Scratch Storage

HUIT Open OnDemand gives every user a private scratch directory at /scratch/$USER, where $USER is your username. Use scratch storage for fast, temporary storage during a compute job. Typical uses include intermediate files and large datasets that your job reads and writes.

Scratch storage is ephemeral

Scratch storage does not keep multiple copies of your data. A file in /scratch/$USER can disappear at any time, even outside of any job. Do not use scratch storage for long-term storage. Save any results that you want to keep in your home directory instead.

To find your username, run this command in a terminal on the cluster:

echo $USER

If this command returns nothing, run id -nu instead.

How your scratch directory gets created

The system creates /scratch/$USER automatically the first time you run a job, whether through an interactive app, sbatch, or srun. Your scratch directory is private: only you can read or write it.

Using the Terminal app before running a job

The Terminal app connects you to a login node, not a compute node. If you open the Terminal app before you run any job, /scratch/$USER does not exist yet.

To create your scratch directory now, run this command:

srun --pty true

This command runs a simple job on a compute node. The job creates /scratch/$USER. Scratch storage exists on both login and compute nodes, so your scratch directory then also appears in the Terminal app.

Even after /scratch/$USER appears in the Terminal app, do not run compute intensive work there. Login nodes are not for computing. Submit a job instead.