From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from eggs.gnu.org ([140.186.70.92]:53233)
	by lists.gnu.org with esmtp (Exim 4.71)
	(envelope-from <kwolf@redhat.com>) id 1QZJBW-0005ol-Tu
	for qemu-devel@nongnu.org; Wed, 22 Jun 2011 04:53:20 -0400
Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71)
	(envelope-from <kwolf@redhat.com>) id 1QZJBV-00007y-8z
	for qemu-devel@nongnu.org; Wed, 22 Jun 2011 04:53:18 -0400
Received: from mx1.redhat.com ([209.132.183.28]:50769)
	by eggs.gnu.org with esmtp (Exim 4.71)
	(envelope-from <kwolf@redhat.com>) id 1QZJBU-00007R-Sz
	for qemu-devel@nongnu.org; Wed, 22 Jun 2011 04:53:17 -0400
Message-ID: <4E01AE27.1040407@redhat.com>
Date: Wed, 22 Jun 2011 10:56:07 +0200
From: Kevin Wolf <kwolf@redhat.com>
MIME-Version: 1.0
References: <20110621142926.7bc4df99@doriath>
In-Reply-To: <20110621142926.7bc4df99@doriath>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Subject: Re: [Qemu-devel] block: regression: savevm/delvm too slow
List-Id: <qemu-devel.nongnu.org>
List-Unsubscribe: <https://lists.nongnu.org/mailman/options/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=unsubscribe>
List-Archive: <http://lists.nongnu.org/archive/html/qemu-devel>
List-Post: <mailto:qemu-devel@nongnu.org>
List-Help: <mailto:qemu-devel-request@nongnu.org?subject=help>
List-Subscribe: <https://lists.nongnu.org/mailman/listinfo/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=subscribe>
To: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Stefan Hajnoczi <stefanha@gmail.com>, qemu-devel <qemu-devel@nongnu.org>

Am 21.06.2011 19:29, schrieb Luiz Capitulino:
> I'm getting the following times when doing a savevm and delvm in current
> HEAD eb47d7c5d (time in minutes, each time corresponds to a savevm/delvm run):
> 
>  savevm: 5:28m, 11:00m, 11:10m
>  delvm: 4:30m, 4:40m, > 15m
> 
> Now, trying with qemu 0.13.0 I get:
> 
>  savevm: < 1:00m, 4:00m, 4:34m
>  delvm: few seconds for all cases
> 
> Yes, you read it correctly, I tried with 0.13.0 because 0.14.0 also has the
> bug. This is the pattern I see when I run strace against HEAD while running
> the savevm command:
> 
>  pwrite(7, "\0\1\0\1\0\1\0\1\0\2\0\2\0\2\0\2\0\2\0\2\0\2\0\2\0\2\0\2\0\2\0\2"..., 65536, 196608) = 65536
>  fdatasync(7)                            = 0
> 
> Ie. a fdatasync() follows every single pwrite(). Something similar also happens
> with delvm. I don't see this pattern with 0.13.0.
> 
> The good news is that I've tracked it down and Mr. git bisect says that:
> 
> 29c1a7301af752de6721e031d31faa48887204bd is the first bad commit
> commit 29c1a7301af752de6721e031d31faa48887204bd
> Author: Kevin Wolf <kwolf@redhat.com>
> Date:   Mon Jan 10 17:17:28 2011 +0100
> 
>     qcow2: Use QcowCache
>     
>     Use the new functions of qcow2-cache.c for everything that works on refcount
>     block and L2 tables.
>     
>     Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> 
> :040000 040000 83e364185d37845bb27f1dccd1249d14cc7a9a1e 0c91964a52b5869333d4fb2cb0fa83104151359e M	block

Let me guess... You're using cache=writethrough?

Previously, qcow2_update_snapshot_refcount() had implemented its own
kind of writeback cache that would be used even with cache=writethrough.
Now we're using the generic Qcow2Cache, which implements a write-through
behaviour for cache=writethrough.

We could fix this by temporarily switching the cache to writeback mode.

Kevin