How long is the collected data stored and how often routine cleanup processes are performed?

Currently four years.

After four years all collected data is purged, except the information which is required for the reporting process towards funding organizations. The latter is removed after the time defined by the funding organizations expires.

What kind of information is collected currently or could be collected in future in which amount and for which purposes?

To manage efficient distribution of the HPC resources and to assure the users/groups rights on fair-share basis the following data is collected by the employed job scheduler:

• job name,
• user name,
• account (HPC project)
• cluster
• partition (e.g., epyc-test, epyc-mem)
• allocated nodes,
• computational job
  
  • start time,
  • completion time,
  • exit status
  • workind directory
  • command
• requested resources
  • CPUs
  • memory (RAM) (memory_per_cpu, memory_per_node, ...)
  • nodes
  • GPU-s
• time shares ("efficiency")
  
  • system
  • limit
  • total
  • user

This data is also required to generate cluster usage reports (eventually in reporting process for the funding organization) and to identify/plan the appropriate cluster extension or improvement directions.

For storage accountability and future planning the disk space usage and file counts are performed on a regular basis

Who has access (can see) the collected data? 

All cluster users can check/acquire information from the running system about the resources requested, obtained, and used by himself and other members of the cluster user group too.