From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S262085AbUHJId6 (ORCPT ); Tue, 10 Aug 2004 04:33:58 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262114AbUHJId6 (ORCPT ); Tue, 10 Aug 2004 04:33:58 -0400 Received: from e34.co.us.ibm.com ([32.97.110.132]:2438 "EHLO e34.co.us.ibm.com") by vger.kernel.org with ESMTP id S262085AbUHJIdn (ORCPT ); Tue, 10 Aug 2004 04:33:43 -0400 Message-Id: <200408100834.i7A8Y7e06476@owlet.beaverton.ibm.com> To: Andrew Morton cc: linux-kernel@vger.kernel.org, mjbligh@us.ibm.com Subject: Re: 2.6.8-rc3-mm2 In-reply-to: Your message of "Sun, 08 Aug 2004 15:29:36 PDT." <20040808152936.1ce2eab8.akpm@osdl.org> Date: Tue, 10 Aug 2004 01:34:06 -0700 From: Rick Lindsley Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org - Added a little patch to the CPU scheduler which disables its array switching. This is purely experimental and will cause high-priority tasks to starve lower-priority tasks indefinitely. It is here to determine whether it is this aspect of the scheduler which caused the staircase scheduler to exhibit improved throughput in some tests on NUMAq. Here's the results of tests with sdet. The summary is that yes, we did reach the peak we saw with the staircase scheduler (2.6.8-rc2-mm2). Dropping the expired array does make a difference. But it looks like that wasn't all of it, though, because we see it fails to reach the same scores in all *but* the peak run. Schedstats helps show a couple of interesting things but no smoking gun. rc3-mm2 both tried to move around more tasks (calls to load_balance()), and succeeded in finding some to move (calls to pull_task()). But not a LOT more, in either case. And rc3-mm2 took 15% longer to run, even if it did as well in the end. Graphs are at http://eaglet.rain.com/rick/linux/staircase/scase-vs-noscase-vs-1q.html Overall, I'd say I still like the staircase patch as a whole. Not only did it increase the benchmark numbers, but I like the simplicity it (re)introduces for interactive bonus calculations. Rick DISCLAIMER: SPEC(tm) and the benchmark name SDET(tm) are registered trademarks of the Standard Performance Evaluation Corporation. This benchmarking was performed for research purposes only, and the run results are non-compliant and not-comparable with any published results. Sdet Scripts 1 4 16 64 2.6.8-rc2-mm2 32.46% 58.58% 100.00% 74.20% 2.6.8-rc2-mm2-noscase 23.62% 43.95% 90.92% 74.16% 2.6.8-rc3-mm2 16.44% 43.26% 102.95% 71.26%