PD4 : Information on Impact of Optimization Settings

PD4 has three optimization options for a VM.

Adaptive Hypervisor
Optimize Performance for
Optimize Power Consumption for

I wanted to find out the impact of these options on CPU usage during idle conditions so I did some testing. Here are the details.

Systems

System A - Early 2008 Mac Pro, 2 x 2.8 GHz quad core (8 cores total), 10 GB RAM, OS X 10.5.5
System B - Late 2007 Mac Book Pro 15", 1 x 2.2 Ghz dual core (2 cores total), 4GB RAM, OS X 10.5.5

VM

Freshly built, Windows XP Pro with SP3 and all updates minus IE7.

The VM was built on System A with a configuration of 2 processors, 1 GB RAM.

The Windows "Computer" Driver was 'ACPI Multiprocessor PC' due to how the VM was created.

Tests

The tests involved setting the number of processors, enabling/disabling the adaptive hypervisor, and switching the optimization settings for virtual machine vs. OS X.

Each test involved modifying the VM configuration settings, booting the VM and then letting it settle for 10 minutes. Once settled, the CPU usage of both OS X and Windows were measured.

The OS X CPU usage was monitored using iSlayer's iStat Menus tool and then reading % allocated to prl_vm_app.

The Windows CPU usage was monitored using Windows Task Manager inside the VM and then reading the % System Idle.

Test 1 - 1 Processor, Adaptive Hypervisor Disabled, Optimized for Virtual Machine / Better Performance

System A - OS - 100% of 800% used / Windows - 50% idle
System B - OS - 12% of 200% used / Windows - 97% idle

Test 2 - 1 Processor, Adaptive Hypervisor Enabled, Optimized for Virtual Machine / Better Performance

System A - OS - 100% of 800% used / Windows - 50% idle
System B - OS - 12% of 200% used / Windows - 97% idle

Test 3 - 2 Processors, Adaptive Hypervisor Disabled, Optimized for Virtual Machine / Better Performance

System A - OS - 10% of 800% used / Windows - 97% idle
System B - OS - 15% of 200% used / Windows - 97% idle

Test 4 - 2 Processors, Adaptive Hypervisor Enabled, Optimized for Virtual Machine / Better Performance

System A - OS - 10% of 800% used / Windows - 97% idle
System B - OS - 15% of 200% used / Windows - 97% idle

Test 5 - 3 Processors, Adaptive Hypervisor Disabled, Optimized for Virtual Machine / Better Performance

System A - OS - 10% of 800% used / Windows - 99% idle
System B - Not Applicable

Test 6 - 3 Processors, Adaptive Hypervisor Enabled, Optimized for Virtual Machine / Better Performance

System A - OS - 10% of 800% used / Windows - 99% idle
System B - Not Applicable

Test 5 - 4 Processors, Adaptive Hypervisor Disabled, Optimized for Virtual Machine / Better Performance

System A - OS - 10% of 800% used / Windows - 99% idle
System B - Not Applicable

Test 8 - 4 Processors, Adaptive Hypervisor Enabled, Optimized for Virtual Machine / Better Performance

System A - OS - 10% of 800% used / Windows - 99% idle
System B - Not Applicable

Test 9 - 2 Processors, Adaptive Hypervisor Disabled, Optimized for OS X / Longer Battery Life

System A - OS - 10% of 800% used / Windows - 99% idle
System B - OS - 20% of 200% used / Windows - 97% idle

Test 10 - 2 Processors, Adaptive Hypervisor Enabled, Optimized for OS X / Longer Battery Life

System A - OS - 10% of 800% used / Windows - 99% idle
System B - OS - 20% of 200% used / Windows - 97% idle


Conclusions

1. Mac Pro's do not like running a multiprocessor VM with a single processor.
2. Mac Book Pro's seemed to be ok running a multiprocessor VM with a single processor.
3. Changing the optimization settings to favor OS X seemed to have the reverse impact - CPU utilization increased at the host level.

Given the different Intel processors found in the Mac hardware, I can understand results would very between systems. However, I was not expecting to see minimal change or change in the wrong direction.

Thoughts?

 

I know there are more significant issues that require attention by Parallels at the moment. However, I did not want this post to get lost in the process.

I re-read the on-line help and, while the explanations of these settings make sense, I still did not get a sense of how significant the impact should or could be.

Parallels Team - is there additional information you can make available regarding these optimization settings?

 

I second that. P5's memory requirements appear to be vastly higher than P4's. "Optimize for Mac OS X" does not change that as it did with P4. Could someone at Parallels please provide a brief update on the best use of the optimization settings?

 

So, according to the results of tandem test 1 vs 2, 3 vs 4 ...... adaptive hypervisor does nothing?