Research On QoS Intelligent Routing Optimization Based On Software Defined Network

With the rapid development of mobile Internet communication technology,new mobile network applications and services are increasingly available,and the traditional best-effort routing algorithm cannot meet the network Quality of Service(QoS)requirements of these new applications and services.In recent years,the emergence of Software-Defined Networking(SDN)has made the control of networks more flexible and efficient,and the software-defined paradigm has been applied to network routing to improve network QoS to a large extent,but QoS routing optimization schemes based on the software-defined paradigm are often specific to certain specific network scenarios and difficult to adapt to changing business requirements.With the development of artificial intelligence in recent years,machine learning has some significant advantages in the fields of data processing,classification,and intelligent decision making.In order to effectively solve the QoS routing problem under SDN networks,it is more convenient to introduce intelligent algorithms of machine learning technology into the QoS routing optimization of SDN,so as to achieve a more intelligent dynamic routing.However,existing intelligent routing schemes can generally only handle discrete,low-dimensional action spaces and train with offline network data,which cannot be learned online and perform poorly for dynamic network environments.In this paper,we propose a novel intelligent online QoS routing optimization solution by studying the problem of QoS routing optimization with SDN as the base framework and using deep reinforcement learning for intelligent routing decisions.The main contents and innovations of this research paper are as follows.(1)An online QoS routing architecture based on SDN and Deep Reinforcement Learning(DRL)is proposed.The framework uses an SDN controller to provide corresponding network state information for QoS intelligent routing decisions,and issues routing rules based on the decision output of the QoS intelligent routing algorithm.The framework enables real-time collection of network information data and on-policy learning for QoS routing optimization.(2)An intelligent QoS routing optimization algorithm SA3CR based on SDN and Asynchronous Advantage Actor-critic(A3C)is proposed.The SA3CR algorithm introduces the deep learning intelligent model A3C into the QoS network,and redesigns the network structure,input,output,reward functions,etc.are redesigned.The algorithm is able to generate routing policies based on the demand of QoS target flows and takes into account a variety of QoS metrics.The SDN controller is used to use the collected network state information as the input state of the A3C deep strong learning model,and after online training,a series of link weights are obtained,and then the optimal path is calculated according to the intelligent routing algorithm SA3CR to optimize the QoS routing.(3)QoS intelligent routing optimization system based on the softwaredefined network is implemented.The system contains a network sensing module,QoS parameter measurement module,network monitoring module,QoS intelligent routing module,and flow table distribution module.The whole system can realize real-time capture of network data,creation of smart routing policy.The three links cooperate with each other to continuously detect the QoS network status and update the routing policy to complete the smart routing process,so as to realize the in-line routing function for network data flow.(4)By using OS3E network topology to evaluate the architecture and algorithm proposed in this paper,the SA3CR intelligent routing optimization algorithm reduces the average delay by 10%,improves the network throughput by 8.7%,and reduces the network packet loss rate by7.2% compared to the ECMP,KSP,NAF?R,and DDPG?R algorithms on.