
>>   be issued after the client knows the server's address.  The server
>>   then unicasts a DHCP Reply, possibly via a relay.  At this point in
>>   the flow all data has been transmitted and is presumed to have been
>>   received.
>
>This is idempotent and I can see no where we break the multiple server
>model.  The protocol keeps on ticking whether there is 1...N links and
>1...N servers.  

It doesn't break the multiple server model, but (as I interpret the text)
does not explicitly point out that the client may interact with multiple
servers.  This might be a good place to add text about how the spec
allows for concurrent interaction with multiple servers.

>4.3. DHCP Request Message Format
>
>   In order to request configuration parameters from a server, a client
>   sends a DHCP Request message, and MAY append extensions [15].  If
>   the client does not know any server address, it MUST first obtain
>   one by multicasting a DHCP Solicit message (see Section 4.1).
>
>Here is where we might add that this solicit in fact will support
>multiple instants of solicit on different links (subnets in IPv4 terms)
>or in section 4.1.  


OK...

>5.3. Receiving DHCP Advertise Messages
>
>   After a client has received a DHCP Advertise message, it has the
>   address of a server for subsequent DHCP Request messages.
>
>   [...]
>
>   Unless a DHCP Advertisement is received with a preference equal
>   to 255 (see Section 4.2), the client MUST wait CLIENT_ADV_WAIT
>   seconds after issuing the DHCP Solicit message in order to receive
>   the Advertisement with the highest preference.  After waiting for
>   that period of time, a client MUST select the highest preference
>   server as the target of its DHCP request.  If a client receives an
>   advertisement with a preference of 255, it does not have to wait for
>   any more advertisements.
>
>Here what we are saying was an addition from the Chicago meeting.
>
>We assume the client on the same link could hear from multiple servers
>as we were told by the working group in Chicago.  Per the working group
>we added the preference field to the Advertisement message.  The way the
>code and implementation could work is as follows:
>
>  boolean head_cheese_server = FALSE;
>  int *head_cheese_server_data
>  pthread_start_client_solicit;  /* this is not how it looks but */
>  time_interrupt_set (CLIENT_ADV_WAIT);
>  while (pthread_CLIENT_ADV_WAIT_not_back) {
>    get_advertisement_message(server_msg_block);
>    If (server_msg_block->preference_value = 255) {
>       head_cheese_server = TRUE;
>       head_cheese_server_data = server_msg_block;
>       break;
>    } else 
>        store_this_server_for_later_sort (server_msg_block);
>  }
>  /* not putting in check here for discussion if no advertisements
>     came back */
>  If (head_cheese_server)
>   use_the_head_cheese_server_for_request(server_msg_block);
>  else
>   sort_and_pick_highest_preference_server(server_msg_block);
>
>But the point is that we listen to multiple servers and select among
>them based on the preference field from the server unless a server says
>I am the boss or head cheese use me.  The first server to be heard that
>does this wins.

I don't see how this mechanism allows for use of more than one server.
The mechanism explicitly says "pick the one head_cheese or highest
priority server".  What crieria does the client use to decide to use
and choose a second server? 

>The client may store all server advs for later use too.
>
>   If a client sends a DHCP Request to a highly preferred server but
>   fails to receive a DHCP reply from that server after following the
>   retransmission algorithm in section 8, the client MUST then try to
>   send a DHCP Request to a less preferred server.
>
>>(I think this section, at least, needs to be rewritten to allow interaction
>>with multiple servers.  Under what conditions may a client disregard the
>>server preference and begin a transaction with a second server with a
>>[presumably] lower preference?)
>
>So do you believe we should say the client SHOULD store servers that
>respond to the solicit and then discuss the case when the client will
>use a server of lower preference?
>
>This I believe resolves your concern and is doable?

No, this doesn't exactly address my concern.  What I read is a mechanism
for *serial contact* with multiple servers, but *interaction* with only
one server at a time; i.e., the client works down the list, one server at a
time, until it successfully completes a transaction with *one* server.

- Ralph


From dhcp-v6@bucknell.edu  Tue Nov 30 15:34:44 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA00322
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 15:34:43 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA20282;
	Tue, 30 Nov 1999 15:31:23 -0500 (EST)
Date: Tue, 30 Nov 1999 15:31:23 -0500 (EST)
Message-Id: <199911301857.NAA22433@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

Oops...the ", as you" in the following sentence from my previous message
should be replaced with ".":

There may be other text required to ensure correct interoperation with
multiple servers, as you 

- Ralph




From dhcp-v6@bucknell.edu  Tue Nov 30 15:34:45 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA00323
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 15:34:43 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA07865;
	Tue, 30 Nov 1999 15:31:11 -0500 (EST)
Date: Tue, 30 Nov 1999 15:31:11 -0500 (EST)
Message-Id: <38441B9D.1ACD7044@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Ralph,

You wrote:

> Charlie, to improve my understanding of DHCPv4 and to be able to tell
> students about when state machines may not be an appropriate specficiation
> or implementation tool, can you give me an example of how the DHCPv4 state
> machine was a problem in your implementation of DHCPv4.  How would the
> implementation have been made easier without the state machine, and how can
> the behavior of the DHCPv4 client be specified clearly and concisely
> without a state machine?

