Optimization Of The Distributed Controller System In Software Defined Networks

Software Defined Networks(SDN) draw a lot of attention from both industry and academy, and thus have substantially broadened SDN researches and products. The key characteristic of SDN is programmability by the virtue of separating control and forward functions. With the programmable control plane, SDN achieves an encouraging improvement on network management. Programmers are able to design network strategies, and SDN benefits from adjusting network strategies in a dynamic manner. Therefore, SDN stands as a fundamental platform for evaluating new protocols and innovations presented by both researches and vendors.In paradigmatic SDN scene, the control logics are derived from physical devices and deployed on the individual controller(also called network operating system) in a centralized manner. The research on controller is really hot due to its importance for network management. Distributed Controller system acts as the inevitable choice for managing large scale network. There is thus a burst on paradigms of distributed controller. When the control distribution structure is implemented as multiple physical nodes, a new issue shows up as the systemic resource management for distributed controllers. Current solution for distributed controllers have not been able to tradeoff among multiple resources while reducing unnecessary overhead. An ideal solution is supposed to deal with comprehensive indicators in a global view so that the controllers can effectively manage the practice and status of network.In this dissertation, two typical distributed controllers’ architectures are fully analyzed, and a systemic model with several primary indicators is proposed for depicting the overhead and availability of controller systems. Furthermore, we design an integrated solution for the purpose of optimizing the performance of distributed controller system. The solution consists of application management and network view management. This research is based on the scalable Openflow distributed controllers structure. Many resource indicators such as storage, synchronization traffic and response time are concerned. According to the online monitoring data, the distributed controller system optimized performs better on strategy execution.This research takes full advantage of SDN’s programmability in order to improve the effectiveness of resources. Comparative experiment results validate the availability overhead and trigger condition of recalculating optimization strategy. In the end, future work towards a practicable direction are discussed.