From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932223AbVHRNcG (ORCPT ); Thu, 18 Aug 2005 09:32:06 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932225AbVHRNcG (ORCPT ); Thu, 18 Aug 2005 09:32:06 -0400 Received: from mail02.syd.optusnet.com.au ([211.29.132.183]:45026 "EHLO mail02.syd.optusnet.com.au") by vger.kernel.org with ESMTP id S932223AbVHRNcF convert rfc822-to-8bit (ORCPT ); Thu, 18 Aug 2005 09:32:05 -0400 From: Con Kolivas To: Martin =?utf-8?q?MOKREJ=C5=A0?= Subject: Re: openafs is really faster on linux-2.4. than 2.6 Date: Thu, 18 Aug 2005 23:31:55 +1000 User-Agent: KMail/1.8.2 Cc: LKML References: <43032109.6030709@ribosome.natur.cuni.cz> <200508182257.35544.kernel@kolivas.org> <43048D81.80402@ribosome.natur.cuni.cz> In-Reply-To: <43048D81.80402@ribosome.natur.cuni.cz> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8BIT Content-Disposition: inline Message-Id: <200508182331.56217.kernel@kolivas.org> Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 18 Aug 2005 23:30, Martin MOKREJŠ wrote: > But that is very short and does not affect the interpretation here. Crap. The sync can take ages. > The throughput is clearly lower on 2.6 kernel and definitely the > CPU is in my eyes unnecessarily blocked... Why is the CPU in the > wait state instead of idle (this is teh problem on 2.6 series > but CPU is free on 2.4 series)? That's the main problem I think at the > moment. There is no wait state accounted for in 2.4 so you won't see it. Con > Con Kolivas wrote: > > On Thu, 18 Aug 2005 22:48, Martin MOKREJŠ wrote: > >>I think the problem here is outside afs. > >>Just doing this dd test but writing data directly to the ext2 > >>target gives same behaviour, i.e. on 2.4 kernel I see most of the > >>CPU idle but on 2.6 kernel all that CPU amount is shown as in > >>wait state. And the numbers from 2.4 kernel show higher throughput > >>compared to the 2.6 kernel (regardless the the PREEMPT or no PREEMPT > >>was used). > > > > Don't forget to include sync time.