Swarm Intelligence-Based Workload Management in Computer Clusters

Abstract

Efficient workload scheduling in heterogeneous distributed systems remains a complex challenge due to variations in computing capacities and unpredictable task arrivals. Traditional scheduling algorithms, including conventional Particle Swarm Optimization (PSO), often suffer from premature convergence and limited adaptability under dynamic workload conditions. To address these limitations, this study proposes an Adaptive Particle Swarm Optimization (APSO) algorithm that dynamically adjusts the inertia weight parameter to maintain an effective balance between exploration and exploitation during the search process. The adaptive mechanism allows particles to respond more effectively to workload fluctuations and prevents stagnation in local optima. Experiments were conducted on a simulated heterogeneous cluster environment consisting of multiple computing nodes with varying processing speeds and workloads. The performance of the proposed APSO was evaluated using three primary metrics: makespan, CPU utilization, and load imbalance. The results demonstrate that APSO successfully reduced makespan from 350.0 s to 287.0 s, achieving an improvement of approximately 18%, and increased CPU utilization from 77.8% to 83.4% compared to the Round Robin baseline. These findings confirm that the adaptive parameter control significantly enhances scheduling efficiency, improves resource utilization, and provides a more robust and reliable solution for dynamic heterogeneous distributed systems.