Windows 10 Pro workstation node on HPC Pack 2012R2 server RRS feed

  • Question

  • Windows 10 workstation node idle detection issue with HPC Pack 2012R2 head node.

    We have a groups of users who are currently being upgraded to Win 10 from 7 and have lost the ability to detect idle time on the Win 10 workstation nodes causing jobs to land on workstation that are currently being used on site.

    Seems to function as a valid workstation node other than the idle detection.

    Is there any work around or patch I can look at before looking into migrating everything to HPC pack 2016?

    WS node version:  Cluster software version is 4.4.4864.0

    Tuesday, April 23, 2019 4:24 PM

All replies

  • Currently there is a workaround worth of try:

    Upgrade the cluster to HPC Pack 2012 R2 Update 3 at least, you can download the latest build here: https://www.microsoft.com/en-us/download/details.aspx?id=49922

    https://docs.microsoft.com/en-us/previous-versions/windows/it-pro/hpc-server-2012-R2-and-2012/mt595796(v%3dws.11) described below new feature

    Customizable idle detection logic – Until this release the workstation nodes and unmanaged server nodes have been treated as idle based on keyboard or mouse detection or CPU usage for processes other than those for HPC Pack. Now we add these capabilities:

    • You can whitelist processes you want to exclude from calculating the node CPU usage by adding below registry key values (Type: REG_MULTI_SZ) HKLM\Software\Microsoft\HPC\CpuUsageProcessWhiteList.

    • When you don’t specify keyboard or mouse detection or a CPU usage threshold, you can provide your own node idleness logic by creating a file with the name IdleDetector.notidle in the %CCP_HOME%Bin folder. HPC Pack checks whether this file exists and reports to the scheduler every 5 seconds.

    And meanwhile, we could investigate why idle detection failed with HPC Pack on win10.

    Qiufang Shi

    Wednesday, April 24, 2019 2:58 AM
  • Do you have the scheduler log on headnode and node manager log on workstation node?


    Sunday, April 28, 2019 2:05 PM
  • They didnt really give me a time frame of when they started migrating to Win10. 

    The only thing I caught in the "Windows HPC Server" log on the Head Node was frequent node time outs (9s) but I'm not sure if that has to do with those users doing something to remove or disable their workstation (compute node) when they are using it, otherwise a jobs could land on it while working. I dont have insight on if that node has been upgraded to Win10 or not to clarify it the time out has to do /w Win 10 nodes only. The head node is not on that local network and is located in another state so that could also cause some latency to produce time outs from the nodes as well from time to time. Otherwise its been working great for them for several years now until the Win10 upgrades started on their workstations/pc's.

    <Event xmlns="http://schemas.microsoft.com/win/2004/08/events/event">
    - <System>
      <Provider Name="Microsoft-HPC-Scheduler" Guid="{5B169E40-A3C7-4419-A919-87CD93F2964D}" />
      <TimeCreated SystemTime="2019-04-29T15:26:50.162931500Z" />
      <Correlation />
      <Execution ProcessID="4540" ThreadID="5004" />
      <Channel>Windows HPC Server</Channel>
      <Security UserID="S-1-5-18" />
    - <EventData>
      <Data Name="NodeName">CVP00155</Data>
      <Data Name="Timeout">90000</Data>

    Tuesday, April 30, 2019 9:06 PM
  • The scheduler log is under %CCP_DATA%LogFiles\Scheduler on headnode, the node manager log is under the same path on compute node (workstation node). It would help to investigate the issue if these logs are provided. You could place it somewhere for downloading. Or mail hpcpack@microsoft.com for help.
    Thursday, May 16, 2019 12:37 PM