Using the Roar Collab Open Queue

All RC users have access to the open queue, which allows users to submit jobs free of charge. The open queue has a maximum time request limit of 48 hours and also has several other resource limits per user. The per-user resource limits for the open queue can be displayed with the following command:

$ sacctmgr show qos open format=name%10,maxtrespu%40

Jobs on the open queue will start and run only when sufficient idle compute resources are available. For this reason, there is no guarantee on when an open queue job will start. All users have equal priority on the open queue, but open queue jobs have a lower priority than jobs submitted to a paid compute allocation. If compute resources are required for higher priority jobs, then an open queue job may be cancelled so that the higher priority job can be placed. The cancellation of a running job to free resources for a higher priority job is called preemption. By using the --requeue option in a submission script, a job will re-enter the job queue automatically in the event that it is preempted. Furthermore, it is highly recommended for users to break down any large computational workflows into smaller, more manageable computational units so jobs can save state throughout the stages of the workflow. Saving state at set checkpoints will allow the computational workflow to return to the latest checkpoint, reducing the amount of required re- computation in the case that a job is interrupted for any reason. RC has somewhat low utilization, however, so the vast majority of open queue jobs can be submitted and placed in a resonable amount of time. The open queue is entirely adequate for most individual users and for many use cases.