From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1946465Ab2LFSNS (ORCPT ); Thu, 6 Dec 2012 13:13:18 -0500 Received: from mail-da0-f46.google.com ([209.85.210.46]:37489 "EHLO mail-da0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755259Ab2LFSNR (ORCPT ); Thu, 6 Dec 2012 13:13:17 -0500 Date: Thu, 6 Dec 2012 10:13:10 -0800 From: Tejun Heo To: Jeff Moyer Cc: Jens Axboe , "linux-kernel@vger.kernel.org" , "linux-mm@kvack.org" , Zach Brown , Peter Zijlstra , Ingo Subject: Re: [patch,v2] bdi: add a user-tunable cpu_list for the bdi flusher threads Message-ID: <20121206181310.GR19802@htj.dyndns.org> References: <50BE5988.3050501@fusionio.com> <50BE5C99.6070703@fusionio.com> <20121206180150.GQ19802@htj.dyndns.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hello, On Thu, Dec 06, 2012 at 01:08:18PM -0500, Jeff Moyer wrote: > > As for the original patch, I think it's a bit too much to expose to > > userland. It's probably a good idea to bind the flusher to the local > > node but do we really need to expose an interface to let userland > > control the affinity directly? Do we actually have a use case at > > hand? > > Yeah, folks pinning realtime processes to a particular cpu don't want > the flusher threads interfering with their latency. I don't have any > performance numbers on hand to convince you of the benefit, though. What I don't get is, RT tasks win over bdi flushers every time and I'm skeptical allowing bdi or not on a particular CPU would make a big difference on non-RT kernels anyway. If the use case calls for stricter isolation, there's isolcpus. While I can see why someone might think that they need something like this, I'm not sure it's actually something necessary. And, even if it's actually something necessary, I think we'll probably be better off with adding a mechanism to notify userland of new kthreads and let userland adjust affinity using the usual mechanism rather than adding dedicated knobs for each kthread users. Thanks. -- tejun