A discrete PSO-based static load balancing algorithm for distributed simulations in a cloud environment

It is vital to balance the computation and communication load for the satisfactory performance of large-scale parallel and distributed simulations deployed on shared resources in a cloud computing environment. The suitable allocation of simulation components (federates) to hosts is essentially a discrete optimisation problem and the particle swarm optimisation (PSO) algorithm is considered to be highly adequate for this purpose. However, the bionic approach was initially designed for continuous optimisation problems and many PSO-based load balancing algorithms suffered due to the random movement of particles owing to their improper discretisation strategies. Moreover, the method adopted by PSO and most of its variants to update the personal best positions considered only the experience of the particles, which resulted in a bad particle being chosen as the leader. In this study, we propose a new PSO-based static load balancing algorithm named adaptive Pbest discrete PSO (APDPSO) to counter these issues. Good solutions stored in the external archive are utilised when updating the personal best positions of the particles and a probabilityand similarity-based discretisation method for PSO is proposed to update the velocity and position vectors of the particles. Simulation experiments injecting random synthetic tasks are conducted on MATLAB and CloudSim platforms. The results showed that our proposed algorithm improved the convergence and diversity of the swarm significantly and reduced the degree of imbalance of loads efficiently, as compared to the state of the art in this area. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

This study proposes a new PSO-based static load balancing algorithm named APDPSO, which utilizes external archive solutions to update personal best positions of particles and introduces a probability and similarity-based discretization method for PSO. Experimental results on MATLAB and CloudSim platforms demonstrate that the proposed algorithm significantly improves convergence and diversity of the swarm, as well as efficiently reduces load imbalance compared to existing techniques.