From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755318AbZBZLDw (ORCPT ); Thu, 26 Feb 2009 06:03:52 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1752958AbZBZLDn (ORCPT ); Thu, 26 Feb 2009 06:03:43 -0500 Received: from gir.skynet.ie ([193.1.99.77]:48215 "EHLO gir.skynet.ie" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751832AbZBZLDm (ORCPT ); Thu, 26 Feb 2009 06:03:42 -0500 Date: Thu, 26 Feb 2009 11:03:36 +0000 From: Mel Gorman To: Lin Ming Cc: Linux Memory Management List , Pekka Enberg , Rik van Riel , KOSAKI Motohiro , Christoph Lameter , Johannes Weiner , Nick Piggin , Linux Kernel Mailing List , Zhang Yanmin , Peter Zijlstra Subject: Re: [RFC PATCH 00/19] Cleanup and optimise the page allocator V2 Message-ID: <20090226110336.GC32756@csn.ul.ie> References: <1235477835-14500-1-git-send-email-mel@csn.ul.ie> <1235639427.11390.11.camel@minggr> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Disposition: inline In-Reply-To: <1235639427.11390.11.camel@minggr> User-Agent: Mutt/1.5.17+20080114 (2008-01-14) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Feb 26, 2009 at 05:10:27PM +0800, Lin Ming wrote: > We tested this v2 patch series with 2.6.29-rc6 on different machines. > Wonderful, thanks. > 4P qual-core 2P qual-core 2P qual-core HT > tigerton stockley Nehalem > ------------------------------------------------ > tbench +3% +2% 0% Nice. > oltp -2% 0% 0% This is a big disappointment and somewhat confusing that it is so severe. For sysbench I was seeing on six different machines; 50834.14 51763.08 1.79% 61852.08 61966.58 0.18% 5935.98 5980.06 0.74% 29227.78 30167.72 3.12% 66702.67 66534.76 -0.25% 26643.18 26542.59 -0.38% So, two smallish regressions but mainly gains. Then again, I'm becoming more and more convinced that sysbench doesn't really represent a proper OLTP workload. I'd like to understand more how the page allocator at least was being used during your tests. Would it be possible to get a full profile (including instruction if possible and the vmlinux file) for both kernels please? If you can get the profiles, confirm the regression is still there as sometimes profiling can alter the outcome. Even if this happens, the profile will tell me where time is being spent. > aim7 0% 0% 0% > specjbb2005 +3% 0% 0% > hackbench 0% 0% 0% > > netperf: > TCP-S-112k 0% -1% 0% > TCP-S-64k 0% -1% +1% > TCP-RR-1 0% 0% +1% > UDP-U-4k -2% 0% -2% Pekka, for this test was SLUB or the page allocator handling the 4K allocations? > UDP-U-1k +3% 0% 0% > UDP-RR-1 0% 0% 0% > UDP-RR-512 -1% 0% +1% > > Lin Ming > Thanks a million for testing. -- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab