Distributed upload bandwidth assigning algorithm in P2P VoD system

Abstract In a peer-to-peer (P2P) live streaming system, each peer uses greedy strategy to download chunks as many as possible so as to assure smooth playback and promote the chunk propagation. However, diversity of video playback rate, heterogeneity of peer downlink bandwidth and no prefetching limitation determine that greedy strategy is not the best strategy for P2P video-on-demand (VoD) system though this strategy performs well in P2P live system. To reasonably assign upload bandwidth among peers, we first present a stochastic model to relate expected downloading speed (EDS) with buffered data amount (BDA, the amount of data has been fetched in buffer, however has not been played yet by the peer), video playback rate and fluency (the probability of no interruption in a viewing process). On the base of this model, we design an equal fluency bandwidth assigning algorithm (EFBAA). In this algorithm, requesting peers compute EDS according to several local parameters (such as BDA, video playback rate and fluency) dynamically and independently and send the EDS to neighbor peers together with buffer map. Then the neighbor peer proportionally partitions and assigns its upload bandwidth to requesting peers according to their EDS. Extensive simulations demonstrate that EFBAA outperforms greedy strategy in terms of server load and startup latency.