Multiuser

Resource quotas and limitations

Generally speaking, different quotas can be applied for different partitions both at the user level and at the job level, but currently the same limitations are set for each partition:

• limitations per user

  • only 2 jobs can run per user at the same time
  • only 3 jobs can be scheduled by a given user at the same time
  • one user can only take up 100GB of storage on the shared NFS /storage
  • to check the storage space available, please use the quota -u $USER -s command

• limitations per job

  • the grace time (max walltime) for job execution is set per partition — always check the current value with sinfo or scontrol show partition <name> before relying on a number here. As of writing:
  • batch partitions (gpu_g2.2xlarge_32, gpu_wigner_g2.2xlarge_40): 1 day (24 hours)
  • the interactive partition (cpu_m2.2xlarge): 7 days
  • once a job reaches its walltime, it is killed with a TIMEOUT status

This configuration ensures the fair share and utilization of computational and storage resources among the users.

For more details, please visit the List of useful Slurm commands section.

---