Using BeeGFS

BeeGFS is a file system that is created on the nodes on which a job runs. As physical storage, it relies on the node’s local storage, i.e., a partition of the SSD drive in that node.

This implies that the size of the file system is at most the partition size on the SSD, multiplied by the number of nodes on which your job runs.

You can request the creation of such a temporary file system by adding an attribute to your node resource specification, e.g., the following node request would create a BeeGFS file system on the 10 nodes the job will run on:

-l nodes=10:ppn=36:beeond

Warning

A BeeGFS file system is created when your job starts, and destroyed when your job finishes. As such, you must ensure that all relevant data is copied to permanent storage such as $VSC_DATA before your job ends.

The path to the root of this file system is given by the environment variable $VSC_SCRATCH_JOB and is only defined in the process that runs your job script on the compute node.

Note

BeeGFS is a parallel file system, hence MPI applications that use MPI-I/O, or applications that rely on parallel HDF5 may benefit greatly from using it.