Dynamic load balancing in distributed computer systems with star topology

Abstract

We consider sender-initiated dynamic load balancing policies in distributed computer system with star topology. In this system, the processing power of central node may be greater than that of peripheral nodes and the job arrival rate of each node may be different. The performance of load balancing may be very different according to the selection of sender criteria and receiver criteria for job transfer. But the optimal decision of sender and receiver criteria is very difficult in heterogeneous distributed computer systems. In this paper we propose effective destination decision policies using dynamic criteria such as mean job response time of node and estimated response time of job instead of fixed threshold for dynamic load balancing. We also show that the proposed policies produce very interesting results in heterogeneous distributed computer systems with star topology.