A Comparative Study of Load Sharing on Networks of Workstations

Networks of workstations (NOWs) can be used for parallel processing by using public domain software like PVM. There are significant differences between a real parallel system and a NOW-based system. As a result, load sharing on a NOW can be different from that on a parallel machine. Therefore, it is important to understand the factors that influence the performance of load sharing algorithms on NOWs. To gain this understanding, we study performance of a matrix multiplication application on a network of heterogeneous workstations that uses the PVM. We report performance of five load sharing algorithms that use fixed, variable, and adaptive task granularity.