Task Matching and Scheduling in Heterogeneous Systems Using Simulated Evolution

ABSTRACT This paper describes and analyzes the application of a simulated evolution (SE) approach to the problem of matching and scheduling of coarse-grained tasks in a heterogeneous suite of machines. The various steps of the SE algorithm are first discussed. Goodness function required by SE is designed and explained. Then experimental results applied on various types of workloads are analyzed. Workloads are characterized according to the connectivity, heterogeneity, and communication-to-cost ratio of the task graphs. The performance of SE is also compared with a genetic algorithm (GA) approach for the same problem with respect to the quality of solutions generated, and timing requirements of the algorithms