[libvirt] [PATCH] Canonicalize qemu machine types

[Mark McLoughlin]

On Thu, 2009-07-23 at 13:56 -0400, Cole Robinson wrote:
> On 07/23/2009 01:34 PM, Mark McLoughlin wrote:
> > Hey,
> >         Here's a series of patches to implement danpb's suggestion of
> > using 'qemu -M ?' to probe for valid machine types and then to build
> > on that for the machine type canonicalization.
> > 
> > Cheers,
> > Mark.
> > 
> 
> What's the speed difference of GetCapabilities after these patches? Not
> that it needs to be fast, but it would be useful to know if it's
> slowwwwwwww.

It's a fair point, especially since it wasn't long ago we discussed
this:

  http://www.redhat.com/archives/libvir-list/2009-April/msg00537.html

I took your little script, made it do 100 iterations, ran that 20 times
on F-11 and averaged the results:

  - Before this series of patches, it's averaging at 10ms, giving 100us
    per call

  - After this series with only qemu-system-x86 installed, it's  
    averaging at 1.7s, giving 17ms per call

  - With qemu-system-arm/mips/ppc/sparc and xenner installed, it's 
    averaging at 5.8s, giving 58ms per call

60ms vs 100us is a huge slowdown, relatively speaking. Is 60ms an
unacceptable worst-case got a GetCapabilities call? Probably not, but it
is certainly approaching unacceptable.

Cheers,
Mark.