ALBPS: An Adaptive Approach on Reduction Redundancy in Latency-Aware Peer-to-Peer Content Delivery System with Network Coding

In recent research on the Peer-To-Peer (P2P) content delivery system, network coding has been applied as a promising approach to improve the performance of P2P content delivery. Many studies have shown that the live P2P streaming system can take great advantage of network coding. It has also been shown that this technology can also resolves the "last block" problem. Accordingly, P2P seed peers can just simply upload the coded data to the downstream peer without being applied a complicated schedule algorithm. This results in a very high utilization of bandwidth as well as the throughput since the seed peers can contribute their bandwidth as much as they can.However, it has been found that in a network coding delivery procedure for requested segment, the seed peers (serving peers) cannot decide wether they should stop pushing coded block to the downstream peer or not. The difficulty is that the seed peers cannot get the decoded state instantly from the downstream peer. It cause a significant redundancy, named redundancy caused by late stoping signal (RLSS), especially when the upload bandwidth is high. We propose an approach to reduce such RLSS and preserve the property of high utilization and throughput though. Such approach is named as adaptive learning-based predictable stoping (ALBPS). In this approach, seed peers predict their uploading bandwidth allocation for each segment peer by learning from the history and allocate the bandwidth via assigning priority to each segment uploading session. We also demonstrate that the RLSS can be reduced via applying the ALBPS by both simulation and reality.