一种基于访问概率预分配的流媒体集群动态副本更新算法

研究了同构流媒体集群服务器环境下的动态副本放置问题,提出一种请求调度和动态内容部署相结合的新颖策略.首先采用滑动窗的方式预测各文件的点播率,以平衡服务器之间的实时累积访问概率为目标,在不进行内容部署的情况下,预分配相同文件不同副本的访问概率;同时进行零迁移代价的动态内容更新,从而能够提前反映出请求的实时倾向性,在降低系统请求拒绝率的同时,有效地减少了请求被分发到过载服务器后产生负载迁移的频度和代价.仿真分析的结果表明这种策略对于流媒体服务系统的研究和实践具有指导意义.

A dynamic replica updating strategy of clustered streaming media system based on accessing probability pre-allocation

The dynamic replica placement problem of a clustered streaming media system composed of homogeneous servers was discussed,and a novel strategy to combine the request scheduling with dynamic content distribution was also presented.Firstly,sliding time window is adopted to predict the relative popularity of all video files.In order to balance the real accumulative accessing probability of each server,different replicas of the same video file are allocated various accessing probabilities under the circumstance of unchanged content distribution.At the same time,a so-called zero migration cost dynamic replica placement strategy is brought forward to embody the real request orientation.Not only the system rejection probability decreases,but also the frequency and cost of request migration caused by too many requests scheduled to already heavily loaded servers are greatly reduced.The simulation analysis showed that the proposed strategy will benefit research and practice of streaming media system.