A Hybrid Dynamic Load Balancing Algorithm for Heterogeneous Environments

Dynamic load balancing has become a necessity in emerging distributed environments to address the inherent heterogeneity in computing resources. The majority of traditional dynamic load balancing algorithms were developed assuming homogeneous set of nodes and suffer significant performance drop under variety of workloads. Moreover, the centralized dynamic approach limits the scalability with the load balancing unit itself becoming a bottleneck. Conversely, the decentralized dynamic approach though overcomes the above problems, suffers from increased communication overhead. Here, we propose the design of a simple yet effective hybrid dynamic load balancing algorithm that overcomes limitations of centralized and decentralized approaches and performs competitively for heterogeneous system. As per our theoretical analysis, under this new hybrid approach, a highly loaded node is expected to discover lightly loaded node in lesser time as compared to decentralized dynamic approach, minimizes the communication overhead, and is also expected to produce improved performance under variety of workloads.