Virtually Vista: real vs. virtual

ZoD submitted a comment on the previous Virtual Vista post about “what kind of results would Super PI get on the Host. I think it’s important to get a sense of what you’re losing(in terms of speed, not graphics) when virtualising Vista.” I don’t know how I forgot to test such a simple concept, but I’m making up for it.

I ran SuperPI on a real Windows Vista partition with exactly the same memory allocation size to give a side-by-side comparison of the speed difference between a virtualized (VMware Server) and real OS environment.

VMware Server with Windows VistaWindows Vista PI test

The real Vista started off 2 seconds faster than VMware and finished 14 seconds earlier, with just a 7% margin. Considering the benefits of security, ease and accessibility virtualization offers, this nothing short of amazing. Kudos to Keith and the VMware team!

Remember, this result was tested from unoptimized hardware configurations. I bet if I had a dedicated drive for the virtual machine and defragmented the virtual disk image, the result would have been even better!

4 insightful thoughts

  1. SuperPI seems to be CPU bound – it calculates the approximation of the pi, after all. The drive/defragmentation shouldn’t have any impact on its performance, assuming that no other activities are taking place in the background of the VM. The program reports 33 megs of allocated memory, so there should be no swapping.

    In other words, it’s not a representative piece of software for testing the virtualization.

  2. But SuperPI is used in many system performance benchmarks. I think it’s rather relevant for a virtual machine benchmark.

    If I were to test 3D, obviously there are none and graphics support on all three platforms vary way too much.

    If I were to test hard-drive, then I’d have to make sure each machine had a defragmented and equally sized hard drive. Not worth testing IMO.

    The only thing left reasonable to test was arithmetic calculations, thus, SuperPI.

  3. This might depend on the hardware.
    Is this running on a machine with a hypervisor?

    If the system doesn’t have Intel’s Vanderpool or AMD’s Pacifica how does Vmware perform?

Comments are closed.