Emerging trends suggest that a significant share of Internet traffic will be due to real-time streaming applications. However, given the unpredictable nature of popularity of any content and the significant swings seen in user demand, provisioning a centralized content distribution system that can guarantee the strict quality of service constraints needed for real-time streaming is a very difficult task. Therefore, this project studies modeling paradigms and systems design for utilizing end-user resources for real-time streaming in a peer-to-peer (P2P) fashion. The rationale is that P2P networks can dynamically adapt to changing network conditions and service requirements more easily than centralized content distribution systems.
This research effort takes a holistic approach in which, on one hand, we study fundamental performance limits of peer-to-peer streaming and design distributed algorithms to achieve these limits. On the other hand, we use these algorithms as building blocks to construct practical protocols that are implemented and validated on an emulation testbed. We bring together a rich set of analytical tools drawn from stochastic processes, optimization and control theory to merge them with real-life multimedia issues of video coding, frame dropping and trans-coding as well as practical implementation issues. Besides the technological impact of developing viable video-streaming ideas for an increasingly networked world, our project also informs and educates both students and the broader society with curriculum development, outreach and integration of under-represented groups.
