2

What ways are there to prevent users from starting long running, resource intensive processes on the headnode of a Rocks cluster?

I've tried:

  • asking politely
  • setting the nice level in limits.conf to 19. Didn't have the expected effect. Processes run at 19 nice level, but machine still becomes unresponsive (maybe memory)

I'm considering:

  • putting time/memory constraints in limits.conf
  • breaking the links to popular programs like /bin/python. Seems clunky

Other thoughts:

  • Banning offending users is not an option
  • Sometimes users need to run interactive sessions. This can be done by launching one through the console with qrsh, so there's really no need to run anything on the head node

1 Answer 1

0

We disallowed logins to our headnode and installed the gridengine client utilities on user workstations. Users can run qrsh from their workstation to get an interactive session on the cluster and still use qsub and qstat.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .