From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: Network latency regressions from 2.6.22 to 2.6.29 Date: Thu, 16 Apr 2009 21:29:02 +0200 Message-ID: <49E786FE.6090304@cosmosbay.com> References: <49E76906.2060205@hp.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Rick Jones , netdev@vger.kernel.org To: Christoph Lameter Return-path: Received: from gw1.cosmosbay.com ([212.99.114.194]:56906 "EHLO gw1.cosmosbay.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755391AbZDPT3s convert rfc822-to-8bit (ORCPT ); Thu, 16 Apr 2009 15:29:48 -0400 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: Christoph Lameter a =E9crit : > On Thu, 16 Apr 2009, Rick Jones wrote: >=20 >> Does udpping have a concept of service demand a la netperf? That co= uld help >> show how much was code bloat vs say some tweak to interrupt coalesci= ng >> parameters in the NIC/driver. >=20 > No. What does service on demand mean? The ping pong tests are very si= mple > back and forths without any streaming or overlay. > -- Would be good to measure a ping flood, because this one doesnt need wakeup of a user process on the remote side. $ ping -f 192.168.20.112 -c 10000 PING 192.168.20.112 (192.168.20.112) 56(84) bytes of data. --- 192.168.20.112 ping statistics --- 10000 packets transmitted, 10000 received, 0% packet loss, time 3337ms rtt min/avg/max/mdev =3D 0.072/0.184/20.011/1.389 ms, pipe 2, ipg/ewma = 0.333/0.078 ms and check if various kernel versions already give you different values = (for avg) When you perform your test, you upgrade kernels of both machines ?