For information about running jobs on the department Linux cluster, please go here.

The following policies pertain to the stand-alone compute servers apart from the Linux cluster, which can be listed using the command

sitehosts compute

 The policies also pertain when running jobs remotely on the Macs in the labs.

Please observe the following policies:

• Access to the compute servers is restricted to individuals who are familiar with running batch jobs in a UNIX/Linux environment. If you are not familiar with running jobs in this way, you can contact consult@stat.berkeley.edu.
• Jobs must be nice'd to a value of 19. In general, jobs on the compute servers are nice'd by default, but if you notice your job is not nice'd, please nice it. Any job that is running on a compute server with a lower priority (without prior authorization) may be terminated.
• Any compute server should have at most as many jobs (across all users) as CPUs on the machine. The number of CPUs on a machine can be seen on the SCF computer grid webpage. If you see (using csstat or top or ingrid-gridmon) that all CPUs are busy, do not run/start a job on that server until resources become available.
• A user should not run so many simultaneous jobs on a compute server such that other users are prevented from starting jobs due to lack of resources. This policy is intended to ensure that CPU cycles and memory are used fairly and effectively. Please contact us if you need temporary or long term access to additional computing resources.