The Pittsburgh Supercomputing Center (PSC) has implemented a flexible job scheduling environment to enable jobs to be placed onto its Cray XT3 system in ways that can help them achieve better performance.
An overview of past work will be provided, discussing changes made to the batch system, work on a graphical machine monitor to help visualize jobs placed on the system, and node assignment changes to help minimize communication overhead to all jobs.
Finally, a summary of work done to accommodate applications that are topology-aware (know in advance how they should be placed on the system) and the system changes that have been made to enable them will be
discussed.