From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Wed, 19 Mar 2003 20:26:58 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Wed, 19 Mar 2003 20:26:58 -0500 Received: from c17870.thoms1.vic.optusnet.com.au ([210.49.248.224]:8599 "EHLO mail.kolivas.org") by vger.kernel.org with ESMTP id ; Wed, 19 Mar 2003 20:26:56 -0500 From: Con Kolivas To: Nick Piggin Subject: Re: [BENCHMARK] 2.5.65-mm2 with contest Date: Thu, 20 Mar 2003 12:37:50 +1100 User-Agent: KMail/1.5 Cc: linux kernel mailing list , Andrew Morton References: <200303201016.54818.kernel@kolivas.org> <3E791241.3070700@cyberone.com.au> In-Reply-To: <3E791241.3070700@cyberone.com.au> MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200303201237.50702.kernel@kolivas.org> Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 20 Mar 2003 11:58, Nick Piggin wrote: > Con Kolivas wrote: > >Contest results for mm2: > > Contest is starting to look good. Especially in > loads and lcpu. Very good point. Some of the loads are exhibiting better overall cpu usage and dropping the compile time (tar loads are a good example) > >no_load: > >Kernel [runs] Time CPU% Loads LCPU% Ratio > >2.5.65 3 80 95.0 0.0 0.0 1.00 > >2.5.65-mm1 3 79 94.9 0.0 0.0 1.00 > >2.5.65-mm2 3 79 94.9 0.0 0.0 1.00 > > AS is now on par with deadline here which is nice. Excellent. > [snip] > > >dbench_load: > >Kernel [runs] Time CPU% Loads LCPU% Ratio > >2.5.65 3 542 14.2 9.0 62.5 6.78 > >2.5.65-mm1 3 361 21.1 6.3 55.4 4.57 > >2.5.65-mm2 3 437 17.4 7.7 60.6 5.53 > > I don't know if this is a good balance shift or not. Maybe not > related to AS but I'll investigate. I dont think there are other tweaks in mm2 that could be responsible. It's extremely hard to know what exactly is the best balance in this load. If dbench 16 should get 80% of the total cpu usage (since the kernel compile has 4 processes) then mm2 is very close at 78%. However I didnt think dbench was supposed to be a particularly cpu bound task. The overall cpu usage is higher which is good though. > Thanks Con A pleasure, Con