none
Workstation Nodes RRS feed

  • Question

  • Hello!
    We have a cluster consisting of 1 head node, 3 compute nodes and several workstation nodes running Windows 7. Is it normal that we can't see any metrics of workstation nodes?
    Tuesday, September 25, 2012 11:57 AM

Answers

  • Thanks for the info. That might well be the root cause of the problem - please also note that the officially supported languages for HPC Pack is English, Chinese and Japanese. As a result i'm afraid you have to stay with English version of Windows for your workstation nodes until we fix it, hopefully in the next release.

    Looking forward to your email!

    Michael

    Monday, October 15, 2012 5:13 PM

All replies

  • Hi,

    what version of HPC pack are you running? and what's the headnode version as well? do you know if you are running 64-bit or 32-bit Win7 on these workstation nodes?

    Michael

    Friday, October 5, 2012 2:52 AM
  • The version of HPC Pack is hpc_2008_r2_sp3

    The workstation nodes run 64-bit OS
    Tuesday, October 9, 2012 9:46 AM
  • That should work fine, but i suggest you upgrade to sp4, as it has more bug fixes and could potentially fix your problem.

    Michael

    Tuesday, October 9, 2012 7:43 PM
  • I updated to SP4, that does no work any way.
    What is interesting, one workstation runs 32-bit OS.And that works fine...
    Wednesday, October 10, 2012 5:50 AM
  • Michael, can I ask you one more question not releted to this topic?

    Is there a way to choose GPU device as a resource type when you submit job like you can choose "core", "node", "socket". The only answer I recieved is: http://social.microsoft.com/Forums/is/windowshpcitpros/thread/8b5b22fb-5d25-4f1d-9365-3399427cd45f . Do you have any other thoughts for this problem?
    Wednesday, October 10, 2012 5:56 AM
  • Hi,

    GPU is not yet supported as a manageable resource. So there's no way to submit jobs towards number of GPU cores, you should in theory still be able to do it per nodes, but it's the responsibility of your code that runs on these nodes to manage the GPU cores yourself.

    On the note of your workstation nodes, i'm talking to the dev team on this and they have some speculations, can you also double check which account the hpc monitoring client service is running under? Please note that it needs to run under local system to gain enough privileges to collect data.

    Please let me know how it works for you.

    Thanks.

    Michael 

    Wednesday, October 10, 2012 9:53 AM
  • Hi)

    The hpc client service is running under local system. But I suppose that the problem can be caused by the language of OS - it is Russian. Tomorrow I'll try to reinstall OS? change it for English version and then I will report you about the results.

    Meanwhile, can you answer,please, what is the maximum number of nodes that could be added to the cluster. The documentation says: more than 1000. We need to have an ability to run the cluster consisting of more than 10000 nodes. Is it possible? If yes, is there any sort of official documentation proving it?

    Looking forward for reply.
    Thanks a lot.
    Thursday, October 11, 2012 1:27 PM
  • Please try English and report, we'd like to figure out what's going on.

    For the maximum number of nodes, do you mean you want to have more than 10,000 nodes or 1000 nodes? If you want to have 10K nodes, we need to have a better understanding of your scenario before giving it a conclusion.

    Thank you again for choosing HPC, i'm looking forward to working with you.

    Michael

    Thursday, October 11, 2012 5:22 PM
  • That works!!!
    I installed Windows 7 Ultimate English. And the previous version was Windows 7 Professional Russian. So, maybe the problem was caused by the language,maybe-by the version (Ultimate or Professional).

    As for number of nodes, the background is the following: we need to create a HPC-cluster consisting of 2K-3K nodes, but one of the requirements is that it should have an ability to scale up to 10K nodes.We chose Microsoft HPC Server as a solution. So, now I wonder if it meets these requirements.Maybe you can send me a link with the appropriate information to learn about the problem?

    Thanks
    Friday, October 12, 2012 2:12 PM
  • Hi,

    I will file a bug for your specific problem and follow up on it.

    As for your scaling requirements, let's talk offline as we want to have a thorough understanding of it - can you shoot me an email at micman@microsoft.com?

    Michael

    Friday, October 12, 2012 4:20 PM
  • OK, I'll e-mail you.

    By the way, one more problem with Russian version of OS occured while trying to change the number of subscribbed sockets. The operation was rejected due to unknown account - the Russian analog of "SYSTEM" (it was interpreted as "???????").When we deleted all the workstations with Windows 7 Professional-Russian edition, the problem was solved.
    Monday, October 15, 2012 4:49 AM
  • Thanks for the info. That might well be the root cause of the problem - please also note that the officially supported languages for HPC Pack is English, Chinese and Japanese. As a result i'm afraid you have to stay with English version of Windows for your workstation nodes until we fix it, hopefully in the next release.

    Looking forward to your email!

    Michael

    Monday, October 15, 2012 5:13 PM