P2P流媒体邻居节点组织与行为预测

Neighbor Peer Organization and Behavior Prediction of P2P Streaming Systems

针对当前流媒体系统用户行为研究缺少单一节点角度的长期行为特性分析及利用的问题,提出一种P2P流媒体邻居节点组织与行为预测策略(POPP).该策略基于流媒体系统中节点长程行为的互相关及自相关特性,通过在数据交互中收集其他节点的在线、延迟及带宽信息来计算邻居节点的综合可用性,维护能够为自身提供优质服务的潜在邻居节点表,并通过基于马氏链的节点行为模型预测邻居节点的未来在线状态.用户日志驱动的仿真实验结果表明,邻居节点组织策略有效提高了邻居节点对自身的服务质量,长期训练后的节点行为预测准确率达97%以上.

A neighbor peer organization and behavior prediction scheme of P2P streaming systems（POPP） is proposed to solve the lack of analysis on single node＇s long-term behavior in current study of streaming system user behavior.The POPP bases on the inter and self behavior correlations of nodes,and collects information of neighbor peers such as behavior,delay and bandwidth in data cooperation process,to calculate their comprehensive availability and to maintain a list of the candidate peers that may provide high quality service.Then,future online status of neighbor peers is predicted by a Markov chain model.Simulations driven by a trace of streaming system show that the neighbor organization of POPP effectively improves the service quality of neighbors,and that the neighbor status prediction accuracy reaches more than 97% after enough trainings.