Default Execution Limits

Execution limits for the three main clusters at HPCVL have been implemented to provide greater flexibility in scheduling user applications.

Users are limited to eight (8) executing jobs at any one time. This means that across the production clusters (M9000 and VictoriaFalls) each user can run up to 8 (total) production jobs simultaneously. Submitted jobs above that limit will remain in the queue until a job slot comes free.

To reflect the differences in processor slots, number of machines, CPU speed and memory available, the total maximum number of processes (threads) that can be run at a given time are as follows

• M9000 Servers (m9k0001-8) 64 (default systems)
• Victoria Falls (vf001-73) 512

Thus, with up to 8 executing jobs, a total of 576 threads/processes are possible at any given time when spread over the three queues as above.

Important Note: Cat. 2 research groups will in many cases require more resources, especially in terms of thread/process numbers, and may fall outside of these parameters. We will work with these users to ensure appropriate access. Any requests for enhanced access should include specifications for number of processors required, total amount of memory, and the expected maximum runtime(s). Contact the User Support group to ensure that this is arranged using the most appropriate resource.

Please note that scheduling of jobs using the commercial software packages Fluent and Abaqus involves a license check and must therefore remain subject to the previously existing limits, i.e. 4 jobs/user, 36 (Fluent) and 20 (Abaqus version 6.5) processes per job. Users of Abaqus version 6.7 (or higher) on the Abaqus Mini Cluster (sw0001-4) are limited to 8 threads/processes as that is the number of single threaded processors available on each node.

These changes makes the utilization of our resources more efficient, while allowing researchers to get their work done or to expand their research and address new problems.