The Research Of SDN Traffic Prediction Based On Deep Learning

With the rapid development of Internet technology,people become more and more dependent on the Internet.The number of applications on the Internet has increased exponentially,this is accompanied by an unusually large flow of data.Unfortunately,the speed of development of the network architecture equipments is far from the network traffic,the imbalance will lead to congestion of network link frequently,user's Internet experience will be affected.Software Define Networking is considered to be the next generation of network architecture,it's controller can issue instructions underlying network equipment operation,the underlying network's operation of the equipment is completely fool type,the centralized control can reduce redundancy,improve efficiency.Based on the above status quo,this thesis will study SDN traffic based on deep learning technology to solve the problem of network traffic prediction.First of all,the thesis introduces the research background and current situation of network traffic prediction,and elaborates the related technologies in detail.Then,came up with a SDN traffic prediction framework based on deep learning,the prediction mechanism of this prediction framework is fully researched and explored,the functions of each module involved in the framework are studied,make modules cooperate with each other to finish the work of network traffic prediction.Finally,based on the TFTS(TensorFlow Time Series)module,a deep learning model suitable for SDN traffic prediction is studied.In this thesis,a large amount of traffic data of SDN environment is collected,and enter it into the deep learning model to training and predicting.The results shows that the prediction performance of the improved TFTS module based on Long Short-Term Memory(LSTM)achieves the highest accuracy of 93.48%what the traffic data is collected once every 30 minutes on the 14 days.The accurate results of traffic prediction can be used as the basis for judging and processing the corresponding network problem of lower-level equipment and upper-level applications.