I don't think that using state machines automatically leads to
inconvenient implementations.  I only know that the implementations
I had to use some years ago had this problem.

Probably it is better to describe states as pertaining to a
transaction, than to a client.  Maybe it would be enough to
specify that any client has to be able to run several state
independent state machines simultaneously.  Maybe it would
be even better to specify that state variables should not have
global process scope.  Thread variables might have to be indexed
into arrays per thread or something.

On the other hand, I think that many protocols can be specified
in a way that is simple enough to allow correct implementations
that don't rely on state machines.

I was happy to read Ted Lemon's note about ISC's implementation.
I wonder if it was the initial design of the ISC system, or instead
a design that was arrived at from experience with other more
restrictive state-based implementations...

If this is too terse, my apologies, but I'll have to wait until
later to write more if needed.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 16:12:14 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA00321
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 15:34:43 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id PAA26625;
	Tue, 30 Nov 1999 15:30:26 -0500 (EST)
Date: Tue, 30 Nov 1999 15:30:26 -0500 (EST)
Message-Id: <3844197D.2E58FBD7@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Thomas Narten wrote:
> 
> Ralph Droms <droms@bucknell.edu> writes:

Hello Thomas,

>                                        The underlying message
> formats certainly could support a client having multiple
> interactactions with different servers simultaneously.  However, the
> spec itself has little text on this and my expectation of what folks
> would actually implement (based on my read of what the current text
> says should be implemented) is that clients will in practice pick one
> server and interact only with it.

Here, I agree with you.  I think this will be the most
typical case.  Except for the boundary cases where a server
crashes and a client has to go find another server, for
example.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 17:11:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA17460
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 17:11:04 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA11588;
	Tue, 30 Nov 1999 17:04:51 -0500 (EST)
Date: Tue, 30 Nov 1999 17:04:51 -0500 (EST)
Message-Id: <199911302049.PAA00400@funnel.cisco.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ralph Droms <droms@bucknell.edu>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Type: text/plain; charset="us-ascii"
Mime-Version: 1.0
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.0

At 03:31 PM 11/30/99 -0500, Charles E. Perkins wrote:
>> [...] clients will in practice pick one
>> server and interact only with it.
>
>Here, I agree with you.  I think this will be the most
>typical case.  Except for the boundary cases where a server
>crashes and a client has to go find another server, for
>example.

But I would not consider this case example to be an
example of *concurrent*
interaction with multiple servers.  I would expect
a client to support one-server-at-a-time interaction
with multiple servers; e.g., server selection by
descending preference.  I think the spec needs to be clearer
about allowing transactions with more than one server
at a time.

- Ralph




From dhcp-v6@bucknell.edu  Tue Nov 30 17:11:06 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA17459
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 17:11:03 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id RAA12950;
	Tue, 30 Nov 1999 17:06:10 -0500 (EST)
Date: Tue, 30 Nov 1999 17:06:10 -0500 (EST)
Message-Id: <199911302100.QAA04834@grosse.manhattan.fugue.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Ted Lemon <mellon@isc.org>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources 
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.


> I was happy to read Ted Lemon's note about ISC's implementation.
> I wonder if it was the initial design of the ISC system, or instead
> a design that was arrived at from experience with other more
> restrictive state-based implementations...

The ISC DHCP client was designed off the protocol spec - neither I nor
Elliot Poger had any experience with other implementations.  The ISC
client does use state machines, BTW - just not a single state machine.
Elliot's original implementation held the state machine global and
forked a new process for each interface, but this seemed unnecessary
to me, so I stuffed the state machine into a per-interface data
structure.  Each network interface then had its own data structure
with timers, parameters to send to the server, and so on.  The whole
thing is event driven - the main program launches the initial event in
each state machine, and then just sits there and handles new events
(timeouts and packet arrivals) as they happen.   Incoming events are
associated with particular state machines based on the transaction
id.

I later (in version 3.0) made a single pervasive change to the client
to support an arbitrary number of state machines per interface.  This
means that if you want to have more than one IP address per interface,
you instantiate a state machine for each address you want, each with
its own unique client identifier (and other parameters, if desired).
Installing IP addresses on the interface gets trickier if you have
more than one per interface, but that is the only real complication.
My motivation in doing this was to implement a DHCP test client, but
since I've done it I've had several people take it and use it to
manage multiple IP addresses per interface.

I would suggest that for DHCPv6, the state machine be associated with
the entity needing an IP address.  That is, if your DHCP client knows
it needs three IP addresses of various kinds, it would set up three
state machines to manage those addresses.  But this is said from a
naive point of view - I have not read the spec closely enough recently
enough to know whether this is the right way of putting it, or that it
makes sense at all.  (And I apologize for that - I'm only
participating right now because I get the feeling I may be of some
service, even if I can't speak authoritatively.)

