From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753165Ab1GFHIG (ORCPT ); Wed, 6 Jul 2011 03:08:06 -0400 Received: from 173-166-109-252-newengland.hfc.comcastbusiness.net ([173.166.109.252]:36526 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752990Ab1GFHIE (ORCPT ); Wed, 6 Jul 2011 03:08:04 -0400 Date: Wed, 6 Jul 2011 03:07:56 -0400 From: Christoph Hellwig To: Dave Chinner Cc: T?r?k Edwin , xfs-masters@oss.sgi.com, Linux Kernel Mailing List , xfs@oss.sgi.com Subject: Re: XFS internal error (memory corruption) Message-ID: <20110706070756.GA25800@infradead.org> References: <4E12A927.9020102@gmail.com> <20110705130932.GF1026@dastard> <4E1313C1.6020309@gmail.com> <20110706040403.GW1026@dastard> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20110706040403.GW1026@dastard> User-Agent: Mutt/1.5.21 (2010-09-15) X-SRS-Rewrite: SMTP reverse-path rewritten from by bombadil.infradead.org See http://www.infradead.org/rpr.html Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Jul 06, 2011 at 02:04:03PM +1000, Dave Chinner wrote: > > Although is there supposed to be a performance benefit from having > > a separate log disk with XFS? > > There used to be. Now everyone just uses delayed logging, which is > far faster and more scalable that even using an external log. Even with delayed logging external logs are a huge benefit if you hit the log hard, e.g. for fsync intensive workloads. E.g. when using fs_mark in fsync mode it gives speedups over 100% for the setups I've tested. You'll see similar speedups for NFS server loads that are log force heavy as well. > > > IIRC it has a disadvantage that you can't use barriers properly. > > That mostly works now (recent kernels), but you take a hit in > journal IO waiting synchronously for the data device caches to be > flushed before writing to the log device. For metadata-heavy workloads where an external log benefits you most you generally just want to disable the volatile write cache anyway.