none
Motivation of using HPC cluster RRS feed

  • Question

  • Hi,

    I attended a presentation on HPC 2008 server last week and am wondering the motivation of using HPC cluster. Btw, I'm new to HPC server.

    1. Is HPC cluster suitable for applications that require frequent file or database accesses? In other words, do I dump everyting into the cluster including file and database access?

    2. Based on the example quoted during the presentation, it looks like there isn't any file or database accesses. Is HPC cluster used solely for number crunching?

    3. Is HPC cluster a shared computing resource that everyone can leverage to perform disparate kind of jobs? If I've dedicated computing resource for my application, do I need HPC cluster?

    Thanks
    kk
     
    Monday, April 6, 2009 1:36 AM

Answers

  • Hi,

    HPC 2008 can use the file system that is local to the compute node or it can access files over a network via SMB or NFS. Another option is to use a third party file system such as Quantum's StorNext, SANbolics Melio, IBrix etc. SQL can also be used over a network.

    If you are using a network connection then frequent file access can be slower than local disk depending on the type of local disk and the type of network.

    An HPC cluster can be a shared or a dedicated resource or both. Some customers run one job at a time on a cluster but can run many jobs thru out the day from many users. Other customers run many jobs on a cluster at the same time. It really depends on the size of the cluster and the requirements of the applications.

    Hope this helped.

    Brian
    HPC
    Wednesday, April 8, 2009 7:16 PM

All replies

  • Hi,

    HPC 2008 can use the file system that is local to the compute node or it can access files over a network via SMB or NFS. Another option is to use a third party file system such as Quantum's StorNext, SANbolics Melio, IBrix etc. SQL can also be used over a network.

    If you are using a network connection then frequent file access can be slower than local disk depending on the type of local disk and the type of network.

    An HPC cluster can be a shared or a dedicated resource or both. Some customers run one job at a time on a cluster but can run many jobs thru out the day from many users. Other customers run many jobs on a cluster at the same time. It really depends on the size of the cluster and the requirements of the applications.

    Hope this helped.

    Brian
    HPC
    Wednesday, April 8, 2009 7:16 PM
  • Hi Brian,

    Thanks for replying.

    Pertaining to your comment on shared and dedicated resource, what is the motivation of creating a HPC cluster if my application requires to hold on to the computing resource permanently (24x7)? Basically, the application consists of several distributed components and when it is started, it will sit there waiting for requests from the clients.

    kk 
    Thursday, April 9, 2009 12:38 AM
  • Hi kk - checkout the SOA (Service Oriented Architecture) in HPC Server 2008. From what you have described it might be a good fit.

    Regards
    Brian
    HPC
    Thursday, April 9, 2009 11:12 PM
  • Hi Brian,

    Thanks. Will do more reading up.

    kk
    Monday, April 13, 2009 8:21 AM