Return-Path: <tytso@thunk.org>
Received: from fort-point-station.mit.edu by po11.mit.edu (8.9.2/4.7) id MAA15212; Sat, 22 Sep 2001 12:54:28 -0400 (EDT)
Received: from snap.thunk.org (SNAP.THUNK.ORG [216.175.175.173])
	by fort-point-station.mit.edu (8.9.2/8.9.2) with ESMTP id MAA24823;
	Sat, 22 Sep 2001 12:54:24 -0400 (EDT)
Received: from think.thunk.org ([216.175.175.162])
	by snap.thunk.org with asmtp (Exim 3.32 #1 (Debian))
	id 15kq2p-0004KN-00; Sat, 22 Sep 2001 12:54:23 -0400
Received: from tytso by think.thunk.org with local (Exim 3.32 #1 (Debian))
	id 15kq2q-0002AP-00; Sat, 22 Sep 2001 12:54:24 -0400
Date: Sat, 22 Sep 2001 12:54:24 -0400
From: Theodore Tso <tytso@MIT.EDU>
To: "McNerney, Tim" <TMcNerney@chipwrights.com>
Cc: "'John Hawkinson'" <jhawk@mit.edu>, "'sipb@mit.edu'" <sipb@mit.edu>
Subject: Re: Who do I talk to about e19-rtr-backbone.mit.edu lossage?
Message-ID: <20010922125424.F853@thunk.org>
References: <93A4BBB9515ED411AE9D00508BD955A5198C20@intranet>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.3.15i
In-Reply-To: <93A4BBB9515ED411AE9D00508BD955A5198C20@intranet>; from TMcNerney@chipwrights.com on Fri, Sep 21, 2001 at 05:07:25PM -0400

On Fri, Sep 21, 2001 at 05:07:25PM -0400, McNerney, Tim wrote:
> 
> I may not be able to be articulate here, but this characterization came
> directly from a friend of mine who designs routers.  In a nutshell, TCP
> flow control was not designed for "modern" network usage (specifically web
> servers), it was designed to "ramp up" slowly, all while determining the
> best size of packets to send.  Now users place different demands on the net
> and this comes at the expense of low-bandwidth TCP connections.
>

OK, you're talking about something completely different here.  TCP
slow start is a function of the netowkring *protocol* as implemented
by clients; that is, TCP assumed large data flows, but the web causes
large numbers of short TCP connections., so TCP never has a chance to
open up its windows completely; but this penalizes the web clients ---
if anything, this actually makes things better for long-lived ssh
connections, since it slows down how many packets that are allowed to
be "in flight" which are caused by the web browers.

> Originally I was, but if the gateway into MIT isn't totally hosed,
> then some sort of backbone priority scheme might work.  My sysadmin
> friends at the AI lab tell me that it is the -backbone- that's
> hosed, so this might be a fertile ground for experimenting with
> better allocation schemes.

Again, you're assuming that the backbone is saturated; however, that's
*not* what was going on.

							- Ted
