基于Q-learning的虚拟网络功能调度方法  

Scheduling Method of Virtual Network Function Based on Q-learning

作　　者：王晓雷[1,2] 陈云杰 王琛[1,2] 牛犇 WANG Xiaolei;CHEN Yunjie;WANG Chen;NIU Ben(National Digital Switching System Engineering Technological Research Center,Zhengzhou 450000,China;National Engineering Laboratory for Mobile Network Security,Beijing 100876,China)

机构地区：[1]国家数字交换系统工程技术研究中心,郑州450000 [2]移动互联网安全技术国家工程实验室,北京100876

出　　处：《计算机工程》2019年第2期64-69,共6页Computer Engineering

基　　金：国家高技术研究发展计划(2015AA01A706);国家自然科学基金(61521003);国家重点研发计划(2016YFB0801605)

摘　　要：针对现有调度方法多数未考虑虚拟网络功能在实例化过程中的虚拟机选择问题,提出一种新的虚拟网络调度方法。建立基于马尔科夫决策过程的虚拟网络功能调度模型,以最小化所有服务功能链的服务延迟时间。通过设计基于Q-learning的动态调度算法,优化虚拟网络功能的调度顺序和虚拟机选择问题,实现最短网络功能虚拟化调度时间。仿真结果表明,与传统的随机虚拟机选择策略相比,该方法能够有效降低虚拟网络功能调度时间,特别是在大规模网络中调度时间可降低约40%。Most of the existing scheduling methods do not consider the virtual machine selection problem of the virtual network function in the instantiation process,a new virtual network scheduling method is proposed.It establishes virtual network function scheduling model based on Markov decision process,minimizes all Service Function Chain(SFC) service delay time.By designing a dynamic scheduling algorithm based on Q-learning,and optimizing the scheduling order of virtual network functions and virtual machine selection,the Network Function Virtualization(NFV) scheduling time is the shortest.Simulation results show that compared with the traditional random virtual machine selection strategy,this method can effectively reduce the virtual network function scheduling time,especally reduce the scheduling time by about 40 % in large-scale networks.

关 键 词：网络功能虚拟化 服务功能链 调度模型 马尔科夫决策过程 Q-学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]