Cluster Partitions

This page will help you select the right partition for your jobs

Overview

Slurm partitions organize and associate compute resources in the cluster. Partitions have a number of attributes that can be adjusted to suit the particular needs of the compute users- primarily we use partitions to adjust the limits and scheduling of compute jobs.

Gizmo Partition Summary

Partition Name	Nodes	Guaranteed	Priority	Default Time	Max Time	Limits
campus-new (*)	All	yes	normal	3 days	30 days	1000 cores per account
short	All	yes	normal	1 hour	12 hours	8000 core-hours committed per account
restart	All	no	low	3 days	7 days	no limit
chorus	harmony	yes	normal	1 days	5 days	min 1 GPU, max 4, max 8 CPU per job

“guaranteed” means that the job is guaranteed all of its requested time

“priority” indicates the relative priority of jobs in one partition versus the others.

Selection of a partition is done when the job is submitted- queued jobs may have the partition changed, but this cannot be changed once the job has started.

campus-new

The partition “campus-new” is the default partition- this is a general purpose partition with a great deal of flexibility in its limits.

short

“short” is a partition intended for workloads with many jobs of short duration. The limit is a little more complicated in that it limits the amount of run time committed to the account, but allows you to run many more cores than under the campus-new partition.

More information is available here

restart

The “restart” partition has no upper limit on the number of cores allocated to an account’s workload. However, these jobs are not guaranteed the requested walltime and can be preempted- the tasks killed and removed from the node- when jobs in other partitions are waiting and can use the resources allocated to restart jobs.

More information is available here

chorus

The chorus partition contains the newest GPU capable nodes. There are more restrictive limits here as this is a scarce resource at the moment. Jobs in this partition must request a GPU.