From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <eb4561603f9193a529b1076db2a1453d@plan9.bell-labs.com>
From: David Presotto <presotto@closedmind.org>
To: 9fans@cse.psu.edu
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Subject: [9fans] (no subject)
Date: Wed, 12 Feb 2003 10:34:36 -0500
Topicbox-Message-UUID: 592e9aaa-eacb-11e9-9e20-41e7f4b1d025

I updates kernel code on sources with window scaling.  The scaling is
automatic;

	if your interface > 100 mbps, you get a 512k window
	if your interface > 10 mbps, you get a 128k window
	otherwise, no scaling except what the other side requests

It seems to have a significant effect in our 1GB intel interfaces,
none elsewhere.  That implies that our bandwidth*rtt product is
sometimes > 64K.  Take it for what its worth.  It moved a single
TCP connection on our test machines from 28 MBps to 50 MBps.

If you have problems, tell me.