Peer-Driven Video Streaming: Multiple Descriptions versus Layering

We propose a peer-driven video streaming system that provides a reliable and high-performance video on-demand service. We encode each video into multiple low bit-rate substreams and distribute copies of the sub-streams from the participating peers. Each client receives the constituent multiple sub-streams from different serving peers, and combines them into the original video stream. When a server peer disconnects, the system attempts to find a replacement peer that has a copy of the missing sub-stream and has sufficient available uplink bandwidth. We consider using multiple description coding and layered coding to generate the multiple sub-streams. We develop a traffic theory for peer-driven video streaming, and examine the system performance analytically and via simulation. We show that peerdriven video distribution can provide excellent performance even if peers disconnect from the network frequently. We also compare the performance of layered coding with multiple description coding.