Deploying HPC Server on Virtual Machines RRS feed

  • Question

  • Hi All,

    Can somebody let me know the detailed procedure or the relavent documents to install Windows HPC server on a Virtual Machines?

    By virtual machine I do not mean Windows Azure nodes. Say, I have set of Virtual Machines which I have created using VMWare and I need to build my cluster using these virtual machines.

    Thanks in advance.



    Tuesday, May 17, 2011 9:52 AM

All replies

  • Tuesday, May 17, 2011 5:07 PM
  • There is already a thread discussing this:





    Hi Mark,

    Thanks alot for the info. However, I wanted to know if there are any technical documents by microsoft which outlines its support for Virtual Machines(not Windows Azure).

    I went through the above thread you suggested and see that the VMs are Hyper-V based. I wanted to know if VMs are created using VMWare will it be different from the ones on Hyper-V.

    To make it short. Will I be able to deploy HPC Server on VMware Virtual machines if yes, how? and which are the supported technical documents; if not, will the support be provided in the future.




    Wednesday, May 18, 2011 5:58 AM
  • There aren't any documents that I know of - my experiences have been through trial and error.

    Having never used VM Ware for Windows HPC Clusters - I can't really comment - although I do have test clusters that I use were Workstation Nodes (running windows 7) are actually hosted on compute nodes that are running additional VMs.

    They are smaller test clusters through - so the configuration has all been done by hand for each machine

    1) configuring the Hyper-V Role (I presume you could install VM ware as an alternative)

    2) Installing HPC Server (as a pre-configured / pre-deployed node)

    3) during the installation - pointing it to your up and running Head Node


    There may be group policy or networking issues preventing the node communicating with your head node - but these would be the same kinds of troubleshooting exercises that you would be doing with any kind of pre-configured node (regardless if it is a VM or not).


    Hope this helps,



    Thursday, May 19, 2011 8:24 PM
  • Why would it be useful to use VMs as compute nodes?  Shouldn't the compute nodes be dedicated physical hardware to get the best performance out of the HPC system?  It seems like a dedicated compute node with 128GB of Ram, and Quad sockets would perform tasks much faster than the same box divided into 20 VM guest compute nodes.  Lots of wasted resources in hypervisor overhead, shared resources competing to get jobs done, doesn't seem all that useful?

    Friday, August 5, 2011 3:23 PM
  • There are many reasons why people may want to use VMs as compute nodes - some of these include:

    - Multiple OS flavours on a single physical machine

    - Partitioning of resources

    - Security - being able to reset / wipe a VM after a job has finished

    - Checkpointing (pausing a VM or taking a snapshot of the VM)

    - Dynamic Configuration Changes - load a different VM vs rebuilding / reinstalling the entire box.



    • Proposed as answer by Mark Staveley Thursday, October 20, 2011 6:38 PM
    Thursday, October 20, 2011 6:38 PM