To: rt@dla1.dl.ac.uk
Cc: P8021@nic.hep.net
Subject: ["Robin Tasker": Priority in 802.1p]
Date: Tue, 09 Apr 1996 14:23:14 -0700
From: Mick Seaman <mjs@NSD.3Com.COM>

Robin,
I take issue with your "almost certainly useless" comment below.

In fact we (802.1) discussed this quite extensively before starting
802.1p and made a couple of observations which I believe are key. The
issue here is not to carve bandwidth out of an oversubscribed service
in the bridges - the issue is to ensure that temporary execursions of
bridge queuing due to undifferentiated data background do not blow
away the timeliness of multimedia traffic. The normal operating point
in terms of load should be very very different to that experienced on
the MBONE.

Inside 3Com we have done a number of trials on the effectiveness of
these very "low-level" techniques on switched LAN segments combined
with router techniques for WAN hops - with very satisfactory results.
In the Maui meeting 802.1 had a presentation from Fouad Tobagi (Stanford)
on carrying traditional video over Ethernet. As a followup exercise we
(3Com) are engaged in some cooperative research work on the loading
levels on networks and the effectiveness of .1p solutions and
supporting technologies (PACE). We are not the only ones looking at
such work, though I should let others speak for themselves.

I believe some switches at key points of bandwidth convergence will
take the router role of being more intrusive into RSVP, but that
doesn't at all negate the considerable benefit we have seen in pure
level 2 switches.

Fouad's work already presented at the Maui meeting did indeed seek to
establish, with considerable sophistication how many flows might be
supported.

If you have a very low cost way of extending RSVP this would be a good
proposal for follow on work. However it seesm very clear that there
will be multivendor support for switches that provide simple priority
services and to only aim at "complete" RSVP solutions will miss the
point - of providing users with standards based rather than proprietary
choices. I don't believe we should go back and change the scope of .1p
at this stage of the game.

Mick

[attached: copy of P802.1-96/036 (see d96/d96n036.txt)]