From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ns.suse.de ([195.135.220.2]:38341 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S965977AbXDGWTF (ORCPT ); Sat, 7 Apr 2007 18:19:05 -0400 From: Andi Kleen Subject: Re: [PATCH 1/4] x86_64: (SPARSE_VIRTUAL doubles sparsemem speed) Date: Sun, 8 Apr 2007 00:18:44 +0200 References: <20070401071024.23757.4113.sendpatchset@schroedinger.engr.sgi.com> In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200704080018.45084.ak@suse.de> Sender: linux-arch-owner@vger.kernel.org To: Christoph Lameter Cc: Andy Whitcroft , Martin Bligh , Dave Hansen , linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org, linux-mm@kvack.org, KAMEZAWA Hiroyuki List-ID: On Sunday 08 April 2007 00:06:13 Christoph Lameter wrote: > Results: > > x86_64 boot with virtual memmap > > Format: #events totaltime (min/avg/max) > > kfree_virt_to_page 598430 5.6ms(3ns/9ns/322ns) > > x86_64 boot regular sparsemem > > kfree_virt_to_page 596360 10.5ms(4ns/18ns/28.7us) > > > On average sparsemem virtual takes half the time than of sparsemem. Nice. But on what workloads? Anyways it looks promising. I hope we can just replace old style sparsemem support with this for x86-64. > Time is measured using the cycle counter (TSC on IA32, ITC on IA64) which has > a very low latency. Sorry that triggered my usual RDTSC rant... Not on NetBurst (hundred of cycles) And on the others (C2,K8) it is a bit dangerous to measure short code blocks because RDTSC is not guaranteed ordered with the surrounding instructions. -Andi