Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751042AbXACSdY (ORCPT ); Wed, 3 Jan 2007 13:33:24 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751044AbXACSdY (ORCPT ); Wed, 3 Jan 2007 13:33:24 -0500 Received: from gate.crashing.org ([63.228.1.57]:49744 "EHLO gate.crashing.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751030AbXACSdX (ORCPT ); Wed, 3 Jan 2007 13:33:23 -0500 In-Reply-To: <20070102235733.GG18033@redhat.com> References: <20070102235733.GG18033@redhat.com> Mime-Version: 1.0 (Apple Message framework v623) Content-Type: text/plain; charset=US-ASCII; format=flowed Message-Id: Content-Transfer-Encoding: 7bit Cc: Linux Kernel , mingo@redhat.com From: Segher Boessenkool Subject: Re: inaccurate migration cost calculation? Date: Wed, 3 Jan 2007 19:33:52 +0100 To: Dave Jones X-Mailer: Apple Mail (2.623) Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 843 Lines: 23 > Across different boots using the same 2.6.19 kernel on a quad-core xeon > I see huge variance in the migration_cost being reported during boot. > > -migration_cost=39,3940 > +migration_cost=25,4941 > > This CPU has a very large cache which could be key here... > L1 Instruction cache: 32KB, 8-way associative. 64 byte line size. > L1 Data cache: 32KB, 8-way associative. 64 byte line size. > L3 unified cache: 4MB, 16-way associative. 64 byte line size. That last cache is shared between CPU cores, which leads to big non-determinism -- not sure if there's much you can do about it. Segher - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/