HGLPSO: Hybrid Genetic Learning PSO and its Applications to Task Matching on Large-Scale Systems

Abstract

Matching tasks to be executed with proper resources is essential for improving the performance of grid systems. Assigning a set of tasks to a set of heterogeneous resources is challenging and becomes more complicated when the number of tasks and resources increases. This problem is known as thetask matching problem and is an NP-hard problem. Swarm Intelligence (SI) methods have been adopted as a solution to this problem. One such algorithm is particle swarm optimization (PSO); however, PSO tends to get stuck at local optima in such complex problems. This paper introduces a hybrid genetic learning PSO (HGLPSO) algorithm for the task matching problem in the grid environment. HGLPSO incorporates two genetic learning schemes to create candidate solutions (exemplars). Accordingly, the resulting exemplars possess the right balance of exploration and exploitation search abilities to direct the particles in the search space. The results demonstrate the effectiveness and efficiency of HGLPSO compared with other PSO variants in a heterogeneous grid environment.