From: Eric Sandeen Subject: Re: [RFC PATCH] ext4: fix 50% disk write performance regression Date: Mon, 30 Aug 2010 14:35:53 -0500 Message-ID: <4C7C0819.8040505@redhat.com> References: <20100829231126.8d8b2086.billfink@mindspring.com> <4C7BE4DD.1060208@redhat.com> <20100830153025.302fa359.bill@wizard.sci.gsfc.nasa.gov> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Bill Fink , "tytso@mit.edu" , "adilger@sun.com" , "linux-ext4@vger.kernel.org" , "Fink, William E. (GSFC-6061)" To: Bill Fink Return-path: Received: from mx1.redhat.com ([209.132.183.28]:16554 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751969Ab0H3TgC (ORCPT ); Mon, 30 Aug 2010 15:36:02 -0400 In-Reply-To: <20100830153025.302fa359.bill@wizard.sci.gsfc.nasa.gov> Sender: linux-ext4-owner@vger.kernel.org List-ID: Bill Fink wrote: > On Mon, 30 Aug 2010, Eric Sandeen wrote: ... >> When I get some time (soon I hope) I'll look into the ramifications >> of this change (i.e. what if wbc->nr_to_write * 8 is more than the dirty >> pages, do things work out ok?) but it seems pretty reasonable. > > In thinking about that issue, my non-expert thought was that > if desired_nr_to_write was larger, it hopefully wouldn't be an > issue since presumably it's only going to actually write at most > the number of dirty pages anyway. And on the flip side, there > doesn't seem to be an issue with desired_nr_to_write possibly > being smaller that it was without the patch, since the empirical > evidence is that the performance significantly improved with the > patch, and the effect of a lower value would presumably be > reduced performance. > > -Bill Ted's suggestion of capturing blktrace data would be a really great step towards understanding what changed, too. Thanks! -Eric