From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932577AbXBOAK5 (ORCPT ); Wed, 14 Feb 2007 19:10:57 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932607AbXBOAK5 (ORCPT ); Wed, 14 Feb 2007 19:10:57 -0500 Received: from mx2.suse.de ([195.135.220.15]:58399 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932577AbXBOAK4 (ORCPT ); Wed, 14 Feb 2007 19:10:56 -0500 Date: Thu, 15 Feb 2007 01:10:53 +0100 From: Nick Piggin To: Christoph Lameter Cc: Linux Kernel Mailing List , Linux Memory Management List Subject: Re: [patch] mm: NUMA replicated pagecache Message-ID: <20070215001053.GB29797@wotan.suse.de> References: <20070213060924.GB20644@wotan.suse.de> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.9i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Feb 14, 2007 at 11:00:02AM -0800, Christoph Lameter wrote: > On Tue, 13 Feb 2007, Nick Piggin wrote: > > > This is a scheme for page replication replicates read-only pagecache pages > > opportunistically, at pagecache lookup time (at points where we know the > > page is being looked up for read only). > > The problem is that you may only have a single page table. One process > with multiple threads will just fault in one thread in order to > install the mapping to the page. The others threads may be running on > different nodes and different processors but will not generate any > faults. Pages will not be replicated as needed. The scheme only seems to > be working for special cases of multiple processes mapping the same file. Yeah like program text, libraries. Not just mapping, also reading, FWIW. I guess if you map something in different positions, then you can have it replicated! But no arguments, this doesn't aim to do replication of the same virtual address. If you did come up with such a scheme, however, you would still need a replicated pagecache for it as well.