Supplemental material for the CHEP23 poster "IceCube experience using XRootD-based Origins with GPU workflows in PNRP"

• runtimes.csv - A CSV file with job data:

  • GPU type name
  • Job end date/time in unix time
  • Wallclock time of the job
  • Number of CPU cores used by the job

• filesizes.csv - A CSV file with output file metadata:

  • Storage site
  • Job index (unique to storage site)
  • Job type (GPU, CPU1, CPU2)
  • Last modified date in unix time
  • File size in bytes

There are three storage sites, with job sets specifically targetted at each site. Within a job set, there are many independent dags which run physics workflows. Specifically, there is a GPU job that runs first, then two CPU jobs.

Not all jobs have successfully completed, due to various errors (some in computing, some in physics code). This data only encompasses completed jobs.