Platform R: Slurm Glossary

The following terms are used frequently in Slurm documentation and also are part of several arguments to Slurm commands.

General Resource (GRES)

In Platform R this refers only to GPUs.

Job

A job is the top-level execution unit that Slurm schedules and runs. A job is most often defined by a batch script run by sbatch, which describes the resources the job and all its tasks will run with.

Jobs have a numerical ID which can be used to run queries about a running job, as well as change the state of the job, including canceling it.

Job Array

Job arrays are used to run identical or nearly identical jobs many times.

Each job of a job array gets its own numerical ID, which is indicated with an underscore: JobID_ArrayID, such as 476599_3, which is array job 3 of the job 476599.

Node

A node is a computer in the Slurm cluster.

Node Feature

A node feature, sometimes just called a "feature," is a capability of the CPUs on a particular node, including the manufacturer, as well as advanced CPU features such as SIMD instruction sets.

See Platform R: Slurm Constraints for selecting features for your jobs.

Partition

A partition is a collection of nodes. Currently, Platform R users' jobs all run on the "shared" partition.

Quality of Service (QoS)

In Slurm quality of service is used to modify the characteristics a job can have. The default QoS in Platform R is normal. You should not need to change your QoS unless directed to do so by Platform R staff.

Task

A task (or job task) is a step within a job, started with the srun command. A job may contain a single task, or several running in sequence, or in parallel using several CPUs. 

Each task of a job gets its own numerical ID, which is indicated with a period: JobID.TaskID, such as 476589.3 which is task 3 of job 476589.