You can request local scratch space on the Linux cluster by adding the following flag to your job script:

#PBS -l nodes=ScratchSpace

This tells the job scheduler to run your job on a system that has local scratch space available.  Please note that currently there is only one node with scratch space, and that each user can currently only use 50GB of scratch space at one time so you need to consider that when submitting your jobs.

To use this space, in your job script you must tell your program to use the directory that’s created for your job, e.g.:

./myprogram /scratch/${USER}/${PBS_JOBID}

This command passes the argument /scratch/${USER}/${PBS_JOBID} to myprogram so that myprogram can use that directory as its scratch space.  Note that the scheduler will automatically substitute your real username for ${USER} and your real job ID for ${PBS_JOBID}.  You may also want to copy data to that space first, e.g.

cp mydatafile /scratch/${USER}/${PBS_JOBID}/mydatafile
./myprogram /scratch/${USER}/${PBS_JOBID}/mydatafile

This would first copy mydatafile to the scratch space and then run myprogram with /scratch/${USER}/${PBS_JOBID}/mydatafile as its argument.

For additional information or assistance, please contact the Tech Desk at 570.577.7777 or techdesk@bucknell.edu.

Keywords: hpc, Linux, linux cluster