References: <419A9151.2000508@atmos.washington.edu> <20041116163257.0e63031d@zqx3.pdx.osdl.net>
Message-ID: <cone.1100651833.776334.15267.502@pc.kolivas.org>
From: Con Kolivas <kernel@kolivas.org>
To: Stephen Hemminger <shemminger@osdl.org>
Cc: linux-kernel@vger.kernel.org
Subject: Re: Network slowdown from 2.6.7 to 2.6.9
Date: Wed, 17 Nov 2004 11:37:13 +1100
Mime-Version: 1.0
Content-Type: text/plain; format=flowed; charset="US-ASCII"
Content-Disposition: inline
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1455
Lines: 33

Stephen Hemminger writes:

> On Tue, 16 Nov 2004 15:46:25 -0800
> Harry Edmon <harry@atmos.washington.edu> wrote:
> 
>> I have a system that is running a program that receives and sends 
>> atmospheric data via TCP.  Most of the data is either in little packets 
>> (between 64 and 127 bytes) and large packets (between 1024 and 1517).  I 
>> am running this on a dual Xeon box (Tyan S2721-533 motherboard) with 2 
>> GB of memory and a Intel gigabit ethernet (82546EB).  I have been 
>> running this under 2.6.7.  When I switch to 2.6.9 on the same hardware, 
>> my network throughtput is cut by more than half.  All I can tell from 
>> looking at "netstat -s" is that my TCP resets are orders of magnitude 
>> higher under 2.6.9 than 2.6.7.  Enclosed is my 2.6.7 and 2.6.9 config 
>> files.  Anyone have any ideas where I should look to find the problem?
> 
> Do an OpenBSD or other firewall in the way that doesn't understand window
> scaling? OpenBSD pf doesn't correctly TCP window scaling so it ruins the
> throughput (typically 1/4 of expected).

Easiest way to see if this is responsible is to disable it and see if the 
throughput improves

echo 0 > /proc/sys/net/ipv4/tcp_window_scaling

Cheers,
Con

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/