From: Eric Sandeen Subject: Re: Performance testing results Date: Wed, 27 Jun 2007 23:52:24 -0400 Message-ID: <46833078.3070609@redhat.com> References: <1692756695.20070628104652@mail.ru> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: valerie.clement@bull.net, linux-ext4@vger.kernel.org To: sftf Return-path: Received: from mx1.redhat.com ([66.187.233.31]:56809 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756751AbXF1DxA (ORCPT ); Wed, 27 Jun 2007 23:53:00 -0400 In-Reply-To: <1692756695.20070628104652@mail.ru> Sender: linux-ext4-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org sftf wrote: > Hi! > IMHO this test is not 100% correct. > ext3 | data=writeback > ext4 | data=writeback,extents,delalloc > xfs | defaults is ordered ! > > So you have compared ext's in writeback (which fastest mode) vs xfs in ordered. Actually xfs's default (only) mode is more like writeback than ordered. So that's a fair comparison. I bet it is xfs's barriers that hurt it, though - while they are pretty much required on a single disk with a volatile write cache, I think xfs's barrier implementation hurts it more than barriers for ext* -Eric