Performance of scheduling scientific applications with adaptive weighted factoring

Abstract

Dynamic scheduling algorithms are a powerful tool towards performance improvement via load balancing scientific applications in heterogeneous environments. However, these scheduling techniques employ heuristics that require a priori knowledge about workload via profiling resulting in higher overhead as problem sizes and number of processors increase. In addition, load imbalance may appear only at run-time, making profiling work tedious and sometimes even obsolete. This paper reports on performance improvements obtained by integrating a recently proposed dynamic loop scheduling technique that addresses these concerns, the Adaptive Weighted Factoring, into two scientific applications. The first application involves computational field simulation on unstructured grids, and the second one is Nbody simulation. Reported experimental results on computational field simulation problems using unstructured grids and ongoing work on N-body simulations confirm the benefits of using this methodology, and emphasize its high potential for future integration in other scientific applications that exhibit substantial performance degradation due to load imbalance.