Tuning performance of P2P mesh streaming system using a network evolution approach

Resilience and startup delay are the most important performance metrics to evaluate the P2P streaming systems. To simultaneously improve the two metrics, we propose several mechanisms at different system evolution stages. At the first stage, media server encodes the stream into multiple sub-streams of the same length. Redundancy is introduced by using Reed-Solomon (RS) coding before distributing the sub-streams to different successors. Each peer in the network establishes a cooperative relationship with others to obtain all required sub-streams. At the stage of new peer arrival, a parent selection algorithm with relatively lower complexity is proposed which takes full advantage of redundant coding. After the peer builds up streaming transmission, it replaces some parents with a latency-based decision mechanism. In case of node failure, a swap-in-turn repairing algorithm between different sub-stream sources is proposed to ensure the high continuity of steaming transmission. Simulation results show that 1) the redundant coding and the parent replace algorithm in case of node failure can effectively reduce interruption of data streams; 2) the codes with higher redundant degree can adapt to more dynamic scenario. Meanwhile, the codes with redundancy does not significantly decrease the effective transmission ratio when network is dynamic; 3) transmission achieves higher performance when the number of substreams is between 8 and 16; and 4) the parent switching mechanism can significantly decrease the startup latency for a big proportion of peers.