There are three types of scratch disk available on each cluster:
/scratch is local disk and is the preferred option for reads and writes that are local to a node. On Hecate, it is accessible from all cores. It has the fastest i/o and best performance.
/scratch/gpfs (on Della, Orbital, and Tiger), and /scratch/pvfs2 (on Adroit) are parallel filesystems that can be accessed from all nodes in the cluster.
/scratch/network is NFS-mounted disk space that is shared between all nodes on a cluster.
- None of the scratch filesystems are backed up.
- Scratch disk is available to all users and if the scratch space is exhausted, jobs may not be able to run.
- Please be considerate and purge any files that are no longer needed or move them to /tigress for longer term storage.