HPC Pack 2012 and Failover Clustering RRS feed

  • Question

  • Hi, new to this forum and to HPC Clustering.

    I have just tried to set up an HPC clustrer with 1 headnode and 4 compute nodes. I also have a Dell Powervault 3200 disc array which I have been trying to get to work as a shared volume such that all nodes can read/write to the volume at the same time and thus all files are available to everyone. The disk array is connected using direct SAS connectors. I am aware that to do this type of sharing you need some form of clusterd file system to arbitrate access to the array from all 5 servers. From what I can see, the only inbuilt way to do cluster shared volumes is using the CSV capability which is only available withing the 'Failover Clustering' feature in Server 2012. However, I do not want or need the 'failover' aspects of this feature, just the shared volume bit.

    I have been trying everything I can think of to get this installed in a suitable manner. I have install Failover clustering first then added the File Server feature and then create a CSV. When I come to install the HPC Pack onto this, it says I need to install it in a special way as I am part of a failover cluster now. When I proceed down that track it says that you cannot have the Head Node and WCF Broker roles on the same machine! I have no other machine to put that on so this may be a stopper for me. I am not entirely sure what the WCF role will do for me anyway so is may be that I do not need it but until I know what I would not be able to do I cannot say.

    I then tried installing the HPC Pack and then adding the failover clustering onto that. I created a failover cluster and it looked for a while that this might actually work ok as I could still run diagnositics from the HPC Cluster mananger but when I rebooted, the HPC Manager could then not access the SDM store as the service was stopped and wouldn't restart. Does anyone know why that might be? I guess it could be that the 2 types of clustering are incompatible in this way and must be installed in a failover way for it to work.

    I am now running out of ideas and options on how to get a shared HPC cluster wide volume to work, surely this must be done around the world where people do not want failover clustering. Can anyone please help on how to do this or on why the SDM service is stopping once a failover cluster is created. The service error is a generic one with code 1064.


    Tuesday, May 28, 2013 2:11 PM

All replies

  • You may consider  LINQ to HPC for your purpuse (just suggestion)

    WCF broker is used for SOA jobs.If you don't plan to submit SOA jobs to your cluster you may configure your head node to not to use it.

    Go to Cluster Manger \Node Management , select your head node machine , take it offline and right click to bring up properties.

    Find "Change Role " and uncheck "WCF broker".You may also uncheck "Compute Node " role if you don't wish to run your jobs on machine with head node role.

    Daniel Drypczewski

    Friday, May 31, 2013 8:23 AM