From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933702AbZIDK0T (ORCPT ); Fri, 4 Sep 2009 06:26:19 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S933514AbZIDK0R (ORCPT ); Fri, 4 Sep 2009 06:26:17 -0400 Received: from hera.kernel.org ([140.211.167.34]:48672 "EHLO hera.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933571AbZIDK0O (ORCPT ); Fri, 4 Sep 2009 06:26:14 -0400 Date: Fri, 4 Sep 2009 10:25:43 GMT From: tip-bot for Ingo Molnar Cc: linux-kernel@vger.kernel.org, ego@in.ibm.com, hpa@zytor.com, mingo@redhat.com, andreas.herrmann3@amd.com, a.p.zijlstra@chello.nl, balbir@in.ibm.com, tglx@linutronix.de, mingo@elte.hu Reply-To: balbir@in.ibm.com, mingo@redhat.com, hpa@zytor.com, ego@in.ibm.com, linux-kernel@vger.kernel.org, a.p.zijlstra@chello.nl, andreas.herrmann3@amd.com, tglx@linutronix.de, mingo@elte.hu In-Reply-To: References: To: linux-tip-commits@vger.kernel.org Subject: [tip:sched/balancing] sched: Turn on SD_WAKE_IDLE for 'close' domains Message-ID: Git-Commit-ID: e54c7073fae0683b90b3157e97142202c7334b79 X-Mailer: tip-git-log-daemon MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Disposition: inline X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.0 (hera.kernel.org [127.0.0.1]); Fri, 04 Sep 2009 10:25:43 +0000 (UTC) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Commit-ID: e54c7073fae0683b90b3157e97142202c7334b79 Gitweb: http://git.kernel.org/tip/e54c7073fae0683b90b3157e97142202c7334b79 Author: Ingo Molnar AuthorDate: Fri, 4 Sep 2009 11:38:07 +0200 Committer: Ingo Molnar CommitDate: Fri, 4 Sep 2009 12:17:06 +0200 sched: Turn on SD_WAKE_IDLE for 'close' domains Turn it on for the thread, core and cpu domain defaults. Same argument as for NEWIDLE: we want to err on the side of over-balancing, and mute the effect of that when it hurts a workload. On a NUMA box (4x4) i get better CPU utilization: Performance counter stats for './hackbench 10' (10 runs): Before: 2124.478000 task-clock-msecs # 12.196 CPUs ( +- 0.614% ) After: 2232.558805 task-clock-msecs # 12.601 CPUs ( +- 0.852% ) It's probably not worth turning it on for the NUMA domain though, as even on this fast-NUMA box it's break-even in terms of cost versus improvement: After #2: 2288.062034 task-clock-msecs # 12.909 CPUs ( +- 0.864% ) Acked-by: Peter Zijlstra Cc: Andreas Herrmann Cc: Andreas Herrmann Cc: Gautham R Shenoy Cc: Balbir Singh LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/topology.h | 6 +++--- 1 files changed, 3 insertions(+), 3 deletions(-) diff --git a/include/linux/topology.h b/include/linux/topology.h index 66774fd..f13ff3a 100644 --- a/include/linux/topology.h +++ b/include/linux/topology.h @@ -95,7 +95,7 @@ int arch_update_cpu_topology(void); | 1*SD_BALANCE_NEWIDLE \ | 1*SD_BALANCE_EXEC \ | 1*SD_BALANCE_FORK \ - | 0*SD_WAKE_IDLE \ + | 1*SD_WAKE_IDLE \ | 1*SD_WAKE_AFFINE \ | 1*SD_WAKE_BALANCE \ | 1*SD_SHARE_CPUPOWER \ @@ -129,7 +129,7 @@ int arch_update_cpu_topology(void); | 1*SD_BALANCE_NEWIDLE \ | 1*SD_BALANCE_EXEC \ | 1*SD_BALANCE_FORK \ - | 0*SD_WAKE_IDLE \ + | 1*SD_WAKE_IDLE \ | 1*SD_WAKE_AFFINE \ | 1*SD_WAKE_BALANCE \ | 0*SD_SHARE_CPUPOWER \ @@ -163,7 +163,7 @@ int arch_update_cpu_topology(void); | 1*SD_BALANCE_NEWIDLE \ | 1*SD_BALANCE_EXEC \ | 1*SD_BALANCE_FORK \ - | 0*SD_WAKE_IDLE \ + | 1*SD_WAKE_IDLE \ | 0*SD_WAKE_AFFINE \ | 1*SD_WAKE_BALANCE \ | 0*SD_SHARE_CPUPOWER \