Multiprocessor Execution

Member Site › Forums › Rosetta 3 › Rosetta 3 – General › Multiprocessor Execution

Viewing 2 reply threads
  • Author
    Posts
    • April 20, 2016 at 3:06 am #2410
      Anonymous

        Hello,

        My apologies in advance if this has already been addressed, but I haven’t found the answer elsewhere.

        I have a powerful workstation with dual Xeons with 16 cores each. dual booting Windows 7 and Ubuntu Linux.  Running on the Windows partition I have used two commercial solutions as well as allowing Rosetta@Home to utilize the system a few hours each night.  Modeling done on all three solutions pegs all 32 processing cores at 100% and the completion times on high quality homology modeling is very low (around 10 hours).

        However, I don’t find the same running on Ubuntu 14.04 LTS.  The most dramatic contrast is Rosetta@home… running on Windows 7, as I mentioned, all 32 processors are pegged and I have contributed several million credits to the project in a short time.  When I run it on Linux, it uses only one processor and the performance difference is predictably deplorable.

        I know homology modeling from the standpoint of the generation of a single model is not inherently improved by multithreading.  However, running a job where hundreds of models are generated obviously is.

        I am not a Linux pro, and honestly, I have no desire to be.  I just want to get my work done.  I need Rosetta generated models.

        Can someone please give me a step by step for getting Rosetta setup to fully exploit the processing power  of my machine?  I’ve tried several iterations, but there are so many pacakges, versions, path variables, etc… 

        Thank you kindly!

        -Scott

      • April 20, 2016 at 4:19 pm #11513
        Anonymous

          https://www.rosettacommons.org/docs/latest/rosetta_basics/MPI

          Rosetta doesn’t natively use more than one processor at a time.  You have to tell it to do so.  MPI is the tool in broad use to do that.  Read through the MPI documentation linked above (and several of the pages it links to) and then come on back with further questions!

          I can’t give you a “step by step” here because A) I don’t know which step you need me to start at (this is an eternal problem for our documentation), and B) it’s always a little different from machine to machine.  We don’t need to go as far as “Linux pro” to get this working, don’t worry.

           

        • April 21, 2016 at 1:51 pm #11516
          Anonymous

            Thank you very much… I’ve been able (I think) to get MPI setup.  I’ll come back when it doesn’t work the way its supposed to because I didn’t set something up right ;)

             

          • April 21, 2016 at 11:00 pm #11517
            Anonymous

              Great!

              Let me gently suggest that you start a new thread if you do have a later, more specific problem.  A “new thread” alert will go to several people, but posting in this thread alerts only me.

          • Author
            Posts
          Viewing 2 reply threads
          • You must be logged in to reply to this topic.