Re: [Qemu-devel] [PATCH 0/3] Add SCSI support for PC target

["Daniel P. Berrange" <berrange@redhat.com>]

On Wed, Oct 31, 2007 at 02:56:53PM +0200, Dan Kenigsberg wrote:
> On Wed, Oct 31, 2007 at 11:17:23AM +0100, Laurent Vivier wrote:
> > Dan Kenigsberg a écrit :
> > >Hi,
> > >
> > >I have some newby's questions: is a speedup expected, comparing to ide
> > 
> > We should have a speedup.
> 
> Would you suggest where it could show? I was trying simple dd and
> bonnie++, and saw big variace between consequtive runs, but little
> change between mounting an image with if=ide and if=scsi. Was I doing
> something wrong?

Using dd is pretty useless as a benchmark. bonnie++ can work sometimes
if you tune the size of its I/O ops wrt to the amount of RAM in your
host and guest. The big variance is likely caused by doing small enough
I/O that the cache in either host or guest can buffer your data at times. 
IOZone is a benchmark I find gives better results, though it does take a 
hell of a long time to run. Tune IOZone params to do increasingly large
I/O chunks, so that it will exceed RAM in both host & guest & let it
repeat many times & you should get to some stable benchmark results.

There were some results at the last Xen summit comparing the IDE & SCSI
drivers in QEMU(KVM) against Xen paravirt - the SCSI results were very
impressive & significantly ahead of IDE.

http://xen.org/files/xensummit_4/xen_summit_2007_spring_hvm_charts_Harper.pdf

Regards,
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  -=|