Job efficiency troubleshooting

Things that might make your job inefficient are:

  • Copying and unwinding large tarballs (especially if they have unneeded files)
  • Not prestaging data files
  • Copying files when streaming them is an option
  • Not using resilient space (if it's available for your experiment) for user code
  • High levels of local IO can cause inefficiency because of disk contention with other jobs
    • If this is the case, you might need to increase your memory request to allow for more disk memory cache
  • Copying files (such as experiment setup scripts) that are already in CVMFS
  • Running very short jobs (<~20 min)

For further assistance, please open a Service Now ticket to Distributed Computing Support