SDN网络中基于深度强化学习的动态路由算法  

Dynamic routing algorithm based on deep reinforcement learning in the SDN network

作　　者：朱国晖 牛皎月 王丹妮 ZHU Guohui;NIU Jiaoyue;WANG Danni(School of Communication and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China)

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121

出　　处：《西安邮电大学学报》2022年第6期1-6,共6页Journal of Xi’an University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61371087)。

摘　　要：对传统网络自身的高分布特性难以实现网络智能化部署以及在传输过程中因网络拥塞带来的高丢包率等问题进行分析,设计一种在软件定义网络(Software Defined Network,SDN)中基于深度强化学习的路由(Deep Reinforcement Learning Routing in Software Defined Network,DRLR-SDN)算法。该算法通过将SDN与深度强化学习算法相结合,实现网络优化与网络资源的智能化管理,用以改善因网络拥塞带来的高丢包率问题。同时,设计了一种可调节的奖励机制实现源交换机和目的交换机之间的上下行调节,通过训练、学习并预测网络的行为,降低网络传输过程中的丢包率。实验结果表明,与非SDN结构相比,DRLR-SDN算法的加入能够降低网络时延、减弱网络抖动、降低网络丢包率,增大网络吞吐量。Due to the high distribution characteristics of the traditional network,it is difficult to achieve intelligent network deployment.The problem of high packet loss rate caused by network congestion during transmission is analyzed.A deep reinforcement learning routing in software defined network(DRLR-SDN)algorithm is designed.By combining the software defined network(SDN)with deep reinforcement learning algorithms,network optimization and intelligent management of the network resources are realized,which can solve the problem of high packet loss caused by network congestion.An adjustable reward mechanism is designed to realize the uplink and downlink adjustment between the source switch and the destination switch.Through appropriate training,it can learn and predict the behavior of the network,reducing the packet loss rate during network transmission.Experimental results show that compared with non-SDN structures,the addition of DRLR-SDN algorithm can reduce network latency,weaken network jitter,reduce network packet loss rate and increase network throughput.

关 键 词：软件定义网络 深度强化学习 丢包率 路由优化 奖励函数 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]