Design And Simulated Implementation Of Intelligent Routing Mechanisms For SDN

As an emerging networking paradigm,Software Defined Networking(SDN),has become a hot topic in network research area.However,with the rapid growth of network traffic,diversification of the network applications,the continuous updating of OpenFlow protocol version,and the increase of the flow table size,SDN is faced with many challenges,including the shortage of the Ternary Content Addressable Memory(TCAM)storage space,the degeneration of the routing and forwarding ability,etc.Therefore,a novel intelligent routing mechanism for SDN is proposed in this thesis.This mechanism takes full advantage of the controller's data collection and process,and uses the artificial neural networks(ANN)to learn historical traffic data to obtain the data flow transmission patterns and replacing the flow table by the ANN.Thus,switches can use the ANN to predict the transmission paths of data flows according to its application requirements,which accelerates the forwarding speed of data flows and finally improves the intelligentization level of the SDN routing.The research content and main contribution of this thesis can be summarized as following:First,the model of controller node and the switch node are designed delicately.The controller is responsible for network traffic collection and data preprocessing,and then it uses the ANN to learn the transmission patterns from historical data flows.The trained ANN will be delivered to the switches in the form of ANN packets.Once the network topology changes or the link load is too heavy or the request timeout occurs,the controller has to retrain the ANN and generates new ANN packets to meet the requirements of network dynamics.Switches employ the ANN to predict the transmission paths in order to achieve fast forwarding of the data flow.When the transmission path of a data flow fails to be predicted,the ingress switch will request the controller to decide how to deal with the data flow.Then,the efficient mechanism for data collection and preprocessing is developed.The data that needs to be collected and the application classification method are determined explicitly,based on which,the training sample set and test sample set are constructed by the controller.More importantly,the routing prediction mechanisms based on back propagation(BP)neural network and radial basis function(RBF)neural network are proposed and implementated respectively.Further,the BP neural network packet formats and RBF neural network packet formats are also designed.The APC-III algorithm and K-means algorithm are combined and improved to facilitate the determination of the hidden layer neurons of RBF neural network.In order to improve the learning and training effect of ANN and avoid the over-fitting problem,the Cross Validation K-fold algorithm is introduced.Finally,the simulation and performance evaluation are carried out based on the CERNET2 topology.The experimental results show that the proposed intelligent routing mechanisms for SDN in this thesis obtain better performance in the aspect of routing table occupied storage space and the time overhead of the data flow forwarding.