Video-on-demand (VoD) is an attractive service that has already gained popularity in the Internet by allowing users to view a video from a large catalog at any time. A limitation of existing VoD services, both using traditional design as well as P2P is their inability to support efficient streamed access to ALL content at high quality, regardless of demand: not just blockbusters that millions want to see, but also niche content that has relatively small demand in comparison to their highly popular counterparts: re-runs of TV shows, non-blockbuster releases, replays of sporting events, and speeches.
The peer-to-peer (P2P) approach is an effective solution for scalable content distribution without imposing a significant burden on a centralized infrastructure. In a P2P-based VoD system, users receive streamed videos from VoD servers as well as from the peers. The ability of peers that are viewing the videos to collaborate with each other reduces the load on serving infrastructure.
This project extends the design of successful P2P VoD systems to enable support for efficient delivery of a large, potentially limitless library of high (movie) quality stored VoD. Regardless of a video's popularity, a high-quality version will be playable by any client attached to the network without interruption within seconds of its request. The research utilizes three design principles: leverage P2P technologies, implement altruistic cross-content caching and serving, and utilize playback-point weighted caching strategies. Results include models that describe desirable caching behavior, distributed algorithms to implement the desired caching behavior, and a functional prototype.
Our research supported by this grant could be divided into two major subproblems. Using peer to peer technology to support efficient delivery of videos that are less popular, and improving on infrastructure based delivery for a group of hetergenous clients by enhancing delivery with peer to peer technology. For the first problem, we showed how, by providing a small additional amount of cache on each peer machine, less popular videos can be efficiently supported. For the second work, we showed how adaptive bit rate video (ABR) and peer to peer technology can significantly improve the performance, meeting the needs of individual clients, while using a small number of servers. The intellectual merit resulted in analytical models, their analysis, and simulation of the various systems. These models and simulators can be used by fellow researchers to further explore our results. The funding helped support two PhD students and trained them in research in this important area. The broader impact was to affect and imrpove VoD delivery. Various organizations, including AT&T (our collaborator) and Time Warner have expressed interest in the research, and we are pursuing further discussions so that the results from our research can be accessed and utilized by these organizations to improve video-on-demand (VoD) quality while reducing costs. One of the students funded by this award was female.
