一种基于拥塞发现的强化学习P2P网络视频点播预取策略  

Video Prefetching Strategy Based on Congestion Finding with Reinforcement Learning in P2P VOD Networks

作　　者：沈项军[1] 姚银[1] 查正军[2] 

机构地区：[1]江苏大学计算机科学与通信工程学院,镇江212013 [2]中国科学院合肥智能机械研究所,合肥230031

出　　处：《计算机科学》2015年第7期258-261,275,共5页Computer Science

基　　金：国家自然科学基金项目(61005017)资助

摘　　要：在P2P(Peer to Peer)网络视频点播中设计合理的视频内容预取策略,可以有效地解决视频观看中出现的缓冲等待时间过长、服务器负载过大等问题。而现有的P2P网络视频预取方法往往考虑了视频内容的发现而忽略了节点本身的状态监控,在网络拥塞发生时存在视频播放观看不佳的情况。提出了一种基于拥塞发现的强化学习P2P网络视频点播预取策略,通过监测节点的拥塞状况以及带宽等参数,采用Q-learning学习算法综合评估网络节点,引导视频预取节点的选择,减少了对网络拥塞节点视频内容的预取。实验结果表明,该方法能提高视频播放流畅性,避免用户在节点发生拥塞时等待时间过长,提高了视频播放效率。A proper video prefetching strategy in P2P（peer to peer）VOD（Video on Demand）networks can effectively solve such circumstances as waiting time for buffering is too long and server is overloaded.While the existing video prefetching methods in P2 Pnetworks just consider video content finding but ignore monitoring nodes status,which makes poor video play when the network congestions occur.This paper presented a video prefetching strategy based on congestion finding with reinforcement learning in P2 P VOD network.Through monitoring nodes status of congestion,bandwidth and other parameters,the Q-learning learning algorithm was used to evaluate network nodes,which can guide nodes selection and reduce the video prefetching from congested nodes.Simulation results show that the proposed method can make video playback smooth.It can also avoid too long waiting time when congestions happen and improve the efficiency of video playback.

关 键 词：P2P网络 视频预取 拥塞发现 强化学习 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]