From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([208.118.235.92]:49492) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1TxEsq-0000u9-TF for qemu-devel@nongnu.org; Mon, 21 Jan 2013 05:45:47 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1TxEsn-0003HI-PE for qemu-devel@nongnu.org; Mon, 21 Jan 2013 05:45:44 -0500 Received: from mailpro.odiso.net ([89.248.209.98]:47290) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1TxEsn-0003H2-K5 for qemu-devel@nongnu.org; Mon, 21 Jan 2013 05:45:41 -0500 Date: Mon, 21 Jan 2013 11:45:35 +0100 (CET) From: Alexandre DERUMIER Message-ID: <005350a9-1990-41bb-86d3-d2a842df84a3@mailpro> In-Reply-To: <50FD19FC.4030209@redhat.com> Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [Qemu-devel] qmp drive-mirror vs qemu-img convert , drive-mirror really slow List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Paolo Bonzini Cc: qemu-devel >>Please try the blkmirror-job-1.4 branch from my github repo=C2=A0 >>(git://github.com/bonzini/qemu.git). That branch uses the qcow2 file's= =C2=A0 >>cluster size as the granularity, and has other optimizations that kick= =C2=A0 >>in when the image is sparse.=C2=A0 I'll try that, thanks Paolo, I'll keep you in touch. ----- Mail original ----- De: "Paolo Bonzini" =C3=80: "Alexandre DERUMIER" Cc: "qemu-devel" Envoy=C3=A9: Lundi 21 Janvier 2013 11:35:40 Objet: Re: qmp drive-mirror vs qemu-img convert , drive-mirror really slow = Il 21/01/2013 10:40, Alexandre DERUMIER ha scritto: > Hi list, > > I'm doing some speed benchmark between > > drive-mirror vs qemu-img convert , > > and drive-mirror seem to be a lot slower. > > > test: > > convert 1 32GB qcow2 file to raw file, local storage (6 disk 15k in raid1= 0 + raid BBU cache) > > time result: > > qemu-img : 30sec > drive-mirror (vm is paused, no writes) : 5min > > > Any idea why drive-mirror is so slow ? (maybe does it use directsync when= mirroring ?) No, it doesn't. Probably it's because the image is sparse? The current code in git master has a very coarse granularity (1 MB). Please try the blkmirror-job-1.4 branch from my github repo (git://github.com/bonzini/qemu.git). That branch uses the qcow2 file's cluster size as the granularity, and has other optimizations that kick in when the image is sparse. Paolo > I think both should have around same speed ? > > > Best Regards, > > Derumier Alexandre > > >