[Ovirt-devel] Appliance disk format

[Daniel P. Berrange]

On Tue, Feb 19, 2008 at 09:54:54PM -0800, David Lutterkort wrote:
> 
> On Tue, 2008-02-19 at 20:36 +0000, Daniel P. Berrange wrote:
> > To see if there's any further compression to be had for helping downloads
> > I tried various compression programs with the following results:
> > 
> >   - gzip  - 661M
> >   - bzip2 - 662M
> >   - p7zip - 617M
> >   - rzip  - 586M
> 
> Did you run these on the raw or compressed disk images ? I did a similar
> comparison a while ago with the surprising result that sometimes
> compressing a raw image gives _much_ smaller images than compressing an
> already compressed image. That, and the fact that VMWare can run raw
> disk images w/o much ado made me use raw disk images for virt-pack[1]

This was all done against  qcow2 images, with compression turned on.
I preferred this to raw, because it saves space at time of deployment
as well as download. eg 680 MB vs  2 GB (uncompressed qcow) vs 4 GB (raw)

> In more detail, I played with the Mantis appliance from JumpBox.
> Original is zip file of 138MB. Try with raw disk files and qcow
> compressed inside zip and tar:
> 
> raw -> qcow  69s (root disk) + 16s (data disk) = 85s
> 
> zip/raw       - miserable, truncates files at 4GB
> zip/qcow      156MB              15s
> tar.gz/raw    140MB              99s
> tar.bz2/raw   123MB             203s
> tar.bz2/qcow  156MB              54s
> tar.gz/qcow   156MB              14s
> 7za/raw        78MB           17m 3s
> 
> As you can see, you actually get better compression if you let
> gzip/bzip2/7zip loose on raw disk images. They can be quite a bit slower
> than qemu-img though, and, of course, you have to uncompress after
> download, not as convenient as qcow. I always imagined though that in a
> managed setting downloaded disk images would be copied into storage
> volumes when a VM is deployed so that uncompressing wouldn't really show
> up as a separate step.

Yes, i guess the question is what do we want the deployable disk format to
be. raw vs qcow2 ?  That in turn impacts the decision on compression needs
for re-distribution.

> [1] Patches I sent to et-mgmt-tools in Dec; I don't think I ever
> committed them.

Hmm, I remember those, but never got time to review that at the time. We
shoudl re-visit them.

Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=|