---------- Forwarded message ----------
Date: Tue, 19 May 1998 13:06:52 -0400
From: Greg Miller <gmiller@mci.net>
To: Fred Baker <fred@cisco.com>
Cc: Ran Atkinson <rja@corp.home.net>, end2end-interest@ISI.EDU
Subject: Re: Current use of RED?
Fred Baker writes:
>At 12:14 PM -0400 5/15/98, Greg Miller wrote:
>>I agree whole-heartedly. This is the very reason that I was uncomfortable
>>about the discussion supporting the recent changes of TCP's initial window
>>from 1 segment to 4. I don't wish to insult anyone doing simulation work, but
>>I have difficulty seeing the relevance of a simulation of 16 long-lived TCP
>>connections in evaluating the proposed change.
>
>That's the reason Mark Allman ran actual (non-simulated) TCP transfers over
>a real (non-simulated) internet to about 700 real (non-simulated) discard
>servers in the process of testing the notion out.
According to Mark's paper, he conducted TCP transfers to 100 distinct hosts.
Unless I'm misunderstanding the methodology, the tests were conducted
serially, so only a single connection with a larger initial window traversed
the internet at a given time. This experiment doesn't address my concern
about what will happen when Microsoft implements the larger initial window,
and suddenly a large percentage of the 200,000-or-so short TCP connections
passing through a single core router begin with 4 times more segments than
they used to.
This is probably a case where simulation is required, although it'd have to
be one enormous simulation . . .
Greg