Water-Level scheduling for parallel tasks in compute-intensive application components

Abstract

The development of complex simulations with high computational demands often requires an efficient parallel execution of a large number of numerical simulation tasks. Exploiting heterogeneous compute resources for the execution of parallel tasks can be achieved by integrating dedicated scheduling methods into the complex simulation code. In this article, the efforts for developing an application from the area of engineering optimization consisting of various individual components are described. Several scheduling methods for distributing parallel simulation tasks among heterogeneous compute nodes are presented. Performance results and comparisons are shown for two novel scheduling methods and several existing scheduling algorithms for parallel tasks. A heterogeneous compute cluster is used to demonstrate the scheduling and execution of benchmark tasks and FEM simulation tasks.