A comparative study of load sharing in heterogeneous multicomputer systems

Abstract

The effective use of a heterogeneous multicomputer system depends on the scheduling policy used. Often the scheduling algorithms used are those that were intended for homogeneous systems. On the other hand, most scheduling algorithms designed with heterogeneity in mind were developed under unrealistic assumptions. To enhance the performance of a multicomputer system, load sharing can be used during scheduling to distribute the workload evenly among the constituent nodes. This paper presents a comprehensive evaluation of load sharing algorithms in heterogeneous multicomputer systems. The simulation results suggest that load sharing policies that perform well in homogeneous systems fail to achieve good performance in heterogeneous systems. The authors also show that the distribution of service time and the communication overhead have significant impact on system performance. They conclude by providing guidelines for selecting load sharing policies suitable for different heterogeneous environments.<>