From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S262698AbUBZFsL (ORCPT ); Thu, 26 Feb 2004 00:48:11 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262699AbUBZFsL (ORCPT ); Thu, 26 Feb 2004 00:48:11 -0500 Received: from sccrmhc13.comcast.net ([204.127.202.64]:44785 "EHLO sccrmhc13.comcast.net") by vger.kernel.org with ESMTP id S262698AbUBZFsJ (ORCPT ); Thu, 26 Feb 2004 00:48:09 -0500 Subject: Re: [RFC][PATCH] O(1) Entitlement Based Scheduler From: Albert Cahalan To: linux-kernel mailing list Cc: johnl@aurema.com Content-Type: text/plain Organization: Message-Id: <1077766232.10393.992.camel@cube> Mime-Version: 1.0 X-Mailer: Ximian Evolution 1.2.4 Date: 25 Feb 2004 22:30:32 -0500 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org John Lee writes: > The usage rates for each task are estimated using Kalman > filter techniques, the estimates being similar to those > obtained by taking a running average over twice the filter > _response half life_ (see below). However, Kalman filter > values are cheaper to compute and don't require the > maintenance of historical usage data. Linux dearly needs this. Please separate out this part of the patch and send it in. Right now, Linux does not report the recent CPU usage of a process. The UNIX standard requires that "ps" report this; right now ps substitutes CPU usage over the whole lifetime of a process. Both per-task and per-process (tid and tgid) numbers are needed. Both percent and permill (1/1000) units get reported, so don't convert to integer percent.