Characterization, adaptive traffic shaping and multiplexing of real-time MPEG-II video

We obtain a network traffic model for real-time MPEG-II encoded digital video by analyzing video stream samples from real-time encoders from NUKO Information Systems. MPEG-II sample streams include a resolution intensive movie, City of Joy, an action intensive movie, Aliens, a luminance intensive (black and white) movie, Road To Utopia, and a chrominance intensive (color) movie, Dick Tracy. From our analysis we obtain a heuristic model for the encoded video traffic, which uses a 15-stage Markov process to model the I, B, P frame sequences within a Group of Pictures (GOP). A jointly correlated Gaussian process is used to model the individual frame sizes. Scene change arrivals are modeled according to a Gamma process. Simulations show that our MPEG-II traffic model generates I, B, P frame sequences and frame sizes that closely match the sample MPEG-II stream traffic characteristics as they relate to latency and buffer occupancy in network queues. To achieve high multiplexing efficiency we propose a traffic shaping scheme which sets preferred I-frame generation times among a group of encoders so as to minimize the overall variation in total offered traffic while still allowing the individual encoders to react to scene changes. Simulations show that our scheme results in multiplexing gains of up to 10% enabling us to multiplex twenty 6Mbps MPEG-II video streams instead of 18 streams over an ATM/SONET OC3 link without latency or cell loss penalty.