Design And Implementation Of Intelligent Traffic Scheduling Platform For SDN

In recent years,with the development and gradual maturity of optical transmission,data communication technology,5G,NFV,SDN network and other technologies,whether it is the bandwidth provided by the network data plane or the control plane,its finegrained control capability and flexibility are very high.high.With the development of the Internet,many new Internet applications,such as high-definition video,AR/VR,Io T,etc.,are also emerging in large numbers.Video traffic accounts for more than 60% of the global Internet downstream traffic.These massive video traffic require a lot of network resources and also require strict quality of service.For example,the flow rate of 1080 P online high-definition video is 3500 kpbs,and the flow rate of 4k online high-definition video is about 25 Mbps,while the bandwidth requirement of strong interactive AR/VR video services will reach 100 Mbps,and the transmission delay needs to be less than 20 ms.In addition,in recent years,the Internet of Things technology has begun to be applied on a large scale,and the various terminals connected to the network are also increasing rapidly.This thesis mainly studies the intelligent traffic scheduling mechanism and platform of SDN network.The main work is as follows:1.The traffic classification mechanism based on machine learning is studied.This classification divides traffic into video,text,voice and other categories to provide guidance for subsequent intelligent traffic planning.The neural network used is the CNN convolutional neural network.CNN can learn the relationship between adjacent elements of the matrix,and the packet characteristics of the traffic can be arranged into a sequence of numbers.It is more appropriate to use CNN for traffic classification research.The simulation results show that the traffic classification mechanism based on CNN has the characteristics of faster calculation speed and higher accuracy than traditional algorithms.2.The intelligent routing model based on machine learning is studied to make routing decisions for traffic.The machine learning model used is a fully connected neural network,and the training data is jointly generated by the traditional routing algorithm K shortest path algorithm and the brute force load balancing algorithm.The training data set integrates various Qo S requirements such as low latency and load balancing.Each pair of nodes has 8 alternative paths stored in the path database marked by binary labels,and the neural network output is the label of the path.Realize the load balancing of video streams and the low latency requirements of voice and text streams.The simulation results show that the traffic planning algorithm based on machine learning has greatly improved in terms of load balancing,delay and packet loss rate.3.Use onos and mininet software architecture to build a large-scale SDN simulation network with 100 nodes,and use wireshark to collect real traffic in the real network and store it as pcap files.These files not only provide data support for the research of intelligent traffic classification,but also play back traffic.transmission in the system.When designing the traffic playback system,the Go language is used to write the playback program,and the collected traffic is played back in the simulation platform using libnet,libnids and other technologies.