From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752755Ab2AaAvi (ORCPT ); Mon, 30 Jan 2012 19:51:38 -0500 Received: from mga01.intel.com ([192.55.52.88]:34684 "EHLO mga01.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751003Ab2AaAvh (ORCPT ); Mon, 30 Jan 2012 19:51:37 -0500 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.71,315,1320652800"; d="scan'208";a="118293448" Subject: Re: [patch v2 0/8]block: An IOPS based ioscheduler From: Shaohua Li To: Vivek Goyal Cc: axboe@kernel.dk, linux-kernel@vger.kernel.org, david@fromorbit.com, jack@suse.cz, zhu.yanhai@gmail.com, namhyung.kim@lge.com In-Reply-To: <20120130153009.GB30245@redhat.com> References: <20120130070213.793690895@sli10-conroe.sh.intel.com> <20120130153009.GB30245@redhat.com> Content-Type: text/plain; charset="UTF-8" Date: Tue, 31 Jan 2012 08:50:40 +0800 Message-ID: <1327971040.21268.24.camel@sli10-conroe> Mime-Version: 1.0 X-Mailer: Evolution 2.32.2 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 2012-01-30 at 10:30 -0500, Vivek Goyal wrote: > On Mon, Jan 30, 2012 at 03:02:13PM +0800, Shaohua Li wrote: > > An IOPS based I/O scheduler > > > > Flash based storage has some different characteristics against rotate disk. > > 1. no I/O seek. > > 2. read and write I/O cost usually is much different. > > 3. Time which a request takes depends on request size. > > 4. High throughput and IOPS, low latency. > > Hi Shaohua, > > Last time we agreed that you will try to extend CFQ iops mode to take care > of this case. I was wondering that if that idea is out of the window? I felt there is complexity to do the merge and mess code even with wrapping some functions as you suggested. Another thing is I want to avoid complexity of CFQ, for example allocating queue and sharing it between tasks. cfq_set_request is one source that queue_lock gets contented, making the code simple can avoid such contention. > Also what's the real workload where this is going to benefit us. I had > struggled to run something which drove constantly deep queue depths to > get the fairness without idling. To be honest, I don't have real workload. Our test environment runs some micro benchmarks, which has some problems. I thought we all agreed CFQ has some limitations. Hoping somebody having real workload can have a look. Thanks, Shaohua