Storage in Slurm

Storage is not managed by Slurm in any way. You will need to be aware of your storage options and the implications of how you run your job.

NFS

For most purposes, you should choose an NFS directory to store your code, data, and results. Depending on the needs of your job, you may need to set certain environment variables to ensure that NFS directory is being used.

For information about availability of NFS storage, see: https://tig.csail.mit.edu/data-storage/nfs/

AFS

AFS is not compatible with Slurm, because your AFS tokens cannot be securely forwarded to a Slurm job. Files in your normal home directory will not be available to your job.

Temporary local storage

Your Slurm job, by default, has a temporary home directory on the node where your job is run. This will be the fastest storage available, but it is only accessible from within that node. Also, it will be deleted on reboot, or possibly sooner as needed.

Job output

Output of a noninteractive job is written to a location specified by the --output= flag. This should be on an NFS filesystem so that you can read your results.