From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Sandeen Subject: Re: ext4_fallocate Date: Wed, 27 Jun 2012 11:14:00 -0400 Message-ID: <4FEB2338.9080008@redhat.com> References: <4FE8086F.4070506@zoho.com> <20120625085159.GA18931@gmail.com> <20120625191744.GB9688@thunk.org> <4FE9B57F.4030704@redhat.com> <4FE9F9F4.7010804@zoho.com> <4FEA0DD1.8080403@gmail.com> <4FEA1415.8040809@redhat.com> <4FEA1F18.6010206@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Fredrick , Ric Wheeler , "Theodore Ts'o" , linux-ext4@vger.kernel.org, Andreas Dilger , wenqing.lz@taobao.com To: Ric Wheeler Return-path: Received: from mx1.redhat.com ([209.132.183.28]:25082 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757016Ab2F0POM (ORCPT ); Wed, 27 Jun 2012 11:14:12 -0400 In-Reply-To: <4FEA1F18.6010206@redhat.com> Sender: linux-ext4-owner@vger.kernel.org List-ID: On 6/26/12 4:44 PM, Eric Sandeen wrote: > On 6/26/12 3:57 PM, Eric Sandeen wrote: >> On 6/26/12 3:30 PM, Ric Wheeler wrote: ... >> I tried running this fio recipe on v3.3, which I think does a decent job of >> emulating the situation (fallocate 1G, do random 1M writes into it, with >> fsyncs after each): >> >> [test] >> filename=testfile >> rw=randwrite >> size=1g >> filesize=1g >> bs=1024k >> ioengine=sync >> fallocate=1 >> fsync=1 I realized the kernel I tested on had a lot of debugging bells & whistles turned on. For a more performance-configured v3.3 kernel, I see much less impact, more or less negligible: unwritten conversion on ext4: WRITE: io=1024.0MB, aggrb=23077KB/s, minb=23631KB/s, maxb=23631KB/s, mint=45437msec, maxt=45437msec WRITE: io=1024.0MB, aggrb=23040KB/s, minb=23593KB/s, maxb=23593KB/s, mint=45511msec, maxt=45511msec WRITE: io=1024.0MB, aggrb=23011KB/s, minb=23564KB/s, maxb=23564KB/s, mint=45567msec, maxt=45567msec overwrites on ext4: WRITE: io=1024.0MB, aggrb=23046KB/s, minb=23599KB/s, maxb=23599KB/s, mint=45499msec, maxt=45499msec WRITE: io=1024.0MB, aggrb=23222KB/s, minb=23780KB/s, maxb=23780KB/s, mint=45153msec, maxt=45153msec WRITE: io=1024.0MB, aggrb=23031KB/s, minb=23584KB/s, maxb=23584KB/s, mint=45528msec, maxt=45528msec So I guess it's interesting information; the debug kernel showed a much bigger difference; the performance-config'd kernel showed almost no difference. FWIW, on this same kernel, unwritten conversion on xfs: WRITE: io=1024.0MB, aggrb=28409KB/s, minb=29091KB/s, maxb=29091KB/s, mint=36909msec, maxt=36909msec WRITE: io=1024.0MB, aggrb=28372KB/s, minb=29053KB/s, maxb=29053KB/s, mint=36958msec, maxt=36958msec WRITE: io=1024.0MB, aggrb=28406KB/s, minb=29088KB/s, maxb=29088KB/s, mint=36913msec, maxt=36913msec overwrites on xfs: WRITE: io=1024.0MB, aggrb=28268KB/s, minb=28946KB/s, maxb=28946KB/s, mint=37094msec, maxt=37094msec WRITE: io=1024.0MB, aggrb=28316KB/s, minb=28995KB/s, maxb=28995KB/s, mint=37031msec, maxt=37031msec WRITE: io=1024.0MB, aggrb=28576KB/s, minb=29262KB/s, maxb=29262KB/s, mint=36694msec, maxt=36694msec -Eric