			       _MelloN_


From dhcp-v6@bucknell.edu  Tue Nov 30 18:53:04 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA08927
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 18:53:02 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA27385;
	Tue, 30 Nov 1999 18:49:03 -0500 (EST)
Date: Tue, 30 Nov 1999 18:49:03 -0500 (EST)
Message-Id: <3846E137.1EB4F770@east.sun.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Michael W. Carney" <michael.carney@east.sun.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: Summary: DC meeting question #1
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.7 [en] (X11; I; Linux 2.2.5-22 i686)
Content-Transfer-Encoding: 7bit

Hi Thomas,


Thomas Narten wrote:
> 
> Mike,
> 
> I've got some clarification questions.

Sure.

> 
> > Semantics of releasable resources:
> 
> > 1) General Form
> 
> >     * Releasable resource extensions need to have a general form so that
> >       clients/servers can manage them in a generic fashion.
> 
> >     * There is a generic bit in the extensions header which is set if the
> >       extension is carrying a releasable resource. When this bit is set,
> >       there are two 32 bit fields present in the extension which represent
> >       the relative times in seconds when the client should start attempting
> >       to extend the lease on the resource, and when the client must stop
> >       using the resource if it is unable to extend it.
> 
> Clarification: are you saying the list discussion is arguing for such
> a bit? Or that one will go into the draft? Or? (I don't recall the
> current draft having such a bit.)

The discussion clearly showed that support for releasable resources was 
something that the WG wanted. That said, what would be the semantics
that would resolve the issues raised during the discussion of this
question? It is this point I'm attempting to address.

General support for releasable resources would require a general method
of communicating between client/server that the resource described in
the extension was releasable. That's what the bit is used for, along
with the lease times. Thus, releasable resources would have a have a
general extension format (A bit in the header which clues the
implementation that
the two lease/lifetimes is present).

Perhaps "summarize" was a poor choice of words.


> 
> >     * Since extensions carrying releasable resources have a general
> >       form/semantics, clients/servers do not need to change when new
> >       extensions are defined which are releasable.
> 
> I believe the above is a goal. Are you saying that this goal has
> already been achieved?

No, although this message's purpose was to propose semantics of such
extensions in pursuit of that goal.

> 
> > 2) How are they allocated/requested, and how and to whom are they
> > released?
> 
> >     * Servers advertise releasable resources. We (WG) need to consider how
> >       releasable resources are advertised.
> 
> >     * Clients MUST request releasable resources, otherwise the server
> >       will not allocate them.
> 
> >     * Servers allocate releasable resources for a period of time set by
> >       administrative policy
> 
> >     * A client can renew releasable resources w/ allocating server, after
> >       the first timer expires. If unable to contact the server, the client
> >       must stop using resource when second timer expires.
> 
> This points out another question. The IP address extension has valid
> and preferred lifetimes. Presumably all releasable resources would
> need to have an associated lifetime. Shouldn't all releasable resource
> extensions then have a somewhat standard format so that clients can
> find the lifetime information in a standard place? (I don't recall the
> current draft doesn't mentioning this.)

Exactly my point! That's the first bullet in #1, General form.

> 
> >     * A client SHOULD release releasable resources to the allocating
> > server.  If it fails to contact the allocating server, it MAY locate
> > another DHCP server and attempt to release them. In any case, the
> > server will recover the resource when its lease has expired. Thus,
> > like DHCPv4, releasable resources (addresses) will ultimately be
> > reclaimed as administrative policy dictates, and it isn't critical
> > for clients to contact the server.
> 
> > 3) Multiple Server Environments
> 
> >     * Servers advertise available IP address(es) and other
> > releasable resources, thus clients know what resources are available
> > from which server.
> 
> >     * We should avoid over specifying how DHCP works within
> > multi-server environments. However, an introductory/scenario section
> > w/ a protocol timeline would help validate the protocol model, and
> > give some guidance to implementors where needed.
> 
> I still don't understand what multi-server aspects we need to
> address. I think this is an important question that needs to be
> resolved. In some earlier messages, it was pointed out that having a
> client get conflicting information from different servers and then
> requiring the client to somehow prioritize which information to use is
> a hard problem. I agree. However, not addressing the problem makes it
> pretty hard to claim that the protocol works in a multi-server
> environment. Before deciding whether to solve or not solve this
> problem, we need to decide whether the problem needs to be
> addressed. I.e., do we think this problem is important?

It is important in DHCPv4 land, and it is just as important in DHCPv6,
IMHO. See Miscellaneous, second bullet, where I discussed Erik's issue
in detail.

