A Data-driven Approach to Speeding Up the Internet
Despite the growing popularity of video streaming and web services over the Internet, problems such as re-buffering and slow page loads continue to plague users. In this talk, I will present a data-driven approach to understanding these problems and analyzing the potential viability of different design choices. I will discuss three projects: First, I will present an end-to-end characterization of Yahoo's video streaming service, analyzing over 500 million video chunks and demonstrate how marrying client side and server side data enables more power diagnosis. Our framework can detect a wide range of new and undetectable problems that can cause re-buffering: I will discuss a subset of these problems. Second, I will present work to proactively reduce web latency by using a data-driven approach to tailor CDN configurations to the customer’s conditions. Initial prototype and large-scale simulations demonstrate that our framework can significantly improve performance. Finally, I will present ongoing work to understand the role of mobile phones endemic within developing regions on web performance. I will show that these developing regions occupy a unique point in the design space enabling us to revisit traditional approaches to web performance. In each component, I will discuss the broader implications of employing a data-driven approach.
Theo is an assistant professor in the Department of Computer Science at Brown University. His group works on designing frameworks and algorithms for solving practical networking problems with an emphasis on speeding up the internet, improving network reliability, and simplifying network management. He has won multiple awards including best paper awards, an applied network research prize, various Yahoo! and Facebook Faculty Awards, and an NSF Career award.