Dynamic and Fair Load Balancing in Heterogeneous Distributed Systems using Hybrid Optimization

Abstract

With the advancement in technology, the need for performance optimisation and efficient load balancing remains challenging in heterogeneous distributed systems due to dynamic workload variations and differences in resource capabilities. To achieve scalable task allocation in heterogeneous environments, a hybrid swarm intelligence-based load-balancing algorithm is proposed that uses PSO and ACO. The proposed work leverages PSO's rapid global exploration to obtain high-quality tasks. The work presented in this paper leverages PSO's rapid global exploration to generate high-quality task-to-node mappings and ACO-based local refinement to improve workload equity and eliminate any remaining imbalances. A multi-objective optimization model for dynamic workload adjustment is constructed, in which task completion time, system cost, and load variance are minimized while Jain's fairness index is maximized. The overhead of task migration and response to changes in workload is also minimized. Static and dynamic workload experiments demonstrate the improved performance of the proposed hybrid PSO-ACO workload distribution compared to traditional load balancing methods. The evaluation results show that the proposed load balancing achieves greater fairness while maintaining similar convergence across varying node counts, swarm sizes, and workloads, with reductions in task completion time and system cost. To verify the feasibility of our proposed method, dynamic workload experiments are carried out.