CAREER: Advances in Dynamic Scheduling Strategies for Scalable Parallel Scientific Applications

Mississippi State University, Mississippi State MS

Investigators

Abstract

This project will strengthen the high performance computing academic program at Mississippi State University by creating a symbiosis between research and education. The research component will ultimately help extend the knowledge of the scientific community towards obtaining accurate theoretical models, so that computer simulations of physical phenomena give correct predictions. The research will do this by studying dynamic scheduling strategies for these computations, which (when properly constructed) provide simple, fast, and effective algorithms for the simulations. The education component will nurture the development of students' fundamental skills in problem solving in both theoretical and practical training. It will do this by improving the curricula by coupling strong foundational courses from different disciplines and providing meaningful interactions between them. Technically, the research will extend the study of dynamic scheduling strategies used in scientific computing with new methods, and evaluate those methods on both analytical and experimental bases. The work here specifically addresses the issues of discovering novel dynamic scheduling strategies that can accommodate highly unpredictable program behavior, providing an integrated dynamic scheduling framework for scientific applications, and analyzing the performance of parallel applications via predictive performance metrics. Doing this will use the expertise and infrastructure of the NSF Engineering Research Center for Computational Field Simulation.