Special Seminar: Balakrishnan Chandrasekaran "A Two-Part Tale on Network Congestion"

A Two-Part Tale on Network Congestion

Balakrishnan Chandrasekaran

Abstract: Network congestion occurs when the traffic or data sent through a communication link exceeds the capacity of that link. Congestion deteriorates the performance of applications using the network. Video traffic constitutes the majority of today’s Internet traffic and, hence, it is likely to suffer from congestion. In the first part of this talk, I discuss Slipstream, a radical approach to video streaming using a partially reliable transport. To combat congestion, Slipstream exploits the insight that not all bits of a video need to be reliably delivered for the user to enjoy a high quality video. In our evaluations, SlipStream outperforms the state-of-the-art in video streaming under a wide range of network conditions, even under stringent playback-buffer sizes, without compromising the perceived video quality. 

The availability of very high-speed links, the decrease in buffer sizes, and the increase in traffic diversity coupled with the stringent goals make congestion control a notoriously hard problem. In the second part, I illustrate how to exploit recent advancements in programmable data planes to overhaul the congestion-control algorithms in end hosts. I will outline the design of a novel network-assisted (explicit) congestion feedback and its potential to cope with small buffers.

Bio: Balakrishnan (Bala) Chandrasekaran received his Ph.D. in computer science from Duke University in Durham (North Carolina), US in 2016 under the supervision of Bruce MacDowell Maggs. He is currently a Senior Researcher at the Max Planck Institute for Informatics and heads the Networked Systems group. His interests are in network measurements and design, with a focus on the performance and security aspects of networked systems. In 2017, his work won the Best Dataset Award at the Passive and Active Measurement Conference (PAM), and the Distinguished Paper Award at the USENIX Security Symposium.