A Load Balancing Aware Task Scheduling using Hybrid Firefly Salp Swarm Algorithm in Cloud Computing

Abstract

Cloud computing is an evolutionary computational model which provides on-demand scalable and flexible resources by the pay-per-use concept. Due to the flexibility of cloud, several organizations are setting up more data centers and switching their businesses to the cloud technology. These industries need a proper load balancing to ensure the efficient resources utilization, which reduces resource wastage and helps to optimize costs. Optimal resource allocation can be achieved through efficient task scheduling and load-balancing. An efficient scheduling with load-balancing allocates resources in a balanced way and optimizes the quality of service (QoS) parameters. Task migration is the best way to balance the load. This paper hybridizes the Salp Swarm Algorithm (SSA) with the Firefly Algorithm (FFA), named as Hybrid Firefly Salp Swarm Algorithm (HFFSSA). This approach utilizes FFA's operators to enhance the exploitation capability of SSA by functioning as a local search. Further, a load balancing (LB) heuristic is proposed and incorporated with HFFSSA, named as Load Balancing Salp Swarm Algorithm (LBFFSSA). For verification, the presented work is evaluated by two experimental series. First HFFSSA is tested on global benchmark functions, where it shows its superiority over other existing metaheuristic approaches such as Firefly Algorithm (FFA), Grey Wolf Algorithm (GWO), Particle Swarm Optimization (PSO), and Salp Swarm Algorithm (SSA). In the second series, the LB-FFSSA is evaluated on real datasets (Planet Lab and NASA) using CloudSim Simulator; again, results outperform similar metaheuristics. The simulation results show that LB-FFSSA significantly reduces makespan and improves resource utilization. Furthermore, the proposed algorithm minimizes the Load imbalance Factor (LIF) by migrating the task from an over utilized virtual machine to an underutilized one. It also shows improvement in waiting time and throughput. Simulation results prove that proposed model improves by an average up to 32.3%, LIF by 50.4%, throughput by 42.1%, resource utilization by 40%, and waiting time by 50%.