> 
> For clarity, I'm talking about the multi-server case where there are
> two different servers with different parameters/resources and the
> client is required to communicate with both in order to get all the
> information it needs.
> 
> > Miscellaneous:
> 
> >       * I think we do need an "extension request extension" - having
> > such an extension would simplify the advertisement of informational
> > extensions, as well as simply a client's method for requesting
> > extensions. Note: We need to specify that a policy switch should be
> > available on the server to determine when an extension is returned
> > (because of administrative policy, or because client requested it?).
> > This point has been a bone of contention in DHCPv4, and we should
> > endeavor to clarify the semantics in DHCPv6.
> 
> >      * Erik raised an interesting point about application layer
> > extensions on machines multiple interfaces: If you have multiple NTP
> > servers available from DHCP information from different interfaces,
> > which one do you select? In our (Sun's) DHCPv4 client
> > implementation, we have a notion of a "primary" interface, from
> > which application layer parameters take precedence over those set in
> > other interfaces. It seems to me that a similar DHCPv6
> > implementation strategy could resolve this issue. Charlie suggested
> > that the server preference value be used to determine
> > preference. His suggestion has the advantage that preference is set
> > by the administrator. I prefer the former solution ("primary"
> > interface).
> 
> I'm not sure the server preferance field should be used for this. My
> understanding of the preference field was that it was to be used to
> select the "highest priority" server a collection of servers, where
> all of the servers in the set are equivalent (i.e., any of them
> returns correct information). These servers presumably have the same
> information and are administered by the same organization. But the
> multi-server problem I mention above also occurs when the multiple
> servers are not under the same administrative domain. i.e., it
> wouldn't make sense to compare server preferences because they are
> assigned idipendently by the different domains.

That's true - this breaks down in multi administrative domains. However,
the primary interface concept does not. What do you think of this
concept?
> 
> Thomas


-- 

Mike Carney


From dhcp-v6@bucknell.edu  Tue Nov 30 18:53:07 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA08984
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 18:53:07 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id SAA17046;
	Tue, 30 Nov 1999 18:49:36 -0500 (EST)
Date: Tue, 30 Nov 1999 18:49:36 -0500 (EST)
Message-Id: <3844614C.7C5E879E@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Ted,

>                                               Incoming events are
> associated with particular state machines based on the transaction
> id.

This makes sense to me.

> I would suggest that for DHCPv6, the state machine be associated with
> the entity needing an IP address.

Would you be willing to support my contention that, instead, the
state machine be associated with the transaction by which the
IP address is expected to be obtained?  This seems to avoid some
of the possible negative interaction between independent requests.

>                             (And I apologize for that - I'm only
> participating right now because I get the feeling I may be of some
> service, even if I can't speak authoritatively.)

It seems to me that you have good ideas to offer.  Thanks for
getting them out there for us to think about.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 20:01:46 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA13308
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 20:01:45 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id TAA12275;
	Tue, 30 Nov 1999 19:58:08 -0500 (EST)
Date: Tue, 30 Nov 1999 19:58:08 -0500 (EST)
Message-Id: <38446AFA.825929CC@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello Richard,

You wrote:

> I've just joined this discussion so I'm not up to speed... Are you saying,
> send different RAs on different links (of course this is OK) or send two
> different RAs on the same link (maybe OK, but most implementations today
> don't support it I would guess).

I don't see why it is wrong to send different RAs on the same link.
In fact, I thought that this possibility was explicitly discussed some
years ago on the mailing list.  There was concern about what might
happen if the advertised parameters (e.g., MTU) were incompatible.

> This is confused - a router either advertises itself as a default router or
> not, there's no concept of being "default router for a prefix".

A router advertises a prefix, and its willingness to be a default
router.  I infer that it wants to be a default router for that
prefix.  I suspect there would be cases where a router would want
to be a default router for some prefix(es) and not for others.
This would be especially true for the multihoming case.  In such
cases, I reckon distinct advertisements would be needed.

Or maybe I am confused.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 20:38:01 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29366
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 20:38:01 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA32325;
	Tue, 30 Nov 1999 20:36:00 -0500 (EST)
Date: Tue, 30 Nov 1999 20:36:00 -0500 (EST)
Message-Id: <38447895.4B49C85@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Ralph Droms wrote:

>                   It is not fair to say that our specification does
>> not support Cisco routers just because we don't say how to use them.
>> It is not fair to say that our spec does not support mobile computers
>> just because we don't say how to use them.  I think your contention here
>> is, I wish I could say it more tactfully but I don't know how, totally
>> wrong.
> 
> I disagree with Charlie, here.  The examples he cites are irrelevant to
> this situation.

True!  However, the point _was_ to show that irrelevant details do NOT
need to be specified.  I strive to eliminate specification of irrelevant
details.  And, just because a detail is irrelevant to the specification,
does NOT mean it is not important for the implementation.

>              The ability to use DHCPv6, as documented in the current
> draft, with multiple servers does *not* imply that it will be implemented
> in a useful, interoperable and useful way.

I think that the current specification promotes interoperability,
so I must disagree with your assessment.  Usefulness is, perhaps, in
the mind of the beholder.  The features in the protocol are, to my mind,
quite likely to be useful now and for many years in the future.
Features that are unlikely to ever be useful should be pointed out
and nominated for removal.


> Let me take this a step further and say that using the dictum of "let the
> market decide" has *not* worked for DHCPv4 *clients*.  There is one
> predominant client, which has some well-documented interesting behaviors
> that have been fixed in later releases.

Hmmm... I thought that many of the problems arose because the
specification was not followed.  Whenever there are ambiguities
in the specification, they should be eliminated.  If implementor A
assumes behavior (X) that is not in the specification, and implementor B
assumes behavior (Y) that is not in the specification, then both
implementors are doing something wrong.  If implementor C assumes
behavior (Z) that contradicts an explicit specification, then
there is really no hope is there?

