Title:  Building Scalable OpenFlow Networks

Committee:

Dr. Douglas Blough, ECE, Chair , Advisor

Dr. Henry Owen, ECE

Dr. Russell Clark, CoC

Dr. Yusun Chang, ECE

Dr. Richard Fujimoto, CSE

Abstract:

Software Defined Networking (SDN) is widely regarded as the next generation networking technique, which can create programmable, flexible, and agile networks whilst reducing costs. The core of SDN is to separate and logically centralize network control from its data plane. To achieve this separation, most SDN implementations use the de facto southbound protocol OpenFlow as the communication interface between the control and data planes. However, the scalability problem bottlenecks the deployment of SDN OpenFlow for large networks. The roots of SDN OpenFlow scalability problem are the centralized architecture of the control plane and the unmatched capabilities of OpenFlow switches to deal with the massive events generated by the fine-grained granularity control mechanism. The objective of this thesis is to address the fundamental problems of scaling SDN OpenFlow networks. On the control plane, this work first investigates the scalability performance of all existing non-centralized control plane architectures versus the centralized one in order to answer the question, ``Which control plane architecture scales best?" The simulation results show that the hierarchical control architecture and the peer-to-peer control architecture with local view  (a.k.a., distributed control plane) are the most two scalable control architectures. With this conclusion, this work then aims to address the most important challenge, eastbound/westbound interface design, for the distributed control plane, which is more feasible for scaling across geographies than the hierarchical one. As for the data plane, this research works around the hardware manufacturing related limitations such as CPU and bus bandwidth by improving the utilization of the existing precise hardware resources and reducing control overheads to mitigate the scalability problem. Specifically, machine learning techniques are exploited to improve proactive flow entry deletion and flow entry eviction. The provided theoretical analysis and simulation results in this thesis lay out the foundation for the deployment of large scale SDN OpenFlow networks.