Abstract: Centralized control over highly programmable networks, or Software-defined Networking (SDN), has enabled data centers to scale to sizes that humans couldn't manage and has allowed wide-area networks to achieve unprecedented utilization levels. Due to SDN's proven benefits, a trend has started to transform all infrastructures, including cellular networks, cyber-physical systems and Internet of Things, into Software-defined Networks. However, due to SDN's centralized nature, SDN increases the risks for outages and makes networking infrastructures less reliable.

 

In this talk, I will describe a couple of directions that my research group has been focusing on to tackle and improve the availability and reliability of the emerging generation of Software-defined Networks. First, I will present work on (re-)designing Software-defined Networking frameworks to support abstractions that decouple failure triggering events and their impact on the infrastructure and the users. These abstractions allow us to safely recover from failures while systematically masking their impact on end users and applications. Second, I will present a principled technique for increasing the efficiency and improving the recovery time of our designs by using static analysis, of the framework’s codebase, to determine where and how to apply our abstractions. Through experiments with realistic Software-defined Networks, I will demonstrate that our approach is 3X times faster than traditional recovery techniques and minimizes application impact by a factor of 8X. In describing my work, I will highlight design choices common to modern Software-defined Networks, and I will conclude by describing future work to apply our approach to other emerging Software-defined Networks.

Short Bio: Dr Benson is an Assistant Professor in the Computer Science Department of Duke University. His research group focuses on solving practical networking and systems problems, with a focus on Software-defined Networking, data centers, clouds, and configuration management. His honors include IBM fellowships, Yahoo FREP awards, Facebook faculty awards, a Yahoo ACE award and multiple award papers at SIGCOMM Internet Measurement Conference and SIGCOMM Workshop: Research on Enterprise Networking.