From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S269469AbUHZVAj (ORCPT ); Thu, 26 Aug 2004 17:00:39 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S269635AbUHZU7u (ORCPT ); Thu, 26 Aug 2004 16:59:50 -0400 Received: from e1.ny.us.ibm.com ([32.97.182.101]:34299 "EHLO e1.ny.us.ibm.com") by vger.kernel.org with ESMTP id S269657AbUHZU4y (ORCPT ); Thu, 26 Aug 2004 16:56:54 -0400 Date: Thu, 26 Aug 2004 13:55:36 -0700 From: "Martin J. Bligh" To: Con Kolivas , "Rafael J. Wysocki" cc: Andrew Morton , linux-kernel@vger.kernel.org, Rick Lindsley Subject: Re: 2.6.9-rc1-mm1 Message-ID: <52540000.1093553736@flay> In-Reply-To: <412E11ED.7040300@kolivas.org> References: <20040826014745.225d7a2c.akpm@osdl.org> <412DC47B.4000704@kolivas.org> <200408261636.06857.rjw@sisk.pl> <412E11ED.7040300@kolivas.org> X-Mailer: Mulberry/2.1.2 (Linux/x86) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Content-Disposition: inline Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org --On Friday, August 27, 2004 02:38:05 +1000 Con Kolivas wrote: > Rafael J. Wysocki wrote: >> On Thursday 26 of August 2004 13:07, Con Kolivas wrote: >> >>> Andrew Morton wrote: >>> >>>> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.9-rc1/2 >>>> .6.9-rc1-mm1/ >>>> >>>> >>>> - nicksched is still here. There has been very little feedback, except >>>> that it seems to slow some workloads on NUMA. >>> >>> That's because most people aren't interested in a new cpu scheduler for >>> 2.6. >> >> >> I am, but I have no benchmarks that give any useful numbers. > > That's because there are none for interactivity; you're simply > reinforcing my point. Rick's schedstats stuff had some ways to measure latency that seemed to work quite nicely. Hard to simulate exactly mozilla, email, etc, but probably close enough to be far more use than "ooh, it feels faster". He did a whole paper at OLS ... Rick ... pointer? >> Actually, with the current scheduler, updatedb really sucks. It's supposed to >> be a background task, but it hogs IO resources and memory like crazy >> (disclaimer: it's my personal subjective observation). > > The cpu scheduler plays almost no part in this. It's the I/O scheduler and the vm. IOnice will help the former _when it comes out_. Dropping the swappiness kind of helps the latter; although there are numerous alternative tweaks appearing for that too. Yup. I can open a large 8Mpixel camera image in "display" and hang the whole system for about 30s too ;-( M.