Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757846AbXESBa1 (ORCPT ); Fri, 18 May 2007 21:30:27 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1754303AbXESBaT (ORCPT ); Fri, 18 May 2007 21:30:19 -0400 Received: from cantor.suse.de ([195.135.220.2]:40989 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753170AbXESBaS (ORCPT ); Fri, 18 May 2007 21:30:18 -0400 Date: Sat, 19 May 2007 03:30:13 +0200 From: Nick Piggin To: David Howells Cc: Linux Kernel Mailing List , Linux Memory Management List , linux-arch@vger.kernel.org Subject: Re: [rfc] increase struct page size?! Message-ID: <20070519013013.GC15569@wotan.suse.de> References: <20070518040854.GA15654@wotan.suse.de> <7554.1179481350@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <7554.1179481350@redhat.com> User-Agent: Mutt/1.5.9i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1464 Lines: 31 On Fri, May 18, 2007 at 10:42:30AM +0100, David Howells wrote: > Nick Piggin wrote: > > > I'd like to be the first to propose an increase to the size of struct page > > just for the sake of increasing it! > > Heh. I'm surprised you haven't got more adverse reactions. > > > If we add 8 bytes to struct page on 64-bit machines, it becomes 64 bytes, > > which is quite a nice number for cache purposes. > > Whilst that's true, if you have to deal with a run of contiguous page structs > (eg: the page allocator, perhaps) it's actually less efficient because it > takes more cache to do it. But, hey, it's a compromise whatever. > > In the scheme of things, if we're mostly dealing with individual page structs > (as I think we are), then yes, I think it's probably a good thing to do - > especially with larger page sizes. Yeah, we would end up eating about 12.5% more cachelines for contiguous runs of pages... but that only kicks in after we've touched 8 of them I think, and by that point the accesses should be very prefetchable. I think the average of 75% more cachelines touched for random accesses is going to outweigh the contiguous batch savings, but that's just a guess at this point. - 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/