Skip NavigationSkip to Content

Before you start

Torque is an open-source job scheduler based on the Portable Batch System (PBS) and somewhat comparable to the commercial product PBS Pro.  Moab is a commercial cluster workload management package derived from the now defunct open-source Maui job scheduler.  Because of the large overlap in functionality provided by Moab and Torque there are often two or more commands that perform the same tasks.  This document will generally give the Moab commands in preference to the equivalent Torque commands.

In order for Moab to execute batch requests, the user environment must be set correctly and the system default settings should work for most users.  To check that a job can be successfully run, give the command

echo 'echo running; sleep 60' | msub -l nodes=1:ppn=1,walltime=2:00 -o test.log -e test.err

The command should return the job id number, e.g., 969.  Check the queues to see that your job is in the system with the command showq.  Look for an entry with the returned jobid and your username.  Next, check the status of your specific job with the command checkjob 969. Finally, after the job completes look for the files test.log and test.err.  The former should contain the word ?running? and the second should be empty.  Contact the helpdesk is any of these checks fail.

Three file systems will be available to you for the job?s output and temporary files.  Your home directory, which is on the /users filesystem, is shared between all nodes in the cluster and is mounted on some other hosts within the ISP.  This directory should be writable only by your account and permissions may be set such that other users have limited or no access to your files.  A disk quota is also in effect on all home directories so large and/or temporary files should be written to one of the scratch areas.  The first scratch area is mounted as /scratch/cluster_tmp and every user has a private folder here named with their username.  This filesystem is mounted on every node and is useful for sharing data across nodes.  The second scratch area is mounted as /scratch/local on each execution node.  This file system uses disks physically attached to the execution node and, while each execution node has a /scratch/local, each is independent from all others.  No data is shared between nodes.  This file system is also smaller than /scratch/cluster_tmp but will provide faster access times, particularly for random access I/O patterns.

IMPORTANT!  Both scratch areas are set up to house temporary files.  No scratch area is backed up in any fashion and inactive files may be deleted without notice.  Do not store any files that you consider important under /scratch without having a copy saved elsewhere.  If files in /scratch are deleted or lost in any fashion, they can not be recovered.