From: Theodore Tso Subject: Re: Enable asynchronous commits by default patch revoked? Date: Wed, 2 Sep 2009 11:03:06 -0400 Message-ID: <20090902150306.GF379@mit.edu> References: <200908241033.10527.Christian.Fischer@easterngraphics.com> <20090824133447.GH23677@mit.edu> <20090824183119.GI5931@webber.adilger.int> <20090824201027.GC17684@mit.edu> <4A92F7E0.9010001@redhat.com> <20090824220738.GG17684@mit.edu> <20090902144833.GA29651@zero> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-ext4@vger.kernel.org To: Tom Vier Return-path: Received: from thunk.org ([69.25.196.29]:35042 "EHLO thunker.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751974AbZIBPDG (ORCPT ); Wed, 2 Sep 2009 11:03:06 -0400 Content-Disposition: inline In-Reply-To: <20090902144833.GA29651@zero> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Wed, Sep 02, 2009 at 10:48:33AM -0400, Tom Vier wrote: > On Mon, Aug 24, 2009 at 06:07:39PM -0400, Theodore Tso wrote: > > Sure, agreed. But let's look a bit closer at what "async commit" > > really means. > > > > What ext3 and ext4 does by default is this: > > > > 1) Write data blocks required by data=ordered mode (if any) > > Shouldn't there be a write barrier after data blocks, so that the journal > blocks aren't written first? ie, mark the first journal block write with > barrier flag on. No, it doesn't matter, because the journal blocks are ignored until the commit block is written. So the order of the data blocks required by data=ordered mode, and the journal blocks are written can be freely reordered by the elevator and the hard drive without any risk of data integrity problems. - Ted