How do I submit jobs?

Created by Kali McLennan, Modified on Mon, 16 Dec, 2024 at 4:27 PM by Kali McLennan

The Alpha cluster uses Slurm as the resource manager. Slurm is a very popular and widely used resource manager for HPC, particularly in academic/non-profit settings.

You can make use of a user guide to help a user get a complete understanding of a feature or a product. For example, this is how a user guide might look for using article template

TABLE OF CONTENTS

Partitions
Accounts & Quality of Service
Examples
Informational Links

Partitions

Each institution has a dedicated partition on the Empire AI Alpha system. If your institution is 'cuny' you will only be allowed to submit jobs to the 'cuny' partition.

You can identify which partition to submit your jobs to by running the following command during a SSH session on Alpha:

sacctmgr show assoc user=$USER format='User,Account'

You will only be able to submit jobs to the partition matching the name listed for "Account". For example, if the account listed in the output from this command is "cuny" then you will only be able to submit jobs to the "cuny" partition.

There are two methods to specify the partition to which a job should be submitted:

Option 1: Select a Partition in the Job Script

Add the following line near the top of your job script, replacing both {partition_name} and {account_name} with the name of the partition to submit to:

#SBATCH -p {partition_name}
#SBATCH -A {account_name}

Option 2: Specify a Partition During Job Submission

Alternatively, you can specify the partition by adding "-p" and "-A" flags to the "sbatch" or "salloc" command used to submit a job. See the following commands, and replace "{partition_name}" and "{account_name}" with the name of the partition to submit to:

sbatch -p {partition_name} -A {account_name} my_jobscript.sh
salloc -N 1 -n 1 -p {partition_name} -A {account_name}

Accounts & Quality of Service

Slurm makes use of accounts as well as quality of service options in addition to partitions in order to control access to resources. For Empire AI Alpha your account, quality of service, and partition names will almost always be the same.

Examples

Submit a script named `my_job.sh` to the `cuny` account and partition:

sbatch -A cuny -p cuny my_script.sh

Request an allocation for 10 tasks with a limit of 5 tasks per node in the cornell account and partition:

salloc -A cornell -p cornell -n 10 --tasks-per-node 5

Informational Links

See the sbatch and salloc pages from SchedMD for complete documentation of these commands.