| Even after many years of research, high-quality video-on-demand is still a challenging service that demands a lot of attention. High-definition video requires a high data rate, and the asynchronous nature of video-on-demand requests make data distribution more challenging to optimize. As a result, these services can consume a considerable amount of bandwidth. This is particularly relevant to U.S. telecommunication companies, like the cable, telephone, satellite, and cellular industries. With the growing prevalence of high-bandwidth access networks, these companies are beginning to compete over multimedia services. In doing so, they are placing considerable strain on their network infrastructure, and thus are a rewarding topic for network-specific system optimizations.;Peer-to-peer systems are a potentially lucrative model to improve the efficiency of such a video-on-demand service over these networks. These systems have been shown to reduce traffic on the network core, scale well with increases in subscribers, and decrease reliance on expensive, specialized video servers. Telecommunications companies have been reticent about peer-to-peer systems, as many deployed versions are irresponsible network users and often providing illegal or competing services. However, many recent works have shown that peer-to-peer systems are considerably more efficient when tailor-made for a specific infrastructure. Also, the growing prevalence of set-top boxes, like TiVos and DVRs, provide an option to place a cable-owned device to serve as a peer in each home.;This dissertation contributes to a growing body of research concerning internally controlled peer-to-peer systems tailored to their underlying access network. It describes a set-top box based peer-to-peer service built on top of the hybrid fiber-optic and coaxial star-bus topology used by the U.S. cable industry. By efficiently using the broadcast-based coaxial cables at the edge of the network, the set-top boxes are able to share data with each other efficiently to enhance video-on-demand services. We show that these methods can reduce the traffic load on expensive fiber-optic networks and specialized video servers by limiting traffic to the cheaper coaxial lines.;We extend our work across three important axis to present a comprehensive evaluation of our system. First, while our results are primarily based on a trace of video-on-demand usage from a deployed system, we also explore a number of mathematically generated usage models. This allows us to explore some of the assumptions we make about usage patterns, and better understand where our system might break down. Also, because the cable industry is concerned with high quality-of-service, we present a number of mechanisms to maintain a good viewing experience despite I/O overloads in the system. Finally, we explore the importance of the broadcast technology at the network edge, both to cable and other industries, and advocate for its utility in these types of systems. |
