Re: [Qemu-devel] kqemu benchmarks versus VMWare 5.0

[Thomas Glanzmann]

Hi Matthew,

benchmarking integer and floating point operation in a virtual machine
like qemu or vmware doesn't make sense because they are both executed on
the real CPU without any modification of the original code. It should be
the same speed than on the native machine.

Interesting would be numbers where many page faults and context switches
involved. Also havy IO. The test that comes in my mind is compiling a
kernel. It would be interesting if you could paste the numbers for say a
kernel compilation of 2.2.8 with default config.

time ( tar xfz /path/to/kernel.tgz; cd kernel-version; make oldconfig; make 
bzImage )

Maybe you should meassure the time external because the virtual machine
can loose time ticks. But that would be a good start.

Greetings,
        Thomas