Research And Application Of Distributed Control Platform In Software Defined Networking

Software defined networking (SDN), where network control is decoupled from forwarding and is directly programmable, has been put forward in the last several vears. The new form of network architiecture has been proving the effectiveness, management convenience and enables the programmability.Control platform plays an important role in SDN. However, traditional SDN deployments relying on single centralized controller will limit the scalability of network in large scale network. Distributed control platform, which has multi control nodes in the control platform, is the main technology to address this problems. Therefore, research on distributed control platform has important significance. This paper focus on the research and application of distributed control platform in software defined networking and includes following work:(1) Proposed a SDN control platform architecture based on hierarchical and distributed structure, which divided the control platform into two layer vertically. And the control platform contains upper control nodes and lower control nodes. The architecture is used for improving the scalability of SDN. Afterwards, we also discuss the implementation of the interface between different upper control nodes, the interface between upper control node and lower control node and the interface between lower control node and forwarding node. (2) Designed an upper control node election algorithm base on Paxos. As the upper control nodes are important and less in amount, the algorithm can elect the upper control node if it fails and recover the system rapidly. The result of experiment indicated this algorithm can deal with the upper control node failure effectively.(3) Designed a dynamic mapping allocation algorithm between lower control nodes and forwarding nodes based on cooperative game theory. As the lower control nodes are interact with forwarding plane directly and great in amount, the balance of the load of lower control nodes is important. This algorithm can balance the load of the lower control nodes. The experiment indicated the algorithm can reduce the CPU usage variance of lower control nodes and balance the load.(4) Compared the time complexity of network view obtaining in different distributed control platform architecture with global view transaction and local view transaction. The mathematical proof shows the distributed control platform with hierarchical structure proposed has better time complexity. Then this paper carried out an experiment with Cbench and the result indicated the purposed architecture has performance advantages on throughput compared to single centralized controller and distributed control platform without hierarchical structure.