A Research On SDN-WAN-based Routing Method For LEO Satellite Network

Able to make up for the defects of terrestrial communication system which cannot cover the whole world seamlessly,satellite communication will become an important part of the future network.Low Earth Orbit(LEO)satellite network has become one of the future network technologies that developed countries,represented by the United States,are focusing on because of lower cost and faster data transmission.However,LEO satellite network topology as well as traffic transferred on the inter-satellite link changes rapidly,which may cause network congestion and QoS performance degradation.To solve this problem,it is a must to optimize flow routing dynamically according to the characteristics of satellite network,which requires network control plane to be equipped with a global state view and flexible control capability.Software Defined Networking(SDN)fits this situation very well,so this thesis focuses on the routing optimization method of LEO satellite networks based on SDN.In order to avoid interruptions caused by high-speed changes in LEO network topol-ogy,this thesis first designs a static routing optimization algorithm based on virtual topol-ogy and K-path-routing.It takes advantage of an auxiliary graph to convert link switching cost into the weight of edges in the graph,so as to give the routing sequences with the least number of link switching between each pair of source and destination nodes.The re-sulted routing produced by the algorithm will be set as the default routing of LEO satellite network.However,static routing cannot adapt to the dynamic changes of traffic volume,which could lead to network congestion.In this thesis,a routing optimization algorithm with low time complexity is designed.The algorithm is based on the idea of neighbor-hood search and is able to give optimized solution in a short time.Experiments show that the algorithm this thesis proposed outperforms other algorithms in load balancing and can effectively alleviate the problems of increasing transmission delay as well as packet loss rate caused by network congestion.Then,a supervised learning based routing optimization method is also proposed in this thesis in order to quickly obtain the sub-optimal routing solution when the traffic volume in LEO network changes.To scale the proposed model well,this thesis splits it into several smaller ones,and then reduces the time consumed by training and optimization by means of parallel deployment of these smaller models.Simulation results demonstrate that this method can quickly offer the sub-optimal solutions and effectively alleviate network congestion.Especially in the case of large traffic volume,this method outperforms the traditional shortest path and online algorithm regarding load balancing and as a result ensures the network transmission performance.Finally,this thesis develops a LEO satellite SDN-WAN prototype validation system to make the simulation environment reflects the reality as well as possible and thus we could evaluate the performance of the two routing optimization methods described above more accurately.It is composed of three subsystems:distributed dataplane,traffic injec-tion and real-time QoS analysis.Experiments show that the distributed dataplane designed by this thesis scales very well;the traffic injection subsystem can replay flows of arbitrary characteristics;the QoS analysis system can perform rather accurate estimation of end-to-end transmission delay as well as packet loss rate of flows in real time.