Multicasting for wireless video-on-demand

We consider a video delivery problem in which a server multicasts a single video to several users over a wireless broadcast channel. Users arrive and depart asynchronously, and when a user arrives, he or she wishes to watch the video from its beginning. Under a natural stochastic model for this problem, the optimal scheduling algorithm can be described by dynamic programming, but it is too complex to be practical. We describe a suite of approximations to the optimal scheme that lead to practical scheduling algorithms. The algorithms work with partially observed state, multirate multicast, and a variety of cost functions. Numerical results show significant performance gains over previously known schemes.