The Research On Intelligent Routing Based On Machine Learning In SDN Environment

With the rapid development of information technology,machine learning has made breakthroughs in various fields in recent years,and deep learning has become a feasible method for network operators to configure and manage networks.However,due to the tight coupling between the data plane and forwarding plane of traditional router,the network has bean greatly limited in scalability and flexibility.The independence between traditional network devices also makes it difficult for machine learning technology to be deployed in the network environment.When there is sudden traffic in the links,the existing routing protocol cannot intelligently schedule the traffic according to the real-time state information of the links,so that the network service quality cannot be well guaranteed under these circumstances.The Software-defined network SDN technology broke this bottleneck.It realizes the control and management of the underlying hardware by the way that the centralized controller sends the flow table,which increases the flexibility of logical deployment.Therefore,SDN architecture has a great advantage in dealing with network congestion,and it can easily make use of its characteristics of network programmable and centralized control to realize specific routing decisions.For the congestion problem that is prone to occur on network link,this paper proposes a real time training application based on convolutional neural network in SDN environment.The specific work is as follows.(1)Study the SDN technology of network architecture,SDN controller and the work principle of OpenFlow.Analyzing the possible problems of the link choice in traditional routing protocol,and it can not adjust the routing policy from the current network status information when link congestion occurs,or based on congestion status before routing policy choice of path.The convolutional neural network(CNN)can effectively judge link congestion situation by means of real-time training.(2)Through the description of SDN network flow control and design with unique convolution neural network input and output,take the new data traffic in the network,the switch cache waiting queue and the load of the link as the description of the link state.By comparing the packet latency in the link with the set latency threshold to determine the congestion status of the current network environment.By selecting the appropriate convolution kernel size,convolution layer number,FC layer and activation function,the model can better express the predicted results.Through Depth Firsts Search(DFS)to achieve the optimal path selection.(3)Design and implement the system corresponding to the bearing algorithm.The SDN topology was built on the Ubuntu virtual machine using the Mininet simulator.Design the modules of data acquisition,flow prediction and path selection and this designed function is deployed in the Ryu controller to realize traffic forwarding scheduling for the Open vSwitch.The Experimental results show that the convolutional neural network is more intelligent than the traditional routing scheme,which can effectively reduce the packet delay and packet loss rate,and alleviate the network congestion.