From HPC Wiki
Jump to: navigation, search

Cluster headnodes are a limited shared resource and one researcher's activities could possibly interfere with the efforts of all other users of the cluster, including jobs running on the compute nodes. This makes it necessary to enforce a strict policy controlling how much CPU utilization a single researcher can occupy on the headnode. Longer running jobs should be submitted to the scheduler via "qsub" or "qlogin". On Kong and Stheno the limit is one CPU minute per a process.

If a user exceeds the limit an email is sent to the user and system administrators describing the terminated job, reiterating this policy, and asking that qsub or qlogin be used. If the user is a student researcher then the email is copied to his/her sponsoring faculty.

If a user repeatedly exceeds the limit then ARCS will temporary disable the user's login.

If one CPU minute is not adequate for your testing of a process you intend to submit via qsub, then please email for an exception.