I think it is almost impossible to anticipate every wrong possibility
for implementation and doctor up the specification to forestall them
all.  It is far better for the specification to just say what is needed,
and not delve into ambiguities and philosophy any more than is necessary
to get the ideas across.  The more text, the more places there are to
get stuff wrong.  Duplicating text across independent parts of the
specification is an open invitation to error, as subsequent revisions
cause changes in one clone of the duplicated text to be mistakenly
omitted from another clone of the duplicated text.

> We must be very careful to get the client specification right from the
> beginning...

I sure do agree with you, Ralph,

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 20:38:17 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29489
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 20:38:15 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA22482;
	Tue, 30 Nov 1999 20:35:33 -0500 (EST)
Date: Tue, 30 Nov 1999 20:35:33 -0500 (EST)
Message-Id: <38447386.52627821@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Ralph Droms wrote:

> Rather than associating a resource with a particular *server*, it might be
> possible to associate a resource with a DHCP *service* (which may be
> provided by more than one server).  If servers and resources were
> identified by service, a client could choose among multiple, equivalent
> servers within a service.  In this way, a client could take advantage of
> redundant *servers* within a *service*, and avoid the risk of returning the
> resource to some other server that cannot process the resource.

Once a configuration item is designated a "service", with
multiple interchangeable servers, I think it is better suited
as a candidate for identification by way of Service Location
Protocol.  SLP was designed for exactly this kind of operation.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 20:38:28 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29576
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 20:38:24 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA03066;
	Tue, 30 Nov 1999 20:35:44 -0500 (EST)
Date: Tue, 30 Nov 1999 20:35:44 -0500 (EST)
Message-Id: <384475BB.B3B5EF11@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Ralph Droms wrote:


> While the spec does not specifically require the use of one server at a
> time, much of the language seems to me to imply the use of a single server.
>  And, as far as I can recall, this is the first time we've formally
> discussed the idea of multiple simultaneous transactions with different
> servers.

I didn't go over the archives, but I know that I have discussed this
possibility many times in the past.  And, I was pretty sure that it had
come up in conversations at IETF.  Those hardly qualify as formal
discussion, I know.

>            I think the claim that the spec currently allows the client to
> ask for parameters from more than one server is not entirely supported by
> the text in draft-ietf-dhc-dhcpv6-14.txt; e.g.:
> 
> 3.3. Request/Response Processing Model
> 
>    The request/response processing for DHCPv6 is transaction based and
>    uses a set of best-effort messages to complete the transaction.
> 
>    To find a server, a client sends a DHCP Solicit from the interface
>    which it wishes to configure.  The client then awaits a DHCP
>    Advertise message containing an IP address of a DHCP server.
>    Transactions are started by a client with a DHCP Request, which may
>    be issued after the client knows the server's address.  The server
>    then unicasts a DHCP Reply, possibly via a relay.  At this point in
>    the flow all data has been transmitted and is presumed to have been
>    received.

This paragraph is worded precisely to allow multiple simultaneous
transactions.  A client sends a DHCP Solicit from the network interface.
It could send another Solicit from another interface.  The same
interface
could be configured with multiple prefixes (somehow).  The client could
send a separate Solicit for each prefix, perhaps at the same time.
I do not see any wording that even prevents the client from sending
multiple Solicits from the same interface at the same time for the
same prefix.  However, in that case, certain timing parameters must
be respected.


> 4.3. DHCP Request Message Format
> 
>    In order to request configuration parameters from a server, a client
>    sends a DHCP Request message, and MAY append extensions [15].  If
>    the client does not know any server address, it MUST first obtain
>    one by multicasting a DHCP Solicit message (see Section 4.1).

This specifies what a client MUST (at minimum) do to request
configuration
parameters.  It does not prohibit additional actions.

> 5.3. Receiving DHCP Advertise Messages
> 
>    After a client has received a DHCP Advertise message, it has the
>    address of a server for subsequent DHCP Request messages.

This statement is true regardless of how many advertisements the client
receives.  It can get multiple advertisements from different servers,
with different preferences.  It can get one advertisement only.  It
might
only get one advertisement if it sent out various solicitations, perhaps
due to network congestion.  In fact, it might get zero, in which case
it would resend the Solicitation.

>    Unless a DHCP Advertisement is received with a preference equal
>    to 255 (see Section 4.2), the client MUST wait CLIENT_ADV_WAIT
>    seconds after issuing the DHCP Solicit message in order to receive
>    the Advertisement with the highest preference.  After waiting for
>    that period of time, a client MUST select the highest preference
>    server as the target of its DHCP request.  If a client receives an
>    advertisement with a preference of 255, it does not have to wait for
>    any more advertisements.

This specifies how long a client has to wait before sending its
DHCP Request message.  It does not limit how many interfaces a
client might want to simultaneously configure.

