Post-meeting status of draft-RFC

hplabs!sytek!kzm (Keith Mc Cloghrie) Fri, 16 February 1990 03:13 UTC

Received: from decwrl.dec.com by acetes.pa.dec.com (5.54.5/4.7.34) id AA10751; Thu, 15 Feb 90 19:13:29 PST
Received: by decwrl.dec.com; id AA29785; Thu, 15 Feb 90 19:13:26 -0800
Received: by hplabs.hpl.hp.com (15.11.1.3/15.5+IOS 3.14) id AA12183; Thu, 15 Feb 90 22:13:16 est
Received: from sytek with uucp; Thu, 15 Feb 90 18:39:50
Received: by sytek.hls.hac.com (5.51/5.17) id AA04735; Thu, 15 Feb 90 18:39:50 PST
From: hplabs!sytek!kzm (Keith Mc Cloghrie)
Message-Id: <9002160239.AA04735@sytek.hls.hac.com>
Subject: Post-meeting status of draft-RFC
To: mogul (Jeffrey Mogul)
Date: Thu, 15 Feb 90 18:39:46 PDT
Cc: hplabs!mtudwg
In-Reply-To: <9002142051.AA01873@acetes.pa.dec.com>; from "Jeffrey Mogul" at Feb 14, 90 1251
X-Mailer: ELM [version 2.2 PL0]

Jeff,

In talking to others since the PMTU-Discovery Working Group
meeting last week, I've discovered that not everybody was
left with the same impressions.

The whole question of which method is best seems to me a matter 
of choosing the right trade-off: how much fragmentation is 
avoided v. the amount of processing/state information in various 
places.  I certainly like the fact that using the RF bit avoids 
having to keep state at the destination.  But none of its variants 
that I've heard so far are as good at avoiding fragmentation as 
the proposal in the draft RFC handed out at the meeting.

If the bit is made available for PMTU-discovery, then I look 
forward to seeing how good a proposal that Van/Steve/Mike can 
come up with.  (Note that there are a number of issues that they
still need to resolve; for example, to decide on a method of
determining whether the destination is prepared to send 
Frag-Report ICMP messages; the discussion at the meeting didn't
resolve this and didn't address UDP either.  It occurs to me that
they could send an IP option to ask the destination explicitly, but
I'm sure they'll find something more creative, which will also
cover the possibility of packet loss/etc.).

On the other hand, if the bit is still not available (or if Van, 
Steve & Mike can't find the time to write up their proposal), then 
as I said at the beginning of the meeting last week, I see the 
draft-RFC as the combination which produces the least fragmentation. 
I believe it's worthwhile to evaluate the removal of capabilities 
from this combination, to determine whether such removal results in
a gain in simplification which outweighs the corresponding increase 
in fragmentation.

Keith.