Job Local Storage
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
/tmpfor local storage. The/tmpdirectory 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.
Note for Java, GATK, and Picard Users
As Java (and by extension GATK and Picard) does not follow the TMPDIR convention we set the Java property java.io.tmpdir using the environment variable JAVA_TOOL_OPTIONS. This updates the default value- if you are using GATK or Picard specific options to set the temporary directory, that option value will be used.