[Qemu-devel] Performance difference between QEMU 2.2.0 and 2.6.0

[Qemu-devel] Performance difference between QEMU 2.2.0 and 2.6.0

[Wenwen Wang]

Hello all,

I just found that there is a significant performance difference between
QEMU version 2.2.0 and 2.6.0.

Here are the execution times of the benchmark 458.sjeng from SPEC CINT2006
with test workload using user only emulation from ARM guest to x86-32 host:

qemu-2.2.0: 29 seconds
qemu-2.6.0: 38 seconds

I am sure that the evaluation environments are the same in the two
executions.

I also evaluated other benchmarks, it seems that they also have some
performance differences.

Could anyone tell me what this difference comes from or give me any hint?

Many thanks in advance!

Wenwen

Re: [Qemu-devel] Performance difference between QEMU 2.2.0 and 2.6.0

[Peter Maydell]

On 8 March 2017 at 18:17, Wenwen Wang <wenwang@cs.umn.edu> wrote:
> I just found that there is a significant performance difference between QEMU
> version 2.2.0 and 2.6.0.
>
> Here are the execution times of the benchmark 458.sjeng from SPEC CINT2006
> with test workload using user only emulation from ARM guest to x86-32 host:
>
> qemu-2.2.0: 29 seconds
> qemu-2.6.0: 38 seconds

2.6.0 is now quite old -- it would be interesting to test this on
the current git master.

> Could anyone tell me what this difference comes from or give me any hint?

There are lots of changes between those two releases, and we don't
in general track performance at all, so a regression is not very
surprising to me. If you wanted to try to identify what caused it
you could try a git bisect.

thanks
-- PMM

Re: [Qemu-devel] Performance difference between QEMU 2.2.0 and 2.6.0

[Wenwen Wang]

Hi Peter,

Thanks for your suggestions! Will test some newer versions.

Best,
Wenwen

On Wed, Mar 8, 2017 at 2:42 PM, Peter Maydell <peter.maydell@linaro.org>
wrote:

> On 8 March 2017 at 18:17, Wenwen Wang <wenwang@cs.umn.edu> wrote:
> > I just found that there is a significant performance difference between
> QEMU
> > version 2.2.0 and 2.6.0.
> >
> > Here are the execution times of the benchmark 458.sjeng from SPEC
> CINT2006
> > with test workload using user only emulation from ARM guest to x86-32
> host:
> >
> > qemu-2.2.0: 29 seconds
> > qemu-2.6.0: 38 seconds
>
> 2.6.0 is now quite old -- it would be interesting to test this on
> the current git master.
>
> > Could anyone tell me what this difference comes from or give me any hint?
>
> There are lots of changes between those two releases, and we don't
> in general track performance at all, so a regression is not very
> surprising to me. If you wanted to try to identify what caused it
> you could try a git bisect.
>
> thanks
> -- PMM
>