alt text 

Stanford University Networking Seminar

Ratul Mahajan (Microsoft)
Software-defined networking : Change is hard

12:15pm, Thursday, May 15, 2014
Gates 104

About the Talk

We developed SWAN (software-driven wide area network), a system that increases the utilization of inter-data center networks to almost 100%, from the current typical levels of under 50%. It boosts utilization by frequently changing the data plane state (i.e., how it forwards packets) of the network based on current traffic demand. A major challenge in SWAN, faced in other SDN systems as well, is to quickly and safely change network state. Safety implies maintaining invariants such as packets should not loop or absence of transient congestion. The transition challenge stems from the fact that when many switches need to be updated, there are inevitable delays and failures in updating some of them, and even if all switches respond promptly, it is difficult to simultaneously (i.e., atomically) change the state of multiple switches.

I’ll describe three techniques to enable fast, safe transitions in SDNs: 1) Multi-step updates: using scratch link capacity to update the network in a provably congestion-free manner; 2) Forward fault correction: spreading traffic in the network such that the absence of congestion is guaranteed as long as the number of switches with update faults is below a configurable bound; and 3) Dynamic update ordering: determining the order of switch updates dynamically, based on current conditions, to finish updates quickly. Experiments using real hardware and using large-scale simulations demonstrate the efficacy of these techniques.

About the Speaker

Ratul Mahajan is a Senior Researcher at Microsoft Research and an Affiliate Professor at the University of Washington. His research interests include all aspects of networked systems. His current work focuses on smart home technologies, software-defined networks, and network verification, and his past work spans Internet routing and measurements, incentive-compatible protocol design, practical models for wireless networks, and vehicular networks. He has published over 40 papers in top-tier venues such as SIGCOMM, NSDI, SOSP, MobiCom, and CHI. He is a winner of the ACM SIGCOMM Rising Star award, the William R. Bennett prize, the SIGCOMM best paper award, and Microsoft Research Graduate Fellowship