基于Q-learning算法的vEPC虚拟网络功能部署方法

摘要

针对虚拟化演进分组核心网(vEPC)环境下,现有虚拟网络功能(VNF)部署方法无法在优化时延的同时保证服务链部署收益的问题,提出一种改进的基于Q-learning算法的vEPC虚拟网络功能部署方法.在传统0-1规划模型的基础上,采用马尔可夫决策过程建立了vEPC服务链部署的空间一时间优化模型,并设计了改进的Q-learning算法求解.该方法同时考虑了空间维度下的EPC服务链虚拟映射和时间维度下的VNF生命周期管理,实现了VNF部署的收益一时延多目标优化.仿真结果表明,与其他VNF部署方法相比,该方法在降低网络时延的同时提高了VNF部署的收益和请求接受率.%In the context of vEPC,a method of virtualized network function (VNF) deployment via an improved Q-learning algorithm was proposed to solve the problem that the existing methods cannot achieve the optimization of time delay and revenue of VNF deployment simultaneously.To get the optimal deployment policy in both space dimension and time dimension,a Markov decision process model of vEPC service function chain deployment on the basis of the traditional 0-1 programming model was established and a solution with an improved Q-learning algorithm was proposed.The method had taken full consideration of both virtual network embedding in space dimension and orchestration of VNF life cycle in time dimension,and thus,the multi-objective optimization of revenue and delay could be attained.Simulation shows that the method can reduce network delay while increasing the revenue and the ratio of request acceptance compared with other deployment methods.