I have just created a new virtual machine in Workstation 9.0.1. Both Host and Guest OS is Windows 7 x64. In the guest I ran the Windows Experience Index evaluation and noticed some weird changes (you can find the WEI for both host and guest attached to this post).
1) Hard drive performance in the guest increased from 5.9 to 6.9 - How is that even possible?
2) Memory operations per second in the guest dropped from 7.5 to 4.5 - what could be the reason for such a significant memory performance drop?
3) CPU performance in the guest dropped from 7.3 to 6.4, but I guess this is to be expected.
CPU Intel Core i5-750 2.66 GHz
8 GB RAM
nVidia GeForce GTX 260
HDD: 3xWD Caviar Black (1x750GB, 2x1TB) - I am aware of this being a bottleneck of my host
Hardware compatibility: WS9.0
1 CPU, 2 Cores per CPU
Preferred mode: Intel VT-x/EPT, also checked Virtualize Intel VT-x/EPT
RAM 1536 MB
HDD: SCSI 64 GB, single file, not preallocated
Checked Disable Memory Page Trimming
Can you please explain the above changes in WEI and suggest how to improve especially the memory performance in the guest?
Thanks in advance.
As for a primary hard disk performance, originally I used a sparse disk, but as soon as I have changed it to preallocated, the WEI reported inside guest is now the same as of host.
Still, I would like to have an explanation for the memory performance drop reported inside the guest. Anyone, please?
I have been playing with various guest memory sizes and came to a conclusion that the description of WEI for memory performace is a bit misleading. I thought that the only criteria for the memory performance would be a number of operations per second, but it looks like the WEI scoring algorithm takes absolute memory size into account as well. Here are the WEI values for different guest memory sizes (no other changes have been done to the guest VM):
Guest VM RAM Size (WEI Score)
1.5 GB (4.5)
2 GB (5.5)
4 GB (7.5)
So, I guess the above answers my question.
That is interesting information.
One of the problems with comparing WEI between host and VMs (or even just between different VMs or virtualization platforms) is that the measurement is relatively opaque: We don't know the entirety of the information goes into computing the WEI or how it affects the outcome. More importantly, we don't know if virtualization overheads can be reflected accurately in the result, since Workstation does all sorts of interesting stuff to try to minimize the visibility of those overheads to the guest (i.e. by distorting the guest's perspective of time when the virtualization overhead is high)... so if the guest is timing particular operations, the numbers might not reflect any actual aspect of the VM's real-world performance! Hard drive performance can suffer due to host overhead and sparse-disk overhead, but it can be boosted by host OS disk caching far beyond the caching capabilities of the physical hard drive.
In summary: I would not put any significant weight in the WEI measurements. Instead, tune your VM's parameters to meet the real-world needs of the workload you are running inside it. Using the WEI to guide VM configuration will quite possibly give non-ideal results.