Deprecated: please see new documentation site.


The available resources dedicated to each queue are subject to change, so please double check on the cluster of interest using the command (some queue classes are only available on certain architectures):

llclass -l for P5 575 clusters.

qstat -q for x86 clusters.

Queue Descriptions

On each cluster, the processors on the head node are available via the interactive queue only.

Please note that processors are not exclusive to a particular queue. One example of this is that the set of all processors in the preempt queue is a subset of the checkpt queue. If you submit to the checkpt queue, it must be done with the knowledge that the job may be preempted for a high priority situation.


This queue is useful for debugging parallel application or using multithreaded, interactive applications. It requires no explicit queue script, and it can be accessed via the command line on the head node using the command:

% poe application_exe -rmpool 1 -nodes 1 -procs 1


This queue should be used for standard production runs. Jobs will run in this queue until their requested time is up, the queue time limit is reached, or there is a system failure.


Jobs running on this queue can be preempted for on-demand jobs running under the preempt queue. For this reason, the checkpt queue should only be used by jobs where are capable of writing restart files which can be used to resume a calculation at a later time.


This queue is a subset of cpus in the checkpt queue that may be preempted by a tightly controlled set of users under specific conditions, such as storm surge predictions during a pending hurricane strike on Louisiana shores.

Queue Limits

Your requested walltime tells the job scheduler how much time it will take your job script to run from start to finish. This includes any processing done by the script, as well as the main processing code itself. By setting this to the minimum needed (plus some buffer time) your jobs should be able to be processed sooner. For instance, if you know your code only needs 2 hours to run, ask for 2 hours 30 minutes (buffer time). If you would have asked for longer time, say 5 days. The scheduler would need to set aside x nodes for 5 days.

How many nodes do you need or can use?

One should run computational timing tests to find the most efficient running configuration for a code and the intended calculation. If you ask for more nodes than is allowed for a particular class type, you will receive this error:

  qsub submit.pbs 
  qsub: Job exceeds queue resource limits MSG=cannot satisfy queue max nodect requirement
Queue Class QueenBee other x86s P5 575
interactive n/a n/a 1 hr
workq 2 days 3 days 5 days
checkpt 2 days 3 days 5 days
preempt 2 days 3 days 5 days
priority 2 days 3 days 5 days

Users may direct questions to

Powered by MediaWiki