BwForCluster MLS&WISO Production Slurm

1 General information about Slurm

• Slurm documentation: https://slurm.schedmd.com/documentation.html
• Slurm cheat sheet: https://slurm.schedmd.com/pdfs/summary.pdf
• Slurm tutorials: https://slurm.schedmd.com/tutorials.html

2 Slurm Command Overview

Slurm commands	Brief explanation
sbatch	Submits a job and queues it in an input queue
squeue	Displays information about active, eligible, blocked, and/or recently completed jobs
scontrol	Displays detailed job state information
scancel	Cancels a job

3 Job Submission

Batch jobs are submitted with the command:

$ sbatch <job-script>

A job script contains options for Slurm in lines beginning with #SBATCH and your commands which you wand to execute on the compute nodes. For example:

#!/bin/bash
#SBATCH --partition=single
#SBATCH --ntasks=1
#SBATCH --time=00:20:00
echo 'Hello world'

The jobs requests requests one core (--ntasks=1) for 20 minutes (--time=00:20:00) on nodes provided by the partition 'single'.

If you want to convert Moab batch scripts to Slurm, you can find general information on this page. Specific information for the usage of Slurm on bwForCluster MLS&WISO Production is included in the following chapters.

3.1 Partitions

It is necessary to request a partition with '--partition=<partition_name>' on job submission. Within a partition job allocations are routed automatically to the most suitable compute node(s) for the requested resources (e.g. amount of nodes and cores, memory, number of GPUs). The devel partition is the default partition, if no partition is requested.

The partitions devel, single, and gpu-single are operated in shared mode, i.e. jobs from different users can run on the same node. Jobs can get exclusive access to compute nodes in these partitions with the "--exclusive" option. The partitions multi and gpu-multi are operated in exclusive mode. Jobs in these partitions automatically get exclusive access to the requested compute nodes.

When choosing the partition gpu-single or gpu-multi, the number of GPUs must be requested with the option "--gres=gpu:<number-of-gpus>".

Partition	Node Access Policy	Node Types	Default	Limits
devel	shared	standard	ntasks=1, time=00:10:00, mem-per-cpu=2gb	nodes=1, time=00:30:00
single	shared	standard, best, best-sky, best-cas, fat, fat-ivy	ntasks=1, time=00:30:00, mem-per-cpu=2gb	nodes=1, time=120:00:00
multi	job exclusive	standard, best, best-sky, best-cas	nodes=2, time=00:30:00	nodes=128, time=48:00:00
gpu-single	shared	gpu, gpu-sky, gpu-cas	ntasks=1, time=00:30:00, mem-per-cpu=2gb	nodes=1, time=48:00:00
gpu-multi	job exclusive	gpu	nodes=2, time=00:30:00	nodes=18, time=48:00:00

3.2 Constraints

If a job requires a certain Intel architecture, the architecture of compute nodes can be explicitly requested with "--constraint=<constraint_name>".

Constraint	Meaning
ivy	request nodes with architecture Ivy Bridge
has	request nodes with architecture Haswell
sky	request nodes with architecture Sky Lake
cas	request nodes with architecture Cascade Lake

3.3 Examples

Here you can find some examples for resource requests in batch jobs. Most partitions allow the allocation of various node types. If you need a certain node types, adapt the memory request, the number of cores per nodes, and/or use the "--constraint" option.

3.3.1 Serial Programs

#SBATCH --partition=single
#SBATCH --ntasks=1
#SBATCH --time=120:00:00
#SBATCH --mem=4gb

Notes:

• Jobs with "--mem" below 60gb can run on all node types associated with the single partition.
• If you increase the memory request, less node types may be available for the job.
• If you need best nodes only, request for example more memory "--mem=70gb" and use "--constraint=has" to exclude standard as well as best-sky, best-cas, and fat-ivy nodes.

3.3.2 Multi-threaded Programs

#SBATCH --partition=single
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=16
#SBATCH --time=01:30:00
#SBATCH --mem=50gb

Notes:

• Jobs with "--ntasks-per-node" up to 16 and "--mem" below 60gb can run on all node types associated with the single partition.
• If you increase "--ntasks-per-node" or "--mem", less node types may be available for the job, see 'Number of Cores' and 'Working Memory' in the CPU Nodes Hardware table.

3.3.3 MPI Programs

#SBATCH --partition=multi
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=16
#SBATCH --time=12:00:00
#SBATCH --mem=50gb

Notes:

• "--mem" requests the memory per node.
• Jobs with "--nodes=" up to 2, --ntasks-per-node" up to 16, and "--mem" below 60gb can run on all node types associated with the multi partition.
• If you increase "--nodes", "--ntasks-per-node", or "--mem", less node types may be available for the job, see 'Number of Cores' and 'Working Memory' in the CPU Nodes Hardware table.

3.3.4 GPU Programs

#SBATCH --partition=gpu-single
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=40
#SBATCH --gres=gpu:4
#SBATCH --time=12:00:00
#SBATCH --mem=200gb

Notes:

• The number of GPUs is requested with the option "--gres=gpu:<number-of-gpus>"
• It is also possible to request a certain GPU type with the option "--gres:gpu:<number-of-gpus>:<gpu-type>". For <gpu-type> put the 'GPU Type' listed in the last line of the GPU Nodes Hardware table.

4 Interactive Jobs

Interactive jobs must NOT run on the logins nodes, however resources for interactive jobs can be requested using srun. The following example requests an interactive session on 1 core for 2 hours:

$ srun --partition=single --ntasks=1 --time=2:00:00 --pty /bin/bash

After execution of this command wait until the queueing system has granted you the requested resources. Once granted you will be automatically logged on the allocated compute node.

If you use applications or tools which provide a GUI, enable X-forwarding for your interactive session with:

$ srun --partition=single --ntasks=1 --time=2:00:00 --pty /bin/bash --x11

Once the walltime limit has been reached you will be automatically logged out from the compute node.

5 Job Monitoring

5.1 Information about submitted jobs

For an overview of your submitted jobs use the command:

squeue

To get detailed information about a specific jobs use the command:

scontrol show job <job-id>

5.2 Interactive access to running jobs

If you like to see what happens on the compute node(s), you can enter access the allocated resources of a running job with:

$ srun --jobid=[jobid] --pty /bin/bash

Commands like 'top' show you the most busy processes on the node. To exit 'top' type 'q'.

To monitor your GPU processes use the command 'nvidia-smi'.

6 Job Feedback

Information will be provided soon.

7 Overview about available resources

Information will be provided soon.