none
Child VM only getting 75% Parent CPU testing with Passmark

    Question

  • I have a fresh Windows 2012 Standard with Hyper V installation.

    Created a VM installed Windows 2008 R2 standard. Allocated all 16 Virtual processors to the VM equally according to the dialog 100% of the total system resources.

    Running the CPU Integer Math score on Parent (hypervisor) i get a score of 18630

    Running the CPU Integer Math score on child VM with the 16 Virtual processors, Reserve (100%), Limit (100%) (Weight 10000) i get a score of 11388 roughly 61% of the parent score. During the Passmark test i see in the Hyper-V MMC that the CPU utilization for that VM appears to cap at 75%.

    Is this just the reality of HyperV? Can a single child VM not actually have access to 100% of a CPU's performance?

    I was planning to host this server on Hyper-V for now so that it could easily be scaled and moved to faster hardware but i'm not sure i want to take a 40% performance hit in the mean time.

    My test machine is a Dell R710 with 16GB of RAM and 2 Xeon 5520's.


    Friday, February 08, 2013 11:17 PM

Answers

  • I have a fresh Windows 2012 Standard with Hyper V installation.

    Created a VM installed Windows 2008 R2 standard. Allocated all 16 Virtual processors to the VM equally according to the dialog 100% of the total system resources.

    Running the CPU Integer Math score on Parent (hypervisor) i get a score of 18630

    Running the CPU Integer Math score on child VM with the 16 Virtual processors, Reserve (100%), Limit (100%) (Weight 10000) i get a score of 11388 roughly 61% of the parent score. During the Passmark test i see in the Hyper-V MMC that the CPU utilization for that VM appears to cap at 75%.

    Is this just the reality of HyperV? Can a single child VM not actually have access to 100% of a CPU's performance?

    I was planning to host this server on Hyper-V for now so that it could easily be scaled and moved to faster hardware but i'm not sure i want to take a 40% performance hit in the mean time.

    My test machine is a Dell R710 with 16GB of RAM and 2 Xeon 5520's.



    That much of a hit sounds like something isn't quite right. I don't have access to a system like that but the tests I'm running on my test environment are reliably hanging in the 90%+ ranges.

    Eric Siron
    http://www.altaro.com/hyper-v/

    Sunday, February 10, 2013 11:44 PM
  • the new hyper-v with server 2012 is a bit less overhead compared to the old one

    server 2012 also needs less RAM compared to older releases


    Windows MVP, XP, Vista, 7 and 8. More people have climbed Everest than having 3 MVP's on the wall.

    Hardcore Games, Legendary is the only Way to Play

    Developer | Windows IT | Chess | Economics | Vegan Advocate | PC Reviews

    Tuesday, February 12, 2013 10:00 PM

All replies

  • the overhead of the hypervisor is what is limiting a benchmark


    Windows MVP, XP, Vista, 7 and 8. More people have climbed Everest than having 3 MVP's on the wall.

    Hardcore Games, Legendary is the only Way to Play

    Developer | Windows IT | Chess | Economics | Vegan Advocate | PC Reviews

    Saturday, February 09, 2013 3:08 PM
  • I have a fresh Windows 2012 Standard with Hyper V installation.

    Created a VM installed Windows 2008 R2 standard. Allocated all 16 Virtual processors to the VM equally according to the dialog 100% of the total system resources.

    Running the CPU Integer Math score on Parent (hypervisor) i get a score of 18630

    Running the CPU Integer Math score on child VM with the 16 Virtual processors, Reserve (100%), Limit (100%) (Weight 10000) i get a score of 11388 roughly 61% of the parent score. During the Passmark test i see in the Hyper-V MMC that the CPU utilization for that VM appears to cap at 75%.

    Is this just the reality of HyperV? Can a single child VM not actually have access to 100% of a CPU's performance?

    I was planning to host this server on Hyper-V for now so that it could easily be scaled and moved to faster hardware but i'm not sure i want to take a 40% performance hit in the mean time.

    My test machine is a Dell R710 with 16GB of RAM and 2 Xeon 5520's.



    That much of a hit sounds like something isn't quite right. I don't have access to a system like that but the tests I'm running on my test environment are reliably hanging in the 90%+ ranges.

    Eric Siron
    http://www.altaro.com/hyper-v/

    Sunday, February 10, 2013 11:44 PM
  • I agree Eric i'm ok with Hypervisor overhead but not 40%. Eric on your system you tested can you share any VM settings that may be giving you the 90% performance? What kind of processor are you running? How many cores does it have? I'm wondering if maybe it's more an issue of having 16 cores and the Hypervisor not being good at giving out lots of cores efficiently?
    Tuesday, February 12, 2013 5:55 PM
  • My test cluster nodes are HP ProLiant N40Ls. They each have a single dual-core AMD Turion 2. I have done absolutely nothing to tweak performance on them at all. The guest I tested with is a Windows Server 2012 Standard installation running Active Directory Domain Services. At the time I was running the test, there was also a Windows Server 2008 (R1) Standard Edition guest, but all it runs is file services and since it's a test cluster there are no users, so it should have been at or near 0% CPU the entire time.

    I may be able to access another system that has access to CPU hardware more in line with what you have. I'll see what I can do and report back. Odd that no one else has jumped in on this.

    One thing that I noticed is that the PassMark test runs very quickly. I don't know that I'd be certain it's running long enough to get a truly representative data set. Is there a way to make it run longer that I just didn't see?


    Eric Siron
    http://www.altaro.com/hyper-v/

    Tuesday, February 12, 2013 9:29 PM
  • the new hyper-v with server 2012 is a bit less overhead compared to the old one

    server 2012 also needs less RAM compared to older releases


    Windows MVP, XP, Vista, 7 and 8. More people have climbed Everest than having 3 MVP's on the wall.

    Hardcore Games, Legendary is the only Way to Play

    Developer | Windows IT | Chess | Economics | Vegan Advocate | PC Reviews

    Tuesday, February 12, 2013 10:00 PM
  • So i'm curious in your Hyper-V manager window do you ever see the CPU usage go to 100% on a particular VM? I have run benchmarking software on 2008 R2 and since it was limited to 4 cores out of my 16 i only saw a CPU usage of 25% even when running heavy benchmarking. now that i'm on 2012 and have all 16 cores given by way of Virtual processors to the VM i run the same benchmarking but even during a long benchmarking operation the CPU usage goes right to 75% and stays there and never goes higher.
    Tuesday, February 12, 2013 11:00 PM
  • i use dynamic memory allocation and server 2012 does use a lot when windows update is running

    Windows MVP, XP, Vista, 7 and 8. More people have climbed Everest than having 3 MVP's on the wall.

    Hardcore Games, Legendary is the only Way to Play

    Developer | Windows IT | Chess | Economics | Vegan Advocate | PC Reviews

    Wednesday, February 13, 2013 1:04 AM
  • On my N40L, Hyper-V Manager gets to about 99% CPU.

    I did get to a 16-core system. It also gets a PassMark in the guest that's comparable to the host and Hyper-V Manager shows 99% CPU usage in the guest. Nothing special was configured on that system either.


    Eric Siron
    http://www.altaro.com/hyper-v/

    Wednesday, February 13, 2013 3:21 AM