基于多目标离散差分进化的网络感知虚拟机放置算法  

Network-aware Virtual Machine Placement Algorithm Based on Multi-objective Discrete Differential Evolution

作　　者：臧韦菲 兰巨龙[1] 胡宇翔[1] 陈文涛 ZANG Weifei;LAN Julong;HU Yuxiang;CHEN Wentao(National Digital Switching System Engineering and Technological Research Center,Zhengzhou 450002,China;State Key Laboratory of Mathematical Engineering and Advanced Computing,Wuxi,Jiangsu 214125,China)

机构地区：[1]国家数字交换系统工程技术研究中心,郑州450002 [2]数学工程与先进计算国家重点实验室,江苏无锡214125

出　　处：《计算机工程》2019年第6期96-102,共7页Computer Engineering

基　　金：国家重点研发计划“网络空间安全”重点专项(2017YFB0803204);国家高技术研究发展计划(2015AA016102);国家自然科学基金群体创新项目(61521003);数学工程与先进计算国家重点实验室基金

摘　　要：针对云数据中心网络内部流量快速增长导致链路拥塞及全网通信代价过高的问题,提出一种改进的网络感知虚拟机放置算法。根据虚拟机的硬件资源需求,通过种群初始化提高算法收敛速度,采用离散差分变异和交叉操作保证种群多样性。综合考虑通信代价、最大链路利用率、硬件资源约束违反度和链路容量约束违反度4项指标,提出基于ε松弛的多子群精英选择策略,选择最优虚拟机放置方案,增强算法全局搜索能力。仿真结果表明,该算法能够有效降低全网通信代价,并实现网络负载均衡。In view of the problems of link congestion and high network communication cost that result from the rapid growth of traffic within the cloud Data Center(DC),this paper proposes an improved network-aware virtual machine placement algorithm.According to the hardware resource requirements of virtual machines,the population initialization is used to improve the convergence speed of the proposed algorithm.Then,discrete differential mutation and crossover operations are used to ensure the diversity of the population.With a comprehensive consideration of 4 evaluating indicators including communication cost,maximum link utilization,hardware resource constraint violation degree and link capacity constraint violation degree,this paper proposes a multiple subgroup elitist selection strategy based on ε relaxation to select the optimal virtual machine placement scheme and enhance the global exploitation ability.Simulation results show that the proposed algorithm can effectively reduce the network communication cost and achieve the load balancing of network.

关 键 词：数据中心 网络感知 多目标 虚拟机放置 差分进化 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]