Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757226Ab0DFCkz (ORCPT ); Mon, 5 Apr 2010 22:40:55 -0400 Received: from fg-out-1718.google.com ([72.14.220.153]:37096 "EHLO fg-out-1718.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757202Ab0DFCks convert rfc822-to-8bit (ORCPT ); Mon, 5 Apr 2010 22:40:48 -0400 MIME-Version: 1.0 In-Reply-To: <20100406021559.GD3630@balbir.in.ibm.com> References: <1270496004-9715-1-git-send-email-mike@android.com> <20100406021559.GD3630@balbir.in.ibm.com> Date: Mon, 5 Apr 2010 19:40:43 -0700 Message-ID: Subject: Re: [RFC][PATCH] sched: cpuacct: Track cpuusage per cpu frequency From: Mike Chan To: balbir@linux.vnet.ibm.com Cc: Paul Menage , cpufreq@vger.kernel.org, linux-kernel@vger.kernel.org Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 8BIT Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2111 Lines: 48 On Mon, Apr 5, 2010 at 7:15 PM, Balbir Singh wrote: > * menage@google.com [2010-04-05 12:52:57]: > >> On Mon, Apr 5, 2010 at 12:33 PM, Mike Chan wrote: >> > New file: cpuacct.cpufreq when CONFIG_CPU_FREQ_STATS is enabled. >> > >> > cpuacct.cpufreq accounts for cpu time per-cpu frequency, time is exported >> > in nano-seconds >> >> Can you clarify the wording of this (and describe it in the relevant >> Documentation/... file)? It's not clear. >> >> From the code, it appears that the file reports a breakdown of how >> much CPU time the cgroup has been consuming at each different CPU >> frequency level. If so, then you probably want to reword the >> description to avoid "per-cpu", since that makes it sounds as though >> it's reporting something, well, "per CPU". >> >> Also, what's the motivation here? If it's for power monitoring >> purposes, might it be simpler to just report a single number, that's >> the integral of the CPU usage by frequency index (i.e. calculated from >> the same information that this patch is already gathering in >> cpuacct_charge()) rather than dumping a whole table on userspace? > > As utilization increases, won't the integral quickly overflow? BTW, > Mike have you looked at the scaled accounting infrastructure we have > in taskstats? > I just looked (thanks for pointed that out, it was new to me, thx!). The problem here is that the accounting scales time based of cpu speed. As it currently stands it will not represent power tracking accurately as the power consumption of running at 1ghz / 2 != 500mhz. We would have to register a weight factor for each speed and those would have to be registered from the board files (at least in the ARM world), since voltage levels can vary across projects. -- Mike > -- > ? ? ? ?Three Cheers, > ? ? ? ?Balbir > -- 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/