An Efficient Load Balancing Technique for Mutation based Particle Swarm Optimization

Abstract

Load balancing techniques play significant role in cloud computing environment because it directly affects the performance of cloud data centers. An efficient load balancing technique not only provides high availability of resources to cloud users, but also enhances the performance of cloud data centers. Load balancing techniques are a typical NP-hard problem. Currently, many researchers have solved load balancing problem by considering well-known metaheuristic techniques. However, these techniques suffer from one of these issues: premature convergence, poor convergence speed, initially selected random solutions and stuck in local optima. To handle the issues associated with existing metaheuristic techniques, in this paper, a mutation based particle swarm optimization based load balancing technique is proposed. The proposed technique has an ability to overcome several issues associated with existing techniques such as premature convergence, poor convergence speed, initially selected random solutions and stuck in local optima issues. Also, multiobjective fitness function is designed as a minimization problem. Multi-objective fitness function considers energy consumption, makespan and load imbalance rate parameters. The proposed technique outperforms existing load balancing techniques in terms of makespan, speedup, communication overheads, efficiency, utilization, mean gain time, load imbalance rate and energy consumption. Index terms: Load balancing, Particle swarm optimization, Cloud computing, Energy efficiency. I. INTRODUCTION Cloud computing is growing phase having a world view of substantial scale distributed computing in internet era [1]. It is a combination of grid and 'X' as a resources in wireless environment. Here, 'X' may be infrastructure, platform, software, data, hardware etc. [2]. The primary objective of scheduling is the assignment of jobs to available set of servers so that execution time can be minimized. Another feature of scheduling is decision-making process and is mostly used in service and manufacturing organizations. Scheduling takes place by taking help of load balancing algorithms [3]. Job is a term related with scheduling used at application level and is a script or program to execute a specific set of jobs. Load balancing algorithm is a technique to which jobs are optimistically assigned to data center resources. There is no completely perfect scheduling mechanism available due to different scheduling objectives. Scheduling algorithms can be executed or implemented under suitable conditions according to assigned applications by a good scheduler. Scheduling algorithm is a mechanism that solves a problem in seconds, minutes or even hours. Time used for execution of particular algorithm measures the efficiency of that algorithm and so time complexity can be measured from the efficiency. Time complexity plays a significant role in time execution of an algorithm. There are some time complexity algorithms used for job execution. The problem is feasible, traceable, fast and efficient in case ...