HPC Pack 2008 Tool Pack, Release 1 RRS feed

  • General discussion

  • From time to time we come up with little add-ons or tools that might make your life, or at least cluster, a little better. We bundle these up in to a 'tool pack' and post it up on the MS.com Download Center. (You can find other HPC Pack-related items, like whitepapers and hotfixes there too be searching on keyword 'HPC')

    This release contains a great new tool that we're all very fond of: Lizard.

    Lizard is a little tool that you run on your headnode, which runs LINPACK across your cluster, iterating through various parameters and adapting them to find a close-to-optimal set to get the best LINPACK perf and efficiency results. (One might even say that this tool is a LINPACK tuning wizard)

    Many of you know that good LINPACK results are what helps get you in to the Top500, that running LINPACK is a great way to identify some problems with your cluster, to break the cluster in, or even to just do comparisons between systems that you control upgrades on. Taking some of the guess-work and time-consuming manual tweaks, Lizard will be happy to give you a hand here.

    Download it now from http://www.microsoft.com/downloads/details.aspx?displaylang=en&FamilyID=3313856b-02bc-4bdd-b8b6-541f5309f2ce
    Thursday, June 25, 2009 6:57 PM

All replies

  • Hi Don
    If I hadn't spent a good part of the past couple of weeks manually compiling, running and tuning LINPACK I would have said that this tool was awesome ;)
    Seriously though, this is fantastic stuff thanks very much!
    Friday, June 26, 2009 8:37 AM
  • Hi Don
    I've noticed that Lizard will not run when headnodes are configured in a high availability configuration, reporting 
    "A Microsoft HPC Pack 2008 head node not found. Check Microsoft HPC Pack 2008 is installed as a head node befor running lizard"
    I guessed this was a result of compliance checking by the application to prevent execution from a machine other than a head node, and I figured it was probably because the cluster name in this type of configuration is not the same as the (calculated?) server name which the application runs on. A bit of process monitor digging led me to believe you check the HKLM\SOFTWARE\Microsoft\HPC\ClusterName key for compliance. From this I was able to workaround the issue by temporarily changing the key value to the physical headnode name, rather than the failover cluster network name. 
    Success, but the cluster in question is pre release so fortunately I'm able to experiment. I wonder if a more suitable solution could be found at the Lizard application level please?
    Thanks again for the in all other ways brilliant tool.

    Monday, June 29, 2009 11:16 AM
  • Hey Dan, you're right about there being a bug. We'll get it fixed up for the next revision of the Tool Pack. Thanks for the report!
    Tuesday, June 30, 2009 10:08 PM
  • Cheers Don, much appreciated.
    Tuesday, July 7, 2009 9:36 AM