none
Hyper-V 2012 VM's various slow performance issues RRS feed

  • Question

  • Recently we've set up a two Windows Server 2012 Datacenter as Hyper-V nodes in a cluster.
    The nodes have dual Intel E5-2620 CPU's at 2.00GHz for a totall of 24 logical processors.
    Each node has 16 DIMMs of 8GB for a totall of 128GByte RAM. Low voltage DDR3 modules.

    Both nodes have a dual port QLogic 3242 convergered ethernet adapters.
    The ports are set up using NPAR to each have four virtual NICs, of which both last two are set to a max bandwidth of 9Gbit/s.
    There are four teams created over the four vNICs. Management, Virtual Switch for Hyper-V, Live Migration, and iSCSI.
    All are set up as Failsafe Teams, meaning only one port is used actively.

    iSCSI teams are set up to use a Max Ethernet Frame Size of 9614, and pings to the Netapp SAN respond to a 8000 packet size ping.

    The nodes are all very fast and responsive. It is the VM's that I find too slow in several ways.

    On of the main things I've bee noticing with Hyper-V so far, and we still have a 2008 R2 cluster from which I am migrating to the new 2012 cluster, is the relative slow performance of the VM's. Sometimes they are slow to respond to a RDP connection, slow response with starting programs.

    The most odd thing I have witnessed so far is Quick formatting a virtual hard drive inside a Windows Server 2012 Standard VM.
    A 40GByte drive (GPT, NTFS with Default sector size) takes about 5 minutes. Never had that problem on 2008 R2.
    Both clusters connect to the same SAN using the same NIC configuration.

    It sounds to me like a I/O problem, but I can't figure out why.
    For example: When I migrate VM's from the 2008 R2 cluster to the new 2012 cluster (manually, since the Failover Cluster wizard cannot handle VM's created in SCVMM 2008 R2), and then Convert-VHD them to another drive to a VHDX file, it's runs at 1 to 1.4Gbps in both ways. Oddly enough the Move-Item back to the Original location does it in half the speed (which I assume is a bug in Move-Item).

    This makes me believe the I/O is just fine for the host, so why is it so damn slow in the VM?
    Although when I copy a file from the local hard drive of the host to a hard drive inside the VM, from within the VM, it shows a 1 to 1.6Gbps speed. Totally fine.

    Is it just some weird bugs that exist in Hyper-V 2012, and will those be fixed in R2 or is it something else that I am missing?

    Wednesday, July 17, 2013 9:50 AM

Answers

  • For the Networking part, you might want to disable VMQ (Virtual Machine Queues) on the Physical Host NIC Advanced Settings.


    Regards, Vik Singh "If this thread answered your question, please click on "Mark as Answer"

    Wednesday, July 17, 2013 10:00 AM

All replies

  • When the VM in question is doing the quick format, the whole system becomes almost unresponsive.
    But when the Task Manager is finally loaded. which takes a while, the CPU is at 0% and the memory use is nothing special too.

    The host is also picking it's nose, even those it was about 10 VM's running. iSCSI traffic of course is there, but nothing special too.

    Wednesday, July 17, 2013 9:56 AM
  • For the Networking part, you might want to disable VMQ (Virtual Machine Queues) on the Physical Host NIC Advanced Settings.


    Regards, Vik Singh "If this thread answered your question, please click on "Mark as Answer"

    Wednesday, July 17, 2013 10:00 AM
  • Hi, 

    I know this is a very old thread, but your post caught my eye, because we're considering using NPAR on our QLOGIC 8200 adapters - mainly to separate the cluster traffic from the VM traffic, as we're seeing some issues on 2012R2 with vNics all connected to a single hyper-v switch.

    Microsoft recommends this setup, but I don't get WHY you would route cluster heartbeat and live migrations through the hyper-v switch - it makes absolutely no sense to me!

    Anyway, the quick format issue you are describing is caused by the TRIM feature in Windows. I've seen this on several different SANs now - some does it, some doesn't. Perhaps a storage guy can explain what's going on. It's a windows 2012(R2) thing. You can "fix" it by disabling trim: fsutil behavior set DisableDeleteNotify 1 

    I have yet to figure out, if disabling TRIM has any adverse effects on the VM, so I usually re-enable it after I've formated the LUNs. However, I'm tempted to leave it disabled, since something is obviously "mismatching" between the SAN and Windows config...

    Anyway - still happy with NPAR?

    THanks ;-)

    Friday, September 16, 2016 8:39 AM