From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753435Ab1AYUDH (ORCPT ); Tue, 25 Jan 2011 15:03:07 -0500 Received: from mx1.redhat.com ([209.132.183.28]:15157 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751827Ab1AYUDF (ORCPT ); Tue, 25 Jan 2011 15:03:05 -0500 Subject: Re: [PATCH 16/16] KVM-GST: adjust scheduler cpu power From: Glauber Costa To: Peter Zijlstra Cc: kvm@vger.kernel.org, linux-kernel@vger.kernel.org, aliguori@us.ibm.com, Rik van Riel , Jeremy Fitzhardinge , Avi Kivity In-Reply-To: <1295898690.28776.472.camel@laptop> References: <1295892397-11354-1-git-send-email-glommer@redhat.com> <1295892397-11354-17-git-send-email-glommer@redhat.com> <1295893920.28776.468.camel@laptop> <1295895083.15920.9.camel@mothafucka.localdomain> <1295898690.28776.472.camel@laptop> Content-Type: text/plain; charset="UTF-8" Organization: Red Hat Date: Tue, 25 Jan 2011 18:02:36 -0200 Message-ID: <1295985756.15920.33.camel@mothafucka.localdomain> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 2011-01-24 at 20:51 +0100, Peter Zijlstra wrote: > On Mon, 2011-01-24 at 16:51 -0200, Glauber Costa wrote: > > > I would really much rather see you change update_rq_clock_task() and > > > subtract your ns resolution steal time from our wall-time, > > > update_rq_clock_task() already updates the cpu_power relative to the > > > remaining time available. > > > > But then we get into the problems we already discussed in previous > > submissions, which is, we don't really want to alter any notion of > > wallclock time. Could you list any more concrete advantages of the > > specific way you proposed? > > clock_task is the time spend on the task, by not taking steal time into > account all steal time is accounted as service to whatever task was > current when the vcpu wasn't running. > > It doesn't change wall-time in the sense of gtod, only the service time > to tasks. I fail to see how does clock_task influence cpu power. If we also have to touch clock_task for better accounting of other stuff, it is a separate story. But for cpu_power, I really fail. Please enlighten me. I did have slightly better results accounting the whole steal time period away from total_"ticks"(*) than just with units. * Please note again that ticks here is just a (bad, I admit) name. This is not tick accounting, is just an estimate of how much useful work your cpu did.