A Comparison of Static and Dynamic Load Balancing

Abstract

By intuition we anticipate that dynamic load balancing policies outperform static ones, but dynamic policies have to react to the current system state. This makes dynamic policies necessarily more complex than static ones. One concern is that the overhead caused by such complexity may negate the benefits of dynamic load balancing. Other concerns are the effect of occasionally poor load-balancing decisions and the potential for instability in dynamic load balancing because of the inherent inaccuracy of system state information. Even though it is known that dynamic policies have advantages over static ones in homogeneous systems, it is questionable whether it is also true in heterogeneous systems. The homogeneous systems under consideration denote systems where all the nodes in the system are identical in both function and processing capacity. The heterogeneous systems under consideration denote systems where all the nodes in the system have the same functions but might have different processing capacities. In this chapter the above concerns are addressed and the performance of static and dynamic policies in a heterogeneous system is compared. The overheads and the delays in both job transfer and state-information exchange are taken into account.