Keynotes

Prof. Deke Guo

Keynote Topic: Resilient Control Plane of Software-defined Networks

Bio:

Deke Guo is a Professor in the College of System Engineering, National University of Defense Technology. His research interests include distributed systems, computer networks, data center networking, mobile computing, and interconnection networks. He has published more than 120 papers in various journals and conference proceedings, such as JSAC, TON, TPDS, TKDE, TC, INFOCOM, ICDCS. He is a distinguished member of CCF, and a senior member of the IEEE. His work was supported in part by the National Natural Science Foundation for Outstanding Excellent Young Scholars of China.

Abstract:

The design of logically centralized control plane is an essential part of a software-defined networking (SDN). In the setting, multiple distributed controllers are deployed to offer a control plane over the entire network to efficiently manage the network usage. Despite such efforts, SDN, still lack a scalable and resilient control plane. In this talk, we will show how to enable a resilient control plane from three orthogonal aspects.

First, we report the minimal fault-tolerant coverage problem of controllers. It aims to improve the fault-tolerant capability of the control plane using the least number of controllers, such that each network device would be served by at least two controllers. Second, we reveal that each controller usually exhibits a skew distribution of routing response latency with a long tail. Cutting the long-tail latency of response is critical to enable the resilient control plane, yet difficult to achieve due to many factors.

We propose a general delay aware switch-to-controller selection scheme to fundamentally cut the long-tail response latency for the more complicated heterogeneous controller scenario with performance fluctuations. Finally, little is known about how to validate that the control plane offers assurable performance, especially across various failures. We develop an optimization framework to derive the worst-case performance of the control plane. This framework can accommodate a rich set of failure recovery strategies, each of which re-elect the appropriate master controllers for disconnected switches under any failure scenario.