Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932098Ab2JTTGW (ORCPT ); Sat, 20 Oct 2012 15:06:22 -0400 Received: from mail.betterlinux.com ([199.58.199.50]:54180 "EHLO mail.betterlinux.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751560Ab2JTTGV (ORCPT ); Sat, 20 Oct 2012 15:06:21 -0400 X-DKIM: OpenDKIM Filter v2.4.1 mail.betterlinux.com D6C3783302 From: Andrea Righi To: Paul Menage , Ingo Molnar , Peter Zijlstra Cc: linux-kernel@vger.kernel.org Subject: [PATCH v2 0/3] distinct load average per-cpuset Date: Sat, 20 Oct 2012 21:05:59 +0200 Message-Id: <1350759962-7092-1-git-send-email-andrea@betterlinux.com> X-Mailer: git-send-email 1.7.10.4 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2603 Lines: 65 Overview ~~~~~~~~ The cpusets subsystem allows to assign a different set of CPUs to a cgroup. A typical use case is to split large systems in small CPU/memory partitions and isolate certain users/applications in these subsets of the system. Sometimes, to have a quick overview of the state of each partition, we may be interested to get the load average of the CPUs assigned to a particular cpuset, rather than the global load average of the system. Proposed solution ~~~~~~~~~~~~~~~~~ The proposal is to add a new file in the cpuset subsystem to report the load average of the CPUs assinged to a particular cpuset cgroup. Example: # echo 0-1 > /sys/fs/cgroup/cpuset/foo/cpuset.cpus # echo 2-3 > /sys/fs/cgroup/cpuset/bar/cpuset.cpus # echo $$ > /sys/fs/cgroup/cpuset/foo/tasks # for i in `seq 4`; do yes > /dev/null & done ... after ~5mins ... # cat /proc/loadavg /sys/fs/cgroup/cpuset/{foo,bar}/cpuset.loadavg 3.99 2.66 1.24 6/377 2855 3.98 2.64 1.20 0.01 0.02 0.04 In this case we can easily find that the cpuset "foo" is the most busy in the system. ChangeLog v1->v2: - convert rq->nr_uninterruptible to a percpu variable - fix nr_uninterruptible accounting in the wakeup/sleep paths - use DEFINE_PER_CPU() instead of NR_CPUS arrays - in patch 2/3 add a comment to explain the validity of evaluating the cpuset load average as the sum of the individual per-cpu load averages [ Thanks to Peter Z. for the review and suggestions of v1 ] TODO: - report nr_running and nr_threads in cpuset.loadavg, producing the same output as /proc/loadavg; in this way we could do nice things like, for example, "mount --bind cpuset.loadavg /proc/loadavg" in a new mount namespace for a specific user that we want to isolate into a specific cpuset cgroup, etc... [PATCH v2 1/3] sched: introduce distinct per-cpu load average [PATCH v2 2/3] cpusets: add load avgerage interface [PATCH v2 3/3] cpusets: add documentation of the loadavg file Documentation/cgroups/cpusets.txt | 1 + include/linux/sched.h | 6 ++ kernel/cpuset.c | 108 +++++++++++++++++++++++++++++++++++ kernel/sched/core.c | 112 +++++++++++++++++++++++++++++++------ kernel/sched/debug.c | 3 +- kernel/sched/sched.h | 8 +-- 6 files changed, 214 insertions(+), 24 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/