none
CPU Usage: 50% or above on all Server 2016 Hyper-V Clustered nodes with 3 VM's

    Întrebare

  • Hello,

    We have a 3-node Server 2016 cluster, based on Dell R730 servers.All Hosts show a CPU Usage of 50% or above. TaskManager shows the correct value (around 0%)

    Only 3 small test-vm's are running on those hosts. Has anyone seen this behavior?

    We would like resolve this.

    Regards,

    Jan


    Jan

    luni, 13 noiembrie 2017 10:39

Toate mesajele

  • Have you run the cluster validation wizard to see if it alerts you to any errors or warnings in your configuration?  Warnings must be investigated; they often point to issues that need to be resolved, even though they don't prevent the cluster from working.

    You tell us that Task Manager is showing the expected value, but you don't tell us what you are using that produced the 50% reading.


    tim

    luni, 13 noiembrie 2017 14:13
  • Hi Tim,

    I ran the cluster validation wizard and found 2 warnings;

    - The integration components of the 2 VM's on the cluster are not up-to-date

    - the nodes do have a pending restart (due to the november updates)

    The Task manager is showing the expected value since we are running only 3 test-Vm's that do not have any load in it. Therefore the task manager is showing 0- to 1 percent load on the CPU.

    The failover cluster manager is showing 50% CPU Usage, a value that is incorrect. i would like to know which value is shown in FCM and how this can be fixed. Other Server 2016 clusters do show the correct value in FCM.

    Jan Kenter


    Jan

    marți, 14 noiembrie 2017 08:01
  • Task Manager and FCM measure two different things when it comes to Hyper-V.  Task Manager measures CPU utilization of the parent partition, i.e. it does not measure CPU used by the VMs..  FCM measures total CPU utilization in the cluster.

    The fact that you have a couple VMs with out-of-date ICs might be an indication that the VMs are working harder than they should be.  I would definitely correct that.  Don't know if it will make the change for you, but it definitely points to something that should be resolved for efficient operations.  And even though I doubt the pending reboots would cause any performance hit on the VMs, it is always a good idea to perform the reboot ASAP because patches don't take effect until the reboot occurs.


    tim

    marți, 14 noiembrie 2017 14:17
  • Hello Tim,

    Thank you for answering this question. We re-installed two of the four nodes with a fresh installation and formed a two node Hyper-V cluster. Even when no VM's are placed on the cluster, and the cluster is not yet configured in SCVMM 2016 the newly installed cluster shows a CPU Usage of 50% or above. :-(

    We are wondering if you (or somebody else) knows which exact counter is used for displaying the CPU Usage in FCM.


    Jan

    marți, 28 noiembrie 2017 19:32
  • Something else must be going on in your environment.  I lost count a long time ago as to the number of Hyper-V clusters I have built.  None of them showed what yours is showing, so I don't have any more ideas about what might be happening in your environment.

    tim

    miercuri, 29 noiembrie 2017 14:59
  • Hello, I have the same problem with Hyper-V 2016 + CU december 2017 with cluster manager on all server HPE Gen 9.

    A new bug with Hyper-V 2016 + Cluster ?

    Thanks.

    joi, 28 decembrie 2017 19:35
  • Hello Noardo, Yes it is a bug!

    The problem only shows on servers with more than 64 Logical processors and 2 cpu's.

    Some calculation goes wrong, when running: wmic cpu get loadpercentage the output for the 2nd cpu always shows 100%.

    C:\wmic cpu get loadpercentage
    LoadPercentage
    2
    100

    FCM then shows 51% load.. Nothing is wrong with your system, you can verify the functioning of your system by disabling hyperthreading; then the number of Logical Processor will (most likely) drop below 64.. and FCM will show the correct value.

    We have an incident opened for this with Microsoft and we are waiting for them to get this fixed.


    Jan

    joi, 28 decembrie 2017 19:55
  • Hello Janneman,

    Thank you for the quick reply.

    On all server HPE Gen9 with 2 CPU (E5-2650L v4 @ 1,70GHz), there are 56 Logical processors (not more than 64 Logical processors).

    Yes for me about output for the 2nd CPU always shows 100%:
    C:\wmic cpu get loadpercentage
    LoadPercentage
    1
    100

    In January, I will also open an incident at microsoft.

    thanks

    joi, 28 decembrie 2017 20:28
  • Same issue on a 4 node HPE DL160 G9 HyperV 2016 cluster 

    Each node has dual E5-2620v4 CPUs  which adds up to 32 Logical Processors per Node.


    • Editat de Mark-03 vineri, 5 ianuarie 2018 15:33
    vineri, 5 ianuarie 2018 15:32
  • Same issue on a 4 node Cisco UCS C460 Hyper-V 2016 cluster as well.

    Node configuration(s)

    E7-8800 v4 @ 2.2Ghz x 4 sockets = 88 cores * 4 nodes = 176 logical processors

    miercuri, 14 februarie 2018 14:15
  • Also same issue

    2 node, HPE DL 380 Gen10, Xeon 6146 48 logical cores per server (dual cpu)

    It hovers about 50-51% which is clearly wrong (almost no load on the cluster) Also veeam one (monitoring solution) says there's about 10% load on each server.

    This is running latest febuary CU.

    Surely this must be a bug in windows, have anybody opened a case with MS yet?


    Best regards Jonas Akrouh Larsen
    IT-Consultant
    http://www.techbiz.dk

    joi, 15 martie 2018 08:31
  • Also same issue, but we have Hyper-V replication vs. Hyper-V cluster. 2 x HP DL380 dual 14 core, 56 logical cores per server, 128GB RAM, all SSD.   50% cpu utilization on both hosts with no VMs running. 

    Does anyone including Microsoft have a fix?

    Thank you!

    Sam Beatty

    IT-Consultant

    vineri, 27 aprilie 2018 19:11
  • Hello, we have the same situation, but I been researching and a I think so is only a wrong report of the console of Fail Over Cluster, because I think it's reading the Maximum Frequency of CPU. 

    Other monitoring tools give the same data than taskmanager. 

    But I need a confirmation to be sure. 


    joi, 12 iulie 2018 15:16
  • I can't beleve that here we are a year after this was first raised and its still an issue - how old is Server 2016? 2.5 years or so?

    I've just set up a two node cluster with Server 2016 DC and before I've even added any guests / workloads the CPU Usage in FCM > Nodes for both nodes is running between 50 & 53%

    Poor, very poor not to have a fix by now.....

    I know its only cosmetic and its reading incorrectly, but would you put up with your cars speedometer reading 50mph when stopped? No, I didn't think so.

     
    vineri, 14 septembrie 2018 14:16