From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754272AbdDEIJd (ORCPT ); Wed, 5 Apr 2017 04:09:33 -0400 Received: from outbound-smtp10.blacknight.com ([46.22.139.15]:47858 "EHLO outbound-smtp10.blacknight.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754144AbdDEIJc (ORCPT ); Wed, 5 Apr 2017 04:09:32 -0400 Date: Wed, 5 Apr 2017 09:09:30 +0100 From: Mel Gorman To: Srikar Dronamraju Cc: Ingo Molnar , Peter Zijlstra , LKML , Rik van Riel Subject: Re: [PATCH] sched: Fix numabalancing to work with isolated cpus Message-ID: <20170405080930.vzz6wd7qfhvjckpb@techsingularity.net> References: <1491326848-5748-1-git-send-email-srikar@linux.vnet.ibm.com> <20170404203714.c4mexerfq3ztanpm@techsingularity.net> <20170405015006.GA4941@linux.vnet.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Disposition: inline In-Reply-To: <20170405015006.GA4941@linux.vnet.ibm.com> User-Agent: Mutt/1.6.2 (2016-07-01) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Apr 05, 2017 at 07:20:06AM +0530, Srikar Dronamraju wrote: > > > > > > To avoid this, please check for isolated cpus before choosing a target > > > cpu. > > > > > > > Hmm, would this also prevent a task running inside a cgroup that is > > allowed accessed to isolated CPUs from balancing? I severely doubt it > > Scheduler doesn't do any kind of load balancing for isolated cpus. > I was referring specifically to numa balancing, not load balancing but I should have been clearer. > > matters because if a process is isolated from interference then it > > follows that automatic NUMA balancing should not be involved. If > > Yes, as an extension of the above, numa balancing should not be > involved. > If anything, it arguably is a more sensible fix. If tasks on isolated CPUs should not be interfered with then that should include the NUMA scanner running in task context. IIf the PTEs are not updated then the faults are not incurred which would be a much larger saving in overhead overall. There would be a potential corner case where two threads in the same address space run in separate cpusets but it would be somewhat of an odd cornercase. -- Mel Gorman SUSE Labs