From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: UDP packet loss when running lsof Date: Tue, 22 May 2007 08:47:40 +0200 Message-ID: <4652920C.8020608@cosmosbay.com> References: <20070521113503.6bb70ae4.dada1@cosmosbay.com> <20070522021259.95ukifwz8kc4gccg@www.pochta.ru> <46528973.50809@cosmosbay.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: John Miller , netdev@vger.kernel.org To: Eric Dumazet Return-path: Received: from gw1.cosmosbay.com ([86.65.150.130]:43527 "EHLO gw1.cosmosbay.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757628AbXEVGrt (ORCPT ); Tue, 22 May 2007 02:47:49 -0400 In-Reply-To: <46528973.50809@cosmosbay.com> Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Eric Dumazet a =C3=A9crit : > John Miller a =C3=A9crit : >> Hi Eric, >> >>> I CCed netdev since this stuff is about network and not >>> lkml. >> >> Ok, dropped the CC... >> >>> What kind of machine do you have ? SMP or not ? >> >> It's a HP system with two dual core CPUs at 3GHz, the >> storage system is connected through QLogic FC-HBA. It should >> really be fast enough to handle a data stream of 50 MB/s... >=20 > Then you might try to bind network IRQ to one CPU > (echo 1 >/proc/irq/XX/smp_affinity) >=20 > XX being your NIC interrupt (cat /proc/interrupts to catch it) >=20 > and bind your user program to another cpu(s) >=20 > You might hit a cond_resched_softirq() bug that Ingo and others are=20 > sorting out right now. Using separate CPU for softirq handling and yo= ur=20 > programs should help a lot here. You might try this patch, now that Ingo "Signed-off-by" it. http://marc.info/?l=3Dlinux-kernel&m=3D117981607429875&w=3D2 I guess that with a correct softirq resched, no need to play with IRQ=20 affinities, unless you really want to push performance.