The following job submission queues are configured on ECCO :
| Queue name | Processor * hours limit | Wallclock limit | Memory per job limit |
Max cores per job limit |
Jobs per user limit |
Nodes in resource set | Groups with access |
| interactive | n.a. | 4 hours | 4GB | n.a. | 1 | compute-0-[0,1] | All |
| regular* [default] | 48 | n.a. | 64GB | 8 | 1 | compute-0-[2-8] | All |
| faculty | 384 | n.a. | 256GB | 8 | 8 | compute-0-[2-8] | Faculty, Premium members |
| premium | none | 7*24 hours | none | 32 | none | compute-0-[2-8] | Premium members |
| HPC | none | none | none | none | none | all | On demand, with approval, no automatic execution! (see below) |
| harvest | 1 | 1 | 8GB | 8 | 24 | compute-0-[2-8] | All |
| sas | none | none | none | 32 | none | compute-0-4 | Used for running SAS jobs |
| Class queues | none | none | 64GB | 4 | none | varioius | Used for class instruction, check with your instructor |
Defaults
By default, all non-interactive jobs have a wallclock limit of 30 minutes - if you need to submit longer jobs, you will need to write a qsub script (at present our q-utilities do not support increasing wallclock limits), see page on Advanced qsub.
By default, all interactive jobs that are submitted through the NX interface are scheduled through the interactive queue, and all other jobs, submitted through the q-utilities or through a command-line qsub submission are submitted to the regular queue.
Submitting to other queues
In order to submit to any of the other queues, assuming you have access rights, you need to specify
-q (queue name)
on the qsub command line, e.g.
qsub -q faculty myjob.qsub
or embed it into the qsub script as
#PBS -q (queue name)
Submitting to the HPC queue
If none of the queues provide enough resources for your jobs, you can submit a job to the HPC queue
qsub -q hpc myjob.qsub
However, you will need to contact the ECCO administrators to request that the job be released from the queue - jobs do not simply run on their own.
- Please provide information on the amount of resources required (walltime, ncpus, mem)
- Access to the HPC queue is contingent on system-wide resources being available. In case of contention, premium and faculty jobs get priority.