Intelligent Network Traffic Control With Deep Reinforcement Learning

With the development of mobile Internet,Internet of Things,big data and cloud computing,the number of network users and the network scale are growing rapidly,At the same time,network services are increasingly diversified and the network structure is becoming more and more complex.Faced with the explosive growth of traffic in the network,how to achieve network traffic load balancing,avoid network congestion,ensure network service quality and improve network throughput with reasonable control and scheduling methods has become a more and more important research topic in the field of computer networks.Traditional traffic congestion control solutions are becoming inadequate both in data center networks and backbone networks which have been continuously updated over the past decade.This is because traditional solutions are often targeted at specific network and traffic scenarios.As new network structures and traffic patterns appear,solutions for old specific network scenarios is no longer suitable for new environments.Therefore,it is necessary to propose an intelligent network flow control scheme with wide applicability.In recent years,due to the improvement of computer hardware,Deep Learning(DL)has made a great success.Deep Reinforcement Learning(DRL),which is made up of Deep Learning and Reinforcement Learning(RL),has made breakthroughs in many fields.Its performance in the field of robotics and industrial automation proves Deep Reinforcement Learning has great potential in intelligent control.At the same time,the development of Software Defined Network(SDN)makes network control more efficient and convenient.Both of them provide new opportunities for smarter network traffic control solutions.This paper makes a research on the problem of network traffic control,and proposes a new intelligent network flow control scheme based on deep reinforcement learning routing optimization and SDN flow control.The main research contents and innovations of this paper are as follows:1.Network traffic scheduling algorithm based on deep reinforcement learningThis paper first discusses the load balancing and traffic scheduling problems in data center networks,and analyzes several existing traffic scheduling schemes and their problems.On this basis,this paper build the modeling of network traffic scheduling problem aiming at the network routing strategy optimization,and proposes a network traffic scheduling algorithm based on deep reinforcement learning.Then,this paper proposes the traffic scheduling priority algorithm to determine the order of traffic scheduling.2.Intelligent network traffic control scheme based on deep reinforcement learning and SDN architectureOn the basis of aforementioned network traffic scheduling algorithm,this paper proposes a systematic intelligent network traffic control scheme based on SDN architecture.This scheme searches the global optimal routing strategy of the network with deep reinforcement learning algorithm.The SDN controller provides corresponding network traffic information for the traffic scheduling decision,and performs corresponding traffic scheduling and control according to the output decision.These enable load balancing of the network.3.Comparative simulation test under different topological networksIn order to verify the practicability of network traffic scheduling algorithm based on deep reinforcement learning for different topological networks,this paper tests the algorithm in two simulation network environments:Fat-tree and random topology.This paper compares several traditional traffic scheduling algorithms to evaluate the effect of our traffic scheduling algorithm.The simulation results show that the intelligent network traffic scheduling algorithm based on deep reinforcement learning can improve the average network throughput rate and improve the network quality more effectively than the traditional traffic scheduling algorithm both in typical data center network and random topology network.