Using the cluster

  1. Create a script to run your analyses
  2. Test the script and determine what resources it needs (esp. RAM and CPU time) using the System monitor (“Applications”/“System Tools”/“System Monitor”) or top in the Shell. Check for “Resident Memory”!
  3. Submit the script to the scheduling system and request specific cluster resources
    • Number of cores (typically 1 per job)
    • RAM
    • Processing time („walltime”)
      • After the walltime elapses, the job is killed.
      • If the job finishes earlier, resources will be released.

MATLAB has an integrated support for running jobs in a scheduling system via Distributed Computing Server (DCS) and Parallel Computing Toolbox (PCT) and supports SoGE scheduler at limited degree. At the end of the day, MATLAB jobs are translated into a series of qsub commands with arguments according to the configuration.

How to submit jobs

  • MATLAB PCT functions with RHUL-Psycho constructor at /usr/local/apps/psycapps/cluster/PsychoSGE.m
  • parfor loops with RHUL-Psycho constructor or directly with RHUL-Psycho cluster profiler at /usr/local/apps/psycapps/cluster/PsychoSGE.settings
  • Automatic Analysis at /usr/local/apps/psycapps/automaticanalysis/automaticanalysis

Top (Cluster)