From owner-freebsd-hackers  Mon May  8 07:58:58 1995
Return-Path: hackers-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id HAA02221
          for hackers-outgoing; Mon, 8 May 1995 07:58:58 -0700
Received: from thepoint.net (dg.thepoint.net [198.6.9.2])
          by freefall.cdrom.com (8.6.10/8.6.6) with SMTP id HAA02215
          for <freebsd-hackers@FreeBSD.org>; Mon, 8 May 1995 07:58:56 -0700
Received: by thepoint.net (5.4R3.00/)
	id AA01594; Mon, 8 May 1995 10:56:22 -0400
Date: Mon, 8 May 1995 10:56:22 -0400 (EDT)
From: Arlie Davis <arlie@thepoint.net>
Subject: Re: Help! My FreeBSD TCP stack is broken?
To: Robert Withrow <witr@rwwa.com>
Cc: freebsd-hackers@FreeBSD.org
In-Reply-To: <199505081433.KAA24568@rwwa.com>
Message-Id: <Pine.3.89.9505081050.A29919-0100000@dg.thepoint.net>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: hackers-owner@FreeBSD.org
Precedence: bulk


> I have had similar behavior: ICMP_ECHO indicated low latency, 
> TCP throughput almost nonexistant.  I thought it was caused
> by the BPF (because it happened when I compiled in BPF).  But
> it went away after I rebooted.  I'm glad this happened to
> you because in confirms my problem.  I'm sorry I can't help
> though...  (BTW, my problems were with 2.0R (with the one
> line patch to PPP).  My network link was PPP, but since
> you had the problem with ethernet, it seems to be somthing
> higher up than the IF code...

Several people pointed out the source of the problem: RFC1323, which is big 
TCP window support.  Turning this off (via "sysctl -w net.inet.tcp.rfc1323=0")
dramatically improves the system's useability.

Can someone explain why this happens?  Big TCP windows is a Good Thing, 
but why does it sometimes break?

-- arlie