云计算中基于NSGA-Ⅲ的动态虚拟机调度方法

Dynamic Virtual Machine Placement Based on NSGA-Ⅲin Cloud Computing

如何将大量的虚拟机有效分布到多个物理节点上是云计算面临的主要难题之一。基于应用负载的动态变化预测信息,将虚拟机在物理节点上的分布问题转化为一个多目标优化问题,利用遗传算法来寻求该多目标问题的近似最优解,在遗传算法的交叉与变异阶段结合类型匹配的规则,在选择阶段利用基于参考点的快速非支配排序遗传算法(Non-dominated Sorting Genetic AlgorithmⅢ,NSGA-Ⅲ)选择出较优子代种群。仿真结果表明,基于NSGA-Ⅲ的遗传算法(Multi-Objective Genetic Algorithm based on NSGA-Ⅲ,MOGAⅢ)在解决虚拟机调度问题时,在云平台的稳定时间、物理节点的激活数量和虚拟机的迁移次数等优化目标之间做了较好的平衡。和基于NSGA-Ⅱ的遗传算法(Multi-Objective Genetic Algorithm based on NSGA-Ⅱ,MOGAⅡ)相比,算法MOGAⅢ的平均功耗降低了4.46%。

How to effectively map virtual machines(VMs)to physical nodes is one of the most important challenges tackled in cloud computing.Based on prediction information of applications’workloads,VM distribution on nodes was defined as a multi-objective optimization problem and genetic algorithm was employed to find the approximate optimal solution.Type matching was used in operator crossover and mutation.Non-dominated sorting genetic algorithmⅢ(NSGA-Ⅲ)was used for operator selection.Simulation results demonstrate that,multi-objective genetic algorithm based on NSGA-Ⅲ(MOGAⅢ)can well balance the relationship among stabilization time of the cloud,active physical nodes,and the number of VM migration.Compared with multi-objective genetic algorithm based on NSGA-Ⅱ(MOGAII),MOGAⅢdecreases power consumption 4.46%on average.