A new clustered load balancing approach for distributed systems

A distributed system consists of several autonomous nodes, where some of the nodes may be overloaded due to a large number of job arrivals while others nodes are idle without any processing. Load Balancing is used for effectively distributing the load among the nodes. Centralized load balancing schemes are not scalable as the load balancing decision depends on a central server. In contrast fully distributed schemes are scalable but they do not produce a balanced load distribution as they use local information. In this paper we propose a clustered load balancing policy for a heterogeneous distributed computing system. Our algorithm estimates different system parameters like CPU Utilization, Memory Utilization, CPU Queue length, and Response time of the system to decide the workload of each node. The proposed strategy privileges local load balancing over global load balancing and thus reduces communication over global network. Simulation results show that the proposed algorithm performs well.