Offloading servers with collaborative video on demand

The peer-to-peer (P2P) paradigm provides a data distribution model that may be attractive for Video on Demand (VoD) as it allows to decrease the costs and to increase the scalability of video distribution. However, VoD is more challenging for P2P technology than file sharing or live streaming, and so, practically feasible VoD systems proposed to date rely on a backend server infrastructure as a fail-over solution. In this paper we investigate how the dependency on servers can be decreased by optimizing the video piece-selection strategy and by allowing multiple peers to form a collaboration for obtaining a single video. In a set of simulations of a trace-based system model we show that for systems such as YouTube the proposed optimizations would result in savings of as much as 70% of the server bandwidth. These simulation results confirm the conclusions of an analytical study of our optimizations, the essential part of which is also included in this paper.