From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1759625AbcDET3E (ORCPT ); Tue, 5 Apr 2016 15:29:04 -0400 Received: from mx0a-00082601.pphosted.com ([67.231.145.42]:58241 "EHLO mx0a-00082601.pphosted.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751511AbcDET3C (ORCPT ); Tue, 5 Apr 2016 15:29:02 -0400 Date: Tue, 5 Apr 2016 15:28:29 -0400 From: Chris Mason To: Bastien Bastien Philbert CC: Peter Zijlstra , Ingo Molnar , Matt Fleming , Mike Galbraith , Subject: Re: [PATCH RFC] select_idle_sibling experiments Message-ID: <20160405192829.vjv7z4xzpr64kcwe@floor.thefacebook.com> Mail-Followup-To: Chris Mason , Bastien Bastien Philbert , Peter Zijlstra , Ingo Molnar , Matt Fleming , Mike Galbraith , linux-kernel@vger.kernel.org References: <20160405180822.tjtyyc3qh4leflfj@floor.thefacebook.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.23.1 (2014-03-12) X-Originating-IP: [192.168.52.123] X-Proofpoint-Spam-Reason: safe X-FB-Internal: Safe X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10432:,, definitions=2016-04-05_12:,, signatures=0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tue, Apr 05, 2016 at 02:43:09PM -0400, Bastien Bastien Philbert wrote: > On Tue, Apr 5, 2016 at 2:08 PM, Chris Mason wrote: [ ... ] > > > > I tried a few variations on select_idle_sibling() that preserved the > > underlying goal of returning idle cores before idle SMT threads. They > > were all horrible in different ways, and none of them were fast. > > > > The patch below just makes select_idle_sibling pick the first idle > > thread it can find. When I ran it through production workloads here, it > > was faster than the patch we've been carrying around for the last few > > years. [ ... ] > > > Here is my concern, do you test this on standard scheduler workloads > or was this just written for Facebook's internal workloads. I am going > to test this later because frankly this may cause a regression on my > system which has only 4 cores so a idle CPU is probably less common > for a small amount of time. I am wondering however if Ingo has any > complains before I test this to see if it causes a regression or a bug > on my system. Ingo do you have any thoughts on this or would you like > me to just test this? Bastien Pretty much every commit to select_idle_sibling over the last few years was somehow trying to preserve or improve the select-idle-cores-first functionality I just ripped out. So, it's safe to assume it'll break something ;) -chris