【作者】 马军；

【导师】 殷保群；

【作者基本信息】 中国科学技术大学 ， 控制理论与控制工程， 2009， 硕士

【摘要】 随着计算机技术的不断发展以及网络带宽的不断提高,流媒体服务被迅速推广。据最新的网络流量统计显示,流媒体服务已经成为当今互联网流量的主体。目前研究具有高承载能力,高可扩展性,高稳定性,低成本的可大规模商业化运营的流媒体内容分发系统成为全球流媒体应用领域的热点。与基于传统C/S架构的系统,基于CDN架构的系统,以及基于P2P架构的系统相比,近年来提出的基于融合CDN和P2P架构的媒体内容分发系统具有更好的优越性和发展潜力。本文基于这种最新提出的媒体内容分发系统,研究了其中的动态数据部署算法这一关键技术。本文的研究工作主要包括下面几个方面:1.由于融合CDN和P2P技术的媒体内容分发系统比较复杂,本文把系统分成两层,即骨干层和边缘层,分别分析了每层要解决的主要问题。2.发现两层要解决的问题都是NP类问题,本文基于部分可观Markov决策过程(POMDP)以及基于观测的策略迭代优化算法和策略梯度优化算法分别给出了每层的动态数据部署算法。3.为了验证算法的性能,本文利用计算机仿真技术对一个具体媒体内容分发系统从简单情况到复杂情况逐步进行了仿真。4.对仿真结果进行分析和讨论,总结出一些规律和结论。本文首次提出把POMDP的相关理论应用到媒体内容分发系统的动态数据部署算法的研究上,希望本文的研究工作能够拓展数据部署算法的研究思路,为该方面的进一步研究提供一些基础。更多还原

【Abstract】 With the ever-increasing development of computer technology and improvement of network bandwidth, streaming media service is rapidly and widely used. According to the latest data of network flow, streaming media service has already been the main part of network flow nowadays. Currently, studies on content delivery network system of streaming media with high carrying ability, high expandability, high stability and low cost which can be commercially operated on a large scale has been a hotspot in the applied field of global streaming media. Compared with systems based on traditional C/S, CDN or P2P, the media content delivery system based on the combination of CDN and P2P put forward in recently years is of more advantages and development potential. The present paper, on the basis of the latest media content delivery system, examines one of the most important technologies: dynamic data deployment algorithm.The present paper includes studies on the following aspects. Firstly, media content delivery system is very complex because of combining CDN and P2P, and the present paper divides the system into two layers (backbone layer and edge layer) and analyzes main problems to be solved in each layer respectively. Secondly, it is found that NP issues are the common problems to be solved in each layer and the present paper proposes dynamic data deployment algorithms for each layer respectively on the basis of POMDP, policy iteration algorithm and policy gradient algorithm. Thirdly, in order to verify the performance of the algorithm, the present paper employs computer simulation technology to conduct simulation on a concrete media content delivery system gradually from simple condition to more complicated ones. Fourthly, the present paper analyzes and discusses the simulation results and makes some conclusions.The present paper applies relevant theories of POMDP to the study of dynamic data deployment algorithm of media content delivery system for the first time and endeavors to broaden the study thoughts of data deployment algorithm and lays a little foundation for the further study of the field.更多还原