>    If a client sends a DHCP Request to a highly preferred server but
>    fails to receive a DHCP reply from that server after following the
>    retransmission algorithm in section 8, the client MUST then try to
>    send a DHCP Request to a less preferred server.
> 
> (I think this section, at least, needs to be rewritten to allow interaction
> with multiple servers.  Under what conditions may a client disregard the
> server preference and begin a transaction with a second server with a
> [presumably] lower preference?)

So far, I didn't see what was wrong.  On the other hand, if you would
like to suggest some clearer wording, I think that would be a good way
to go about resolving any lingering doubts, as well as eliminate them
before they arise for future readers of the specification.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 20:38:32 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29628
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 20:38:30 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA12121;
	Tue, 30 Nov 1999 20:36:10 -0500 (EST)
Date: Tue, 30 Nov 1999 20:36:10 -0500 (EST)
Message-Id: <4D0A23B3F74DD111ACCD00805F31D8101CA21784@RED-MSG-50>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: Richard Draves <richdr@microsoft.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: RE: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
X-Mailer: Internet Mail Service (5.5.2650.21)

> > I've just joined this discussion so I'm not up to speed... 
> Are you saying,
> > send different RAs on different links (of course this is 
> OK) or send two
> > different RAs on the same link (maybe OK, but most 
> implementations today
> > don't support it I would guess).
> 
> I don't see why it is wrong to send different RAs on the same link.
> In fact, I thought that this possibility was explicitly discussed some
> years ago on the mailing list.  There was concern about what might
> happen if the advertised parameters (e.g., MTU) were incompatible.

There's nothing wrong with it, what I meant is that I don't know of any IPv6
router implementations that will actually do it today.

> > This is confused - a router either advertises itself as a 
> default router or
> > not, there's no concept of being "default router for a prefix".
> 
> A router advertises a prefix, and its willingness to be a default
> router.  I infer that it wants to be a default router for that
> prefix.  I suspect there would be cases where a router would want
> to be a default router for some prefix(es) and not for others.
> This would be especially true for the multihoming case.  In such
> cases, I reckon distinct advertisements would be needed.

To repeat myself, there's no concept of being "default router for a prefix".
When a router advertises a prefix (meaning includes a Prefix Information
option in a Router Advertisement), the prefix can mean several things. If
the A bit is set, then the prefix is used for stateless address
auto-configuration. If the L bit is set, then prefix is put in the host's
on-link prefix list. In either case, there's no connection between the
prefix and the router that advertised it.

When you say "default router for a prefix" it sounds like you mean
"destination addresses matching the prefix should be sent to the router"
which is certainly not correct.

Rich


From dhcp-v6@bucknell.edu  Tue Nov 30 20:38:48 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA29775
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 20:38:47 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id UAA07824;
	Tue, 30 Nov 1999 20:35:07 -0500 (EST)
Date: Tue, 30 Nov 1999 20:35:07 -0500 (EST)
Message-Id: <38447243.C27B12C6@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DISCUSSION: Use of Address Extensions with DHCPv6
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit


Hello folks,

I am getting a feeling that things are seeming much more complex
than they really are.  I'll try to bring as much simplicity back
into this conversation as I can possibly do.


> Seems to me (after no small amount of consideration) that a server needs to
> provide more information in its advertisements then we have been proposing
> (note that providing this information doesn't change the protocol as defined
> at all). Consider my summary of question #1 - I've used it in responding to
> this message.

I think that when there are multiple servers, you are right that
a client might wish to get more information in the advertisements.
When there is just one, no additional advertisements would help.
When the client doesn't care, no additional advertisements are
needed.

Allowing extensions to the advertisement message is a suitable way
to allow for administrators to cater to sophisticated clients.  But
on the other hand I don't think it means that all servers need to
advertise everything they might be willing to offer to the clients,
especially because so many clients won't care very much.

> > 1.  Client sends DHCPv6 Request with no address extension.
> >
> > 1a. Server responds with all addresses available to the client
> >     if the sys admin has defined the server can do this.
> >
> > 1b. If the server gets a second message from the client with no
> >     address extension it resends the client addresses it gave
> >     them previously cumultatively (see below).

If the client sends a request with the same transaction ID, it
should get the same reply.  If the client sends the request with
no extensions and a different transaction ID, then the server should
send one or more new configuration items if it is willing to do so
for that client.

If the client wanted to renew a specific address, it would include that
address in an IP address extension.  If it does not include that address
for a valid IP address it is using, it is not asking the server's
opinion about that IP address.

This algorithm seems pretty unambiguous to me.

If the working group decides that IP addresses must be explicitly
requested, then of course the client wouldn't get any IP addresses
if it doesn't include the address extension or the as-yet-undefined
Parameter Request extension.

My first paragraph above is worded in such a way that it should still
hold true even if the working group decides to allow a client to get
addresses without explicitly asking for them.  I would prefer that this
decision be made to allow such freedom, but I don't want to get dragged
again over the coals of consistency with future extensions for
releaseable
resources.  We decided in Minneapolis that all releaseable resources
needed
to be explicitly requested, even if right now everybody knew that IP
address extensions could be designed otherwise.

