Child pages
  • I want to increase my computational research productivity
Skip to end of metadata
Go to start of metadata

Traditionally, this is the reason most High Performance Computing facilities came into existence.  A desktop computer can be the perfect environment for interactive or small scale research computing.  As their research computing requirements grow, researchers often find their desktop computer is a far less than ideal environment for doing the work.  You might ask "what benefit will using JCU HPRC facilities provide for me in terms of productivity increases?".

  1. JCU HPRC infrastructure operates in an environment with redundant power and cooling, so your jobs are more likely to run to completion on JCU HPRC servers.
    • There is very little you need to do in this space.
  2. ICT staff have installed quite a lot of software on HPRC for your use.  You don't have to deal with the pain of trying to get software compiled and working.
    • Check the list of available software.  If your software (or the version you want to use) doesn't appear to be installed, put in a request for HPRC staff to install it.
  3. ICT staff have experience with optimisation of software so that your jobs complete as quickly as possible, given the hardware we have.
    • You could interact with HPRC staff to determine whether pushing the optimization parameters to their limits (sacrificing a little in accuracy) is acceptable to you.
  4. You can run many jobs at the same time on JCU HPRC servers.  There have been times were a single researcher has been able to run 500 jobs at the same time.
    • You will need to ensure that your jobs can be run in parallel - e.g., making sure that there aren't any common input/output files.
    • You might want to explore the level to which you want to scale up - e.g., if your environment can run 100 jobs at a time why not try to get it done.
  5. Does the software you are using have a pre-parallelised version that you could use?
    • There are OpenMP, MPI, and/or PVM options for some applications.
    • If an individual job is taking many days/weeks/months to run, running each job across multiple processors is a great option for reducing the job completion time.
    • Efficiency of the multi-processor job is a key consideration here.  Not all workloads scale up well - it's often dependent on your data.  For example, you may find that trying to run your job on 24 CPU cores only results in a ten-fold decrease in time - running the job on 12 CPU cores may achieve results in roughly the same time.
  6. a

 

  • No labels