Heuristic approach to multiple-job submission: A case study
The number of parallel simulations (jobs) needed to solve a given scientific/engineering problem can be arbitrarily large. The computational performance is expected to depend on how the user submits these jobs using available computing resources. One option is to run one job at a time using the maximum number of processors that can be allocated to execute each job in the shortest time. The other option is to run all jobs simultaneously using the minimum number of processors per job so that each simulation can be very long. Here, we propose a heuristic approach to multiple-job submission in which a subset of jobs can be submitted at a time using some intermediate number of processors to significantly improve the total execution time. The subset size and the number of processors can be decided based on the number of jobs to be completed within a given time. In a shared system, the total job completion time also depends on the system load. We present some idea on choosing the best time for job submission to minimize the waiting time. Our heuristics are based on the scaling results of a parallel simulation program (VASP) for a Linux cluster.
Publication Source (Journal or Book title)
Proceedings of the IASTED International Conference on Parallel and Distributed Computing and Systems
Chintana, Y., & Karki, B. (2011). Heuristic approach to multiple-job submission: A case study. Proceedings of the IASTED International Conference on Parallel and Distributed Computing and Systems, 153-158. https://doi.org/10.2316/P.2011.757-062