HTTP-Based Adaptive Streaming for Mobile Clients using Markov Decision Process

Due to its simplicity at the server side, HTTP-based adaptive streaming has become a popular choice for streaming on-line contents to a wide range of user devices. In HTTP-based streaming systems, the server simply stores the video segmented into a series of small chunks coded in many different qualities and sizes, and leaves the decision of which chunk to download next to achieve a high quality viewing experience to the client. This decision making is a challenging task, especially in mobile environment due to unexpected changes in network bandwidth as the user moves through different regions. In this paper, we consider Markov Decision Process (MDP) to derive the optimum chunk selection strategy that maximizes streaming quality and propose three approaches to reduce the computational cost of MDP. The first approach recomputes the solution after downloading every k chunks. The second approach computes the solution once using global network statistics of a given region. The third approach recomputes the solution every x meters using offline statistics for each x meters of the road. The three approaches are compared using real-world 3G bandwidth and mobility traces. The best performance is achieved with x-MDP.