> #1 goes against the constraint I listed in the Summary: DC meeting question #1
> message: Clients must explicitly request releaseable resources. If the server
> had explicit (through policy) or implicit (dynamic addresses) to offer to the
> client,
> it would do so by adding IP address extensions carrying these addresses to the
> advertisement. The client would request one or all of these IP addreses in its
> request. If the site is using stateless autoconf, then the server would not
> have any IP address extensions in its advertisement, and thus the client could
> not request an IP address.

I think a client should be able to request an IP address even if the
site is using stateless address autoconfiguration.  This does not lead
to any ambiguities to my best current understanding.

Furthermore, I don't think that the client has to request an address
that is listed in an advertisement, and I don't think that the
advertisement
has to have an IP address extension before the client is allowed to get
an IP address.


> > 2.  Client sends Request with 1 or more "empty" address extensions
> >
> > 2a. Server responds, if the client is permitted a new address, with a new
> >     IPv6 address in addition to the ones given to the client previously.
> >     If the client is not permitted a new address then an error is
> >     returned to the client.
> >     The previous addresses are not resent to the client in the
> >     Server Reply.
> >
> > 2b. The client MUST NOT send this through a RELAY-AGENT and the
> >     client is assumed to have an address with enough scope to
> >     reach the server directly.  This permits the use of DHCPv6
> >     authentication or IPsec to be used to access the server.
> >     [In fact it probably should be a MUST have security for
> >      this type of transaction.]

The client should be able to send any DHCP Request through a
relay agent.  Why not?  Authentication is fine, and I don't think
that one should be required to authenticate the request just to get
an address.  There are lots of them.  On the other hand, a server
should be allowed to require authentication if that is site policy.


> I'm assuming that this request is in addition to the IP address(es) offered in
> the advertisement. The client will have knowledge of what networks are
> available
> from the advertisement, and thus would need to tell the server which network it
> 
> would like an IP address on. Thus, an empty IP address extension isn't enough -
> the
> client would need to communicate which network.

The server is supposed to know which network it is by looking at the
relay agent address, or else by looking at the client address.

>    At DC, I proposed a change to the
> IP address extension which would permit the client to identify which network it
> wants
> an additional IP address on. Whether or not it would receive an address is up
> to
> administrator policy. Note that a server could use this modification for
> advertising
> network prefixes, without advertising a specific IP address on that network.

I don't see the need for this.

> I don't see that the transaction id needs to be part of the binding. The XID is
> cached
> to enable a server to distinguish retransmissions from new requests. The
> link-local
> address, agent prefix, and IP address should be enough.

The transaction id has to be stored somewhere for the client, and
the binding is an O.K. place.  Other places are also O.K.  We discussed
this recently on this list, I think.


> > * we can add in the appendix that a server can structure the database
> >   so that the main key is link-local + agent-prefix and sub keys are
> >   the remaining parts.  We should also point out that when dhcpv6
> >   operates in this mode there is more cost for memory and for
> >   performance as the sever must now locate previous transactions
> >   with this behavior or use a binding-cache (see discussion below).
> 
> Do we really want/need to specify this?

I don't think so, myself.


> > 3.  Client sends Request with address extension or extensions
> >     which are not empty.
> >
> > 3a. Server adds these addresses to the sever database for this
> >     client.  Server sends Reply back to the client if it is
> >     permitted to do this with the addresses the client sent
> >     back to the client in the Reply.
> >
> > 3b. The client MUST NOT send this through a RELAY-AGENT and the
> >     client is assumed to have an address with enough scope to
> >     reach the server directly.  This permits the use of DHCPv6
> >     authentication or IPsec to be used to access the server.
> >     [In fact it probably should be a MUST have security for
> >      this type of transaction.]
> >
> > We really need to discuss this scenario.  Do we want to permit this type
> > of behavior in DHCPv6.  This breaks the model of DHCPv4 yet again and is
> > really new stuff???????????  But we could make it work.

I do not understand what is wrong here.  Clients should be able
to ask for IP addresses when they need one.  I disagree with Jim's
proposed restriction about avoiding Relay Agents.

If a client sends a nonempty IP address extension to a server, it
is asking that server to renew or extend the lifetime for that address.
This is very natural.  I don't see what is wrong.


> Jim, I don't think this is a good scenario. A Client should never be involved in
> telling a server about IP addresses (or other releasable resources) it may have
> that the server has no knowledge of. Sure, a client MAY request that a certain
> address be assigned to it, and if the request is valid from a network topology
> point of view and a administrator policy view, then fine. I just don't think we
> should be configuring DHCP servers with DHCP clients, IMHO.

While I agree with your point, Mike, I don't think it is an issue
given my interpretation of what it means for a client to include an
address in its DHCP Request.


> > BINDING-CACHE:  We would move the binding cache discussion to the
> > appendix as the above I think fixes idempotent problems in the protocol
> > standard.  The binding-cache would also lesson the cost/performance hit
> > I speak of above in implementations.  The server would only need to
> > store the Reply that has to be resent and an not the entire client
> > database.
> 
> Binding cache is very useful for preventing unnecessary database lookups. Ah,
> now I see where you're coming from wrt saving the transaction ID - you're
> attempting to record the assignment transaction to avoid confusion between
> retransmissions and new requests.

