Job Local Storage
Updated: May 19, 2023
Edit this Page via GitHub Comment by Filing an Issue Have Questions? Ask them here.In the gizmo compute cluster you will be provisioned a directory on the allocated node’s local disk. This directory exists only for the duration of the job. Once the job exits this directory will be removed.
Please use this space instead of
/tmp
for local storage. The/tmp
directory has less space available and (more importantly) is not automatically cleaned up after the job is finished
This local storage is on an attached SSD and has approximately 7TB available on the K class nodes. This storage is only available to processes running on the allocated node- it isn’t shared between jobs or between jobs running on multiple nodes (e.g. an MPI job)
Accessing Job Local Storage
In the gizmo compute cluster the environment variable $TMPDIR
will be set to the path where this storage has been provisioned. Currently this variable is set using the job ID and uses the form /loc/scratch/<jobid>
. This is a directory unique to the job and owned by you.
We recommend that you use the environment variable in your scripts rather than generating this path as the path to the job local storage may change. Using the environment variable will “future proof” your scripts.
Updated: May 19, 2023
Edit this Page via GitHub Comment by Filing an Issue Have Questions? Ask them here.