【作者】 张建伟；

【导师】 鲁东明； 邢卫；

【作者基本信息】 浙江大学 ， 计算机应用技术， 2013， 博士

【摘要】 P2P(Peer-to-Peer)内容分发技术已发展成为当今互联网上最具影响力的技术之一。相较于传统的C/S架构及CDN架构的内容分发技术,P2P技术具有性能优越、扩展性强、易于部署等优势。其中,BitTorrent已成为应用最为广泛的P2P系统,基于P2P技术的文件共享与实时流媒体服务均已成为互联网流量占用最大的应用类型之一然而,由于P2P网络中节点上传带宽有限且各不相同,不同类型的应用或终端用户对带宽、延迟、业务持续时长等有不同的要求,如何为终端用户提供较高的服务质量仍面临着诸多挑战和难题。基于上述原因,本文围绕如何分析和优化P2P网络中一点到多点的内容分发效率这一问题,针对BitTorrent系统以及文件共享与流媒体直播应用中所涉及的重叠网络拓扑构建、带宽分配、数据调度等关键技术展开了研究,主要工作和研究成果体现在以下三个方面：首先,以节点的上传带宽信息为基础,推导出片段扩散模型及上传带宽利用率模型,并提出了改进的拓扑构建机制和激励机制。推导出的模型有助于更深刻理解BitTorrent系统中TFT周期、最大并行上传数、片段大小等重要参数的作用；提出的基于上传带宽的拓扑构建机制UC-track算法有利于提高BitTorrent网络的拓扑特性和传输效率；提出的包含带宽限制策略与带宽分配策略的激励机制PU算法不仅提高了BitTorrent系统的公平性,同时也大幅度减小了源服务器的负载。其次,将BitTorrent系统泛化为一般的P2P系统。针对文件共享应用,以最终完成时间为系统性能优化目标,在节点带宽异质的情况下,研究了如何合理利用辅助节点的空闲带宽及数据资源以提高系统性能,为任意给定的目标节点集合提供最优的下载服务。针对文件共享的区分服务问题,在节点带宽完全异质的情况下,对P2P网络中的文件传输机制建立了扩展的均等服务模型和区分服务模型,得出了最终完成时间的最优值,并且提出了达到此最优值的优化算法。另外,从分析和仿真两方面对比了基于流的传输模型与基于块的传输模型应用于P2P文件传输中的主要区别,得出了扩展模型的适用条件。利用得出的扩展模型,可实现在P2P网络中对任意划分的节点集合进行任意等级的文件共享区分服务。最后,针对实时性要求更高的流媒体直播应用,以播放延迟与播放概率为系统性能优化目标,在节点带宽同质的Mesh形拓扑条件下,将传输方案分为片段优先方案、节点优先方案与Epidemic方案三类,建立了基于Pull方式的流媒体传输方案分析框架,提出了片段选择策略、节点选择策略、请求选择策略精确的模型；通过改进的分段选片策略与随机推送策略,提出一种基于Push方式的流媒体直播优化策略LR2,使得流媒体直播与文件共享可以在同一框架下优化。本文的结论有着重要的理论和应用价值,一方面可以对P2P网络内容分发问题中带宽分配及数据调度技术提供深入的认识,另一方面可以对提高P2P系统的性能提供有价值的参考。更多还原

【Abstract】 Nowadays peer-to-peer (P2P) content distribution has become one of the most influential technologies on Internet. Compared with the traditional C/S and CDN architectures, P2P technology has the advantages of high efficiency, good scalability and easy deployment. BitTorrent (BT) has been the most popular P2P system, and both P2P-based file sharing and real-time streaming service have become the main application types that consume one of the largest fractions of Internet traffic.However, since peers have very limited and heterogeneous upload bandwidth, different users or applications have different requirements, such as bandwidth, delay, transmission duration, there are still some difficulties and challenges to provide high quality services. Based on above reasons, we focus on how to analyze and optimize the efficiency of point-to-multipoint content distribution in P2P network. We study the key technologies involved in BitTorrent system and the applications of file sharing and live streaming, e.g., overlay construction, bandwidth allocation, data scheduling. More specifically, the contribution of this thesis lies in the following three aspects.First, we deduce the piece diffusion and the upload utilization models based on the upload bandwidth of peers, and propose improved strategies for overlay construction and incentive mechanism. The derived models can reveal the delicate relationship between some key parameters, such as TFT round, maximum concurrent uploading number, piece size; the proposed overlay construction strategy UC-track algorithm can optimize the topological features of the BitTorrent overlay and improve the efficiency of data transmission in BitTorrent system; the proposed incentive mechanism PU algorithm, which consists of bandwidth limitation strategy and bandwidth allocation strategy, can significantly improve the fairness of BitTorrent system whereas the load on the seed is reduced considerably.Furthermore, we generalize the BitTorrent system into a common P2P system. As to file sharing, we focus on the system performance metric of last finish time. In an upload-constrained P2P file sharing system, we study how to allocate the spare upload bandwidth and how to schedule data resource of assistant peers to improve the performance of P2P file distribution systems. We model both the equal service process and the differentiated service process when the initial data distribution of peers satisfies some special conditions, and also show how to minimize the time to get the file to any number of peers. The proposed fluid-based models can reveal the intrinsic relations among the initial data amount, the size of peer set and the minimum last finish time. The closed-form expressions derived from the extended models can closely approximate the chunk-based models and systems as well, especially for relatively large files. As an application of the extended models, we propose a recursive differentiated service strategy to provide differentiated service to multiple peer sets efficiently.Finally, with regard to the live streaming which has more advanced demands of real time, we concentrate on the system performance metrics of playout delay and playout probability. In a mesh-based overlay where peers have homogeneous upload bandwidth, we establish the analytic framework for the pull-based streaming schemes in P2P network, give accurate models of the chunk selection and peer selection strategies, and organize them into three categories, i.e., the chunk first scheme, the peer first scheme and the epidemic scheme. Moreover, we propose the LR2strategy, which incorporates a segmented chunk selection strategy and a random push strategy, then live streaming and file sharing can be optimized in a unified modeling framework.The results of this thesis have important theoretical and practical significance. On one hand, they could provide fundamental insights into bandwidth allocation and data scheduling in P2P content distribution; on the other hand, they could give helpful reference for improving system performance of P2P systems.更多还原