From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S261464AbVGFFyk (ORCPT ); Wed, 6 Jul 2005 01:54:40 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262115AbVGFFxg (ORCPT ); Wed, 6 Jul 2005 01:53:36 -0400 Received: from e31.co.us.ibm.com ([32.97.110.129]:41431 "EHLO e31.co.us.ibm.com") by vger.kernel.org with ESMTP id S261464AbVGFEVn (ORCPT ); Wed, 6 Jul 2005 00:21:43 -0400 Date: Wed, 6 Jul 2005 10:00:56 +0530 From: Suparna Bhattacharya To: Benjamin LaHaise Cc: linux-aio@kvack.org, linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: aio-stress throughput regressions from 2.6.11 to 2.6.12 Message-ID: <20050706043056.GA4223@in.ibm.com> Reply-To: suparna@in.ibm.com References: <20050701075600.GC4625@in.ibm.com> <20050701142555.GB31989@kvack.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20050701142555.GB31989@kvack.org> User-Agent: Mutt/1.4i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Fri, Jul 01, 2005 at 10:25:55AM -0400, Benjamin LaHaise wrote: > On Fri, Jul 01, 2005 at 01:26:00PM +0530, Suparna Bhattacharya wrote: > > On one test system I see a degradation from around 17+ MB/s to 11MB/s > > for random O_DIRECT AIO (aio-stress -o3 testext3/rwfile5) from 2.6.11 > > to 2.6.12. It doesn't seem filesystem specific. Not good :( > > What sort of io subsystem does it have? Also, does changing the aic7xxx. > elevator make any difference? I'm away for the long weekend, but will I tried switching to the noop elevator - the regression was still there. > help look into this next week. Do you see the regression as well, or is it just me ? Regards Suparna > > -ben -- Suparna Bhattacharya (suparna@in.ibm.com) Linux Technology Center IBM Software Lab, India