Data storage

Types of storage

There are several types of storage available depending on the nature and intended use of the data.

  • Scratch
    Several types of scratch disk are available on each cluster. The /scratch/gpfs/<YourNetID> filesystem is where users should be running jobs. These filesystems are fast because they are composed of special purpose hardware that is local to each cluster (see figure below). Parallel I/O is enabled on these filesystems. While your data on /scratch/gpfs is not backed up, the files will not be deleted. Important results should be copied or preferably moved to /tigress. Additionally, each compute node of a given cluster has local scratch at /tmp. Most users will never need to use the local scratch space. If you need to write (or in some cases read) large temporary files then it may be useful to you.
  • /tigress
    /tigress is a shared disk resource for Della, Perseus, Tiger and Tigressdata. /tigress is principally for backing up non-volatile files generated on /scratch/gpfs/<YourNetID>. Do not make the mistake of writing your job output here. That is, you should never be writing the output of actively running jobs to /tigress or /projects. Users are tempted to do this because those filesystems are backed-up while /scratch/gpfs/<YourNetID> is not. However, this is a mistake and you may adversely affect other users by writing or reading job files from /tigress or /projects.
  • DataSpace
    Dataspace offers long-term storage and publication options for datasets, visualizations or reports.

 

For more information