Skip to Main Content
With ever increasing network traffic, dynamic load balancing can achieve improved performance in a typical distributed system to cope with the fluctuating workload. The hybrid dynamic load balancing algorithm stays away from the drawbacks of centralized and decentralized load balancing approaches. In this paper, first we address the two imperative design issues that are crucial to the hybrid algorithm viz. division of distributed nodes into virtual groups (clusters) and cluster head (supernode) selection in each group. We propose a new strategy for clustering of nodes based on the theory of integer partition. We deal with supernode selection in each group by proposing two novel algorithms that describe the tradeoff between message complexity and performance. Subsequently, we evaluate the performance of the hybrid algorithm in heterogeneous distributed system. We observe that our hybrid algorithm potentially outperforms the classical decentralized load balancing algorithm.
Date of Conference: 6-8 Dec. 2012