Virtual Machines comparison - strange results

[oemsuite]

Hi,
Firstly sorry for my English

I'm writing a research-based work about virtual machines.
I have a strange results with comparison of virtual machines( Vmware, Virtual Box, Virtual PC). The disk mark on the host system is lower than virtual. I don' t have an idea why? Can anyone help me?
Thank you.

[David (PassMark)]

Virtual machine cache differently (and in some instances cache disk access when asked not to). If you use the advanced disk test and test with file too large to be cached then you should find the VM is no faster than the host.

The other possible reaon is that the VM might be acting like a compressed volume. So when read / writing data that compresses well the I/O might appear faster.

[TarikB]

I have done quite a lot of benchmarking with virtual machines running on Type 2 hypervisors. I have not seen the disk tests give values higher than for the host OS. It looks from your post that you are also testing Type-2 Hypervisor products. This may, of course, be related to the disk caching already mentioned, though I am slightly concerned that I have not observed these results. What is the underlying physical/logical disk? What size of disk is the VM using?

[oemsuite]

Hi these is my results:
http://img532.imageshack.us/img532/9154/nowyobrazmapybitowej4.jpg

I set virtual disks always the same 12 GB single file.

Thank you for help

[TarikB]

I agree that VPC gives good results on the sequential read and write. I have found VMware to give better results for random seek/rw than you are getting.

Are you using a fixed sized disk in VMware or a growing one?What size are the volumes of the disk used in host, vmware and vpc respectively?

[oemsuite]

I used fixed disk all the time. On the host I have 250 gb , on the guests - 12 GB. Maybe that is the problem?

[TarikB]

This is my mistake, everything looks as I would expect. VMWare random read/write performance is definitely just a little under the disk performance of the Host while VPC is significantly greater.

I also noticed a very high deviation from the mean of the VPC results in my dataset; I'm thinking that VPC is doing quite a bit of caching.