IT Center Help

You are located in service: RWTH High Performance Computing (Linux)

Job Parameters

This page gives a short overview of the most important computing job parameters. For a full list consult the Slurm documentation.

If you need help writing job scripts or submitting jobs to the Slurm queue, please consult the provided tutorial.

Note: Job parameters can be specified in a short and long form. The short form requires a space after the parameter, whereas the long form requires an "=" without any whitespace. Examples are provided below for clarity.

Table of Content

Setting the Job Name

long format: --job-name
short format: -J
example: -J my_job
default: interactive when using salloc and the name of the job script when using sbatch
remarks: Avoid using special characters. Only alphanumeric are recommended.

Setting the Run Time

long format: --time
short format: -t
example: --time=1-05:10:15 (1 day, 5 hours, 10 minutes, 15 seconds)
default: 15 minutes
remarks: Mind the maximum job run time of your computing project.

Selecting the Output File

long format: --output
short format: -o
example: -o /home/ab123456/results.out
default: output_%j.txt where %j is the job reference number
remarks: Do not use ~ or variables like $HOME as path! Use full, explicit paths as shown in the example. Combine STDERR with STDOUT for easier error analysis by avoiding the --error parameter.

Setting the Number of Cores

Depending on your use case one or multiple of the following options can be set:

CPUs per Task (usually is used for OpenMP or Hybrid parallelisation)

long format: --cpus-per-task
short format: -c
example: -c 10
default: 1

Number of Tasks (usually used for MPI parallelisation)

long format: --ntasks
short format: -n
example: -n 10
default: 1

Number of Tasks per Node

long format: --ntasks-per-node
example: --ntasks-per-node 2

Number of Nodes

long format: --nodes
short format: -N
example: -N 1
default: will be set according to number of tasks

Setting the Memory

long format: --mem-per-cpu
example: --mem-per-cpu 2G (2GB per core)
default: Depends of the selected partition.
remarks:
- Usually you do not need to set this parameter, as the default will get you the maximum memory without being billed for more cores. If you need more memory than the default, choose a different partition.
- Mind that memory is assigned per core. If you use hybrid calculations (e.g. MPI + OMP), please consider the number of cores per process used, when calculating the memory each process needs.
- Slurm considers memory units as binary prefixed, meaning that the conversion factor is 1024 and not 1000. As a consequence, requesting 60Gis the same as requesting 61440M. The difference is negligible for small amounts but can affect jobs that require several hundreds of GB of memory.

Using Nodes exclusively

long format: --exclusive
default: Not used
remarks: For larger jobs, this is almost always what you want to use.

Submitting with a Project

long format: --account
short format: -A
example: -A rwthXXXX
default: your personal quota
remarks: You must be a member of the computing time project to charge computing time to it.

Submitting a GPU Job

long format: --gres=gpu:<N>
example: --gres=gpu:2 (requests two GPUs)
default: none
remarks: <N> is the number of GPUs that you want. Mind that your project needs to be allowed to use GPUs.

Submitting to a specific Partition

Depending on your computational needs (e.g. memory) you may want to specify a specific partition on the cluster.

long format: --partition
short format: -p
example: -p c23mm
default: depends on your computing time project

Using BeeOND

long format: --beeond
default: BeeOND is not used
remarks: Details on how to use BeeOND (BeeGFS On-Demand) can be found on the page about the available filesystems. Please mind that your job will become exclusive if you use BeeOND.