From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Michael Chan" Subject: Re: Network latency regressions from 2.6.22 to 2.6.29 Date: Thu, 16 Apr 2009 14:57:17 -0700 Message-ID: <1239919037.12883.15.camel@HP1> References: <49E78A79.6050604@cosmosbay.com> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Cc: "Eric Dumazet" , "netdev@vger.kernel.org" To: "Christoph Lameter" Return-path: Received: from mms1.broadcom.com ([216.31.210.17]:3503 "EHLO mms1.broadcom.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751228AbZDPWDD (ORCPT ); Thu, 16 Apr 2009 18:03:03 -0400 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On Thu, 2009-04-16 at 12:55 -0700, Christoph Lameter wrote: > On Thu, 16 Apr 2009, Eric Dumazet wrote: > > > Is it a bnx2 or tg3 driver ? > > bnx2. > > > It would be sane to always set on both machines : > > ethtool -C eth0 rx-usecs 0 rx-frames 1 > > prior to your tests... > > rx-usecs is set to 18 by default by the driver and rx-frames is set to 6. > We saw some funky behavior with rx-usecs 0 so we left the defaults. We haven't changed these defaults since the beginning, but firmware updates between 2.6.22 and 2.6.29 can affect latency. Yes, Eric is right, using rx-frames 1 will give the best latency number. (rx-usecs won't matter if rx-frames is set to 1). > > But what does it matter? Just changing the kernel should not cause > increases in latency. > It will not matter if you use the same out-of-tree driver when switching kernels. If you use in-tree driver, the firmware may have changed between kernel versions.