WG ACTION: 2 weeks to discuss [LL65] Need a requirement for multihomed hosts

Erik Guttman <[email protected]> Wed, 5 May 2004 01:19:49 +0200
Newsgroups gmane.ietf.zeroconf
Message-ID <a05200f0cbcbdce6ae4f2@[80.139.178.51]>
Please post discussion of this issue to the mailing list over the 
next two weeks
ending May 18, 2004.  In order to accept this issued, we will need a strong WG
consensus given that this is very late in the process.

Please see http://www.drizzle.org/~aboba/ZEROCONF/issues.html for a list of
current issues and their status.

[LL65]

Description of Issue:		Need a requirement for multihomed hosts
Submitter Name:                 Stuart Cheshire
Submitter Email Address:        [email protected]
Date first submitted:           04 May 04
Reference:			LL15, LL32
Comment Type ['t'ech|'e'dit]:	t
Prio ['S' Must|1 should|2 may]:	1
Section:			3.4
Rationale/Explanation:
Lengthy Description:

[Stuart]

>    If a host has two interfaces on the same network

The host may not *know* they are bridged onto the same link. This is the
real problem that can fool you.

I'd prefer a stronger requirement: If a host has more than one interface
(on the same network or not) then it MUST NOT attempt to confgure the
same LL address on more than one interface.

[Erik]

This is a pretty good technical suggestion, but its about 2 years late.
We have been over this ground many times.  The ultimate journey was
issue LL15.  Since then we have revised the text and moved it around
some - it no longer reads 'distinct for each interface', now its 'run
the algorithm independently for each interface.'

Note that this advice at the end of the first paragraph in section 3.4
would allow the algorithm to detect the conflict and resolve it, albeit
less efficiently than if the host had never created the problem in the
first place.

In issue LL32 we decided we would not attempt to solve problems arising
from use of IPv4 LL when used on more than one interface at a time.  For
this reason, it is inappropriate to include any MUST NOT language in
section 3.

[Stuart]

 From my implementation experience, I can tell you that any implementation
that tries to configure the same address on more than one interface will
run into problems.

Requested Change:

TEXT NEEDED