The latter sentence is precisely what the transaction ID is supposed to
be used for.  I don't think of any other function for it, just off the
top of my head right now.

> Is it really necessary to be this heavy-weight? Sure, as specified, there is
> a window where the server's XID timeout cache may timeout cache entries
> while a client is retransmitting (server reboots, loses cache information), but
> that window is very small. In most cases, the XID cache timeout will be large
> enough and the server stable enough to prevent the accidental assignment of
> additional IP address(es) (or other releasable resources). Alternately, a
> server could write its XID cache to stable storage from time to time, and
> reload it upon restart, thus satisfying the idempotent problem w/o having the
> cost of the extra space in the database and extra initial database lookup.

I am failing to see the problem here.  The transaction ID allows a
server
to recognize retransmitted Request messages.  It has a similar role for
Reconfigure messages.  I'm getting perplexed by what might be causing
this
to be confusing.


Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 21:11:39 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA16042
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 21:11:38 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA29604;
	Tue, 30 Nov 1999 21:08:49 -0500 (EST)
Date: Tue, 30 Nov 1999 21:08:49 -0500 (EST)
Message-Id: <38448038.4E330255@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Richard Draves wrote:

> When you say "default router for a prefix" it sounds like you mean
> "destination addresses matching the prefix should be sent to the router"
> which is certainly not correct.

You are right.  I am having a brain check.  Too many keystrokes,
too many messages to read.  I'll try to figure this out again later
after my eyes stop boiling.

Regards,
Charlie P.


From dhcp-v6@bucknell.edu  Tue Nov 30 21:12:20 1999
Received: from mail.bucknell.edu (marge.bucknell.edu [134.82.9.1])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id VAA16400
	for <dhc-archive@odin.ietf.org>; Tue, 30 Nov 1999 21:12:19 -0500 (EST)
Received: from reef.bucknell.edu (reef.bucknell.edu [134.82.7.6])
	by mail.bucknell.edu (8.9.3/8.9.3) with SMTP id VAA21468;
	Tue, 30 Nov 1999 21:07:57 -0500 (EST)
Date: Tue, 30 Nov 1999 21:07:57 -0500 (EST)
Message-Id: <38447CE7.4E59FD5B@iprg.nokia.com>
Errors-To: droms@bucknell.edu
Reply-To: dhcp-v6@bucknell.edu
Originator: dhcp-v6@bucknell.edu
Sender: dhcp-v6@bucknell.edu
Precedence: bulk
From: "Charles E. Perkins" <charliep@iprg.nokia.com>
To: Multiple recipients of list <dhcp-v6@bucknell.edu>
Subject: Re: DC meeting: Discussion question #1, releaseable resources
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
X-Comment: Discussion list for DHCPv6 internet protocol.
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0
X-Mailer: Mozilla 4.5 [en] (X11; I; FreeBSD 2.2.6-RELEASE i386)
Content-Transfer-Encoding: 7bit

Ralph Droms wrote:

> Put succinctly, I know of no abstract reason to preclude the concurrent
> interoperation of a DHCP client with multiple servers.  In my opinion, the
> current spec does not adequately describe the details of concurrent use of
> multiple servers to ensure interoperable DHCPv6 clients that will provide
> predictable, manageable, useful behavior.

I hope you be willing to point out text that is wrong, or suggest
some strategy for us to forestall erroneous interpretations.

> Potentially open issues in the use of multiple servers:
> 
> * Text that implies interaction with a single server through use of
> singular reference to "server" and through omission of any specific
> reference to multiple servers

I think we need to say in the introduction that the specification
explicitly allows multiple simultaneous transactions with different
DHCP servers.

> * Semantics of Solicit message in the presence of multiple DHCP services:
> how does a client know when it should interact with multiple servers; how
> does a client differentiate between redundant servers (from the same
> service) and servers with different resources?

I don't think a client needs to make the differentiation.  If it's
truly a service, it should be resolved using SLP.

I also think that DHCP should be invoked because of client needs.
Thus, when the client needs something, it will take action.  I would
interpret your question to be asking how does a client know when it
needs something.  I don't know how to answer that question, and I hope
we don't have to answer it in order to accomplish our near-term
objectives.

> * Text in section 5.3 describing use of server preference field that
> implies use of single server with highest preference

This is in response to a solicitation.  If the client uses the same
link-local address for each of its multiple solicitations, then there
is ambiguity in determining which advertisement belongs to which
solicitation.   The client can solve this problem by picking different
link-local addresses for each distinct solicitation.  This can be
specified, if the working group wants to make the distinction clear.

I think that these cases are very much in the minority.  I agree that
the current specification could be made clearer on this point, but on
the
other hand there are several ways to solve the problem, and it would not
arise for most clients any time soon.  I'll try to suggest a solution
for the problem, or be open for consideration of anyone else's
suggestion.

Note that Solicitation/Advertisement is not a transaction, and should
not require maintenance of state at the server.

Regards,
Charlie P.


