While test and development jobs can be run interactively on headnodes, longer "production" job runs should be run in a batch mode across a larger segment of the cluster. Batches of jobs are managed by entities known as job schedulers and resource managers. Not only do these tools improve the use of the cluster across a broader workload, but they also ensure that your job gets uninterrupted use of the processors, memory, and bandwidth scheduled for it during that run.
Job scheduling for PACE-managed clusters is managed via the Moab scheduler and Torque resource Manager from Cluster Resources. Those who have used OpenPBS or PBSPro will notice the similarities.