Job queues on the PPPL cluster
On the PPPL cluster, except as noted below, a specific queue does not have
to be specified in your job script.
A submitted job will be automatically sent to the
correct queue by the job scheduler. This scheduler will determine, from
the number of nodes requested and any special memory requests, which
queue should accept the job. For example, if a job requests 1-2
CPUs, the job will go to the default queue. If it requests more than 2 CPUs,
the job will go to the parallel (kestrel) queue.
For special purpose jobs only (like those for Theory jobs, those requiring
infiniband, or destined for the SMP systems),
the queue (kruskal, kite, smpq) would be specified in
your job script.
After submitting your job, you may be curious about where it is being run.
(This information is available using the 'qstat' command). There are
several batch queues on the PPPL cluster,
kestrel
(for parallel jobs),
the kite queue (for jobs which
use an infiniband interconnect) ,
the sque
(for single CPU jobs), kruskal (an Infiniband cluster reserved
for Theory jobs),
smpq (for the stix SMP system),
and mque (for jobs requiring large memory sizes).