Markov Chain Model for Data-Delivery P2P Streaming Applications

In this paper, we proposed a Markov model to evaluate general P2P streaming applications with the assumption of chunk-delivery approach similar to Bit-Torrent file sharing. The state of the system was defined as the number of useful pieces in a peer’s buffer. The model was numerically solved to find out the probability distribution of the number of useful pieces. The central theme of this study revolved around answering the question: what is the probability that a peer can play the stream continuously? This is one of the most important metrics to evaluate the performance of a streaming application. By finding the numerical solution of the Markov chain, we found that increasing the number of neighbors enhances the continuity to a certain threshold, after which the continuity improvement is marginal which complies with empirical results conducted with DONet, a data-driven overlay network for media streaming. We also found that increasing the buffer length increases the continuity but there is a trade-off because peers exchange information about the buffer map, hence increasing the buffer length increases the overhead. Finally we discussed the continuity for both homogeneous and heterogeneous peers regarding the uploading bandwidth.