Sign in
 
HomeWindows Server 2012Windows Server 2008 R2Windows Server 2003LibraryForums
Windows Server TechCenter >
> 
> 
AVX and Hyper-V Failover Cluster issue

已答覆 AVX and Hyper-V Failover Cluster issue

  • Monday, June 18, 2012 12:30 PM
     
     
    Sign In to Vote
    0

    We have a (2) node Hyper-V Failover Cluster created; first node (node1) is an HP DL385 G6 running Server 2008 R2 SP1 Data Center, the second node (node2) a new HP DL385 G7 also running Server 2008 R2 SP1 Data Center, both servers are using AMD Opteron CPUs but from different families.  The cluster validation tests all passed, and the cluster is being managed via SCVMM 2012.

    We ran into our original issue with node2, the HP DL385 G7, when it failed to start/run any VMs.  This was resolved by applying Microsoft Hotfix 2568088 which fixed an issue with Hyper-v and AVX "Bulldozer" processor support.  The current issue is trying to live migrate VMs to between nodes under certain circumstances.  A VM can be deployed directly to or started on node2, it can then be live migrated between nodes without issue.  Note: Processor compatibility feature has been enabled on all VMs.  If the VM is started or deployed on node1 first, it cannot be live migrated to node2. SCVMM presents an error about Processor compatibility and fails to proceed. The migration also fails when attempted from Failover Cluster Manager, so I doubt it's an SCVMM 2012 issue.  However the same VM in a stopped state can be migrated to node2 and started without issue.  After that, live migrations to node1 and back to node2 will succeed.  It's just when the VM starts on node1 that it cannot be live migrated to node2.

    We are looking for solution or viable workaround until 2008 R2 SP2 is released and "hopefully" supports AVX.  I also posted this questions to the SCVMM board but got no responses.

    Thanks, Tim

     

All Replies

  • Monday, June 18, 2012 5:35 PM
     
     Answered
    Sign In to Vote
    0

    Was the Bulldozer hotfix applied to both nodes?

    You may need to do that in order to keep both nodes at the same patch level.

    Verify that all virtualization capabilities are enabled in the BIOS on both servers. Note that the newer system may need certain virtualization acceleration features disabled in the BIOS to maintain some form of compatibility with the older system.

    It's best to use G7 across the board, G6 across the board, G8 across the board. While not mandated, we have run into issues across Intel 5500, 5600, and now E5-2600 series platforms that were software driven.

    Keeping all nodes in the cluster identical as far as service pack level, patches, _and_ hotfixes is also a good idea.

    Philip Elder SBS MVP Blog: http://blog.mpecsinc.ca

     
  • Monday, June 18, 2012 9:57 PM
     
     Answered
    Sign In to Vote
    0

    I suspect you have the same CPU compatibility problem as I do.

    When I boot the VM in one particulary host, say #3, the VM do live migration fine.  However if VM boot on host #1, the VM get a different CPU features.

    Let do testing. Try one (or both):

    1-Install CPU-Z on a VM. Run it and see the CPU features, that is "Instructions"

    CPU-Z

    Now, reboot the same VM on node #2 and see "Instructions". Do you see any difference? CPU instructions should be exactly the same to be live migrated.

    2- Install Virtual Machine Migration Test Wizard (Download from Microsoft) on any PC. Then run the wizard and test the 3 hosts, or at least the 2 host that you know you have problems. The report sent a incompatibiliy problem? Any errors?

    Post any results, please.

    See:

    http://social.technet.microsoft.com/Forums/en-US/winserverhyperv/thread/79c707fa-587c-4f08-9710-ade783c95772