Document Type : Research Paper
Department of Computer Engineering, Babol Branch, Islamic Azad University, Babol, Iran
Department of Computer Engineering, Sari Branch, Islamic Azad University, Sari, Iran
Faculty member, Department of Computer Engineering, Babol Branch, Islamic Azad University, Babol, Iran
Nowadays, energy consumption is curtailed in an effort to further protect the environment as well as to avoid service level agreement (SLA) breach, as critical issues in task scheduling on heterogeneous computing centers. Any reliable task scheduling algorithm should minimize energy consumption, makespan, and cost for cloud users and maximize resource utilization. However, reduction of energy consumption leads to larger makespan and decreases load balancing and customer satisfaction. Therefore, it is essential to obtain a set of non-domination solutions for these multiple, conflicting objectives, as a non-linear, multi-objective, NP-hard problem. This paper formulates the energy efficient task scheduling in green data centers as a multi-objective optimization problem so that fuzzy Non-dominated Sorting Genetic Algorithm 2 (NSGA-II) has been applied using the concept of Dynamic Voltage Frequency Scaling (DVFS). In this procedure, we adopted fuzzy crossover and mutation for optimal convergence of initial solutions. For this purpose, the binary variance function of gene values and the mean variance function of objective values are proposed for fuzzy control of mutation rate, increasing the variation in the optimal Pareto front as well as the correct frequency variance function of the processors engaged in scheduling to control the crossover rate. This serves to add the objective of indirect load balancing to the optimization objectives, thereby to replace the three-objective optimization process with four-objective optimization. In the experiments, the proposed NSGA-II with fuzzy algorithm is compared against the NSGA-II algorithm, involving three scheduling strategies namely Green, Time and Cost Oriented Scheduling Strategy. The simulation results illustrate that the newly method finds better solutions than others considering these objectives and with less iteration. In fact, the optimal Pareto solutions obtained from the proposed method improved the objectives of makespan, cost, energy and load balance by 4%, 17%, 1% and 13%, respectively.