【作者】 张海鹏；

【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， 2014， 博士

【摘要】 本世纪初,随着对等网络(peer-to-peer, P2P)技术在因特网上广泛应用,其良好的扩展性以及较高的资源利用率等优点影响着网络技术的发展,网络发展的趋势也从集中式系统逐步向分布式系统过渡。如今,文件共享、即时通信、社交网络以及在线视频点播等网络服务系统遍布网络,影响着人们的生活、工作以及学习等各个方面。以BitTorrent为代表的P2P文件共享系统通过充分利用其中各用户节点的带宽资源、存储空间来实现文件的共享,而以在线视频点播(如优酷、土豆以及爱奇艺等)为代表的流媒体服务系统通过采用内容分发网络(Content Distribution Network, CDN)技术将服务部署到距离用户仅有“一跳”的网络边缘,并通过采用P2P技术加强边缘服务器之间的交互,从而可以提高服务质量、增大系统吞吐量和降低服务提供商部署成本。本文将着重研究网络服务系统中两种典型系统：BitTorrent和流媒体服务系统。通过分析研究这两种系统的结构、主要流程以及主要影响因素,并将各因素进行数学描述,建立其动力学模型,描述系统的动态演化过程,并进行具体的模型分析以揭示影响系统演化的各因素之间的相互关系以及对系统整体性能的影响,为网络服务系统的研究、设计以及实现提供理论支持。本文首先分析了BitTorrent和流媒体服务系统的结构、主要流程以及影响它们动态演化的主要因素,然后分别对这两种系统进行动力学建模,并在动力学模型的基础上,对流媒体服务系统的稳定性进行讨论。本文的主要工作与创新性表现在以下几个方面：第一,建立了BitTorrent文件共享系统的动力学模型。BitTorrent文件共享系统是P2P系统的典型应用,用户在享受下载的同时还向其他用户提供上传,充分体现P2P系统的对等性和交互性。以往针对BitTorrent的建模研究着重于全系统范围内的节点数目的变化,将影响系统变化(尤其是节点数目的变化)的主要因素归结于带宽资源的限制,忽略了其他因素对于BitTorrent系统的影响,本文采用在较小尺度上针对BitTorrent系统中单个节点的动态演化过程进行建模分析,将影响BitTorrent用户之间交互演化过程的硬件条件、文件属性、用户行为以及算法策略等因素进行数学描述,建立其动力学微分方程模型。该模型能够准确描述各因素对于BitTorrent系统中节点间交互过程的影响,同时也反映了各因素之间的相互影响关系。第二,建立了流媒体服务系统的动力学模型。传统的针对流媒体服务系统的研究侧重于单纯算法策略的研究,而忽略了系统自身的演化过程以及这些算法策略之间的相互影响。本文首先对流媒体服务系统结构进行抽象化描述,用虚拟服务节点代表流媒体服务系统覆盖网络中的边缘服务子系统,进而将流媒体服务系统抽象成一个由虚拟服务节点组成的逻辑网络；然后分析影响虚拟服务节点与用户以及与其他虚拟服务节点交互过程中的影响因素,主要包括硬件条件、用户行为、服务属性以及算法策略等因素,并将这些影响因素进行数学描述,建立流媒体服务系统的动力学微分方程模型。该模型能够准确刻画用户接入到服务系统,以及服务子系统之间的交互过程,并能够反映各因素之间的相互关系以及他们对于系统整体性能的影响。第三,提出了流媒体服务系统中边缘服务子系统的状态稳定性判据。边缘服务子系统距离用户仅有“一跳”之遥,是直接面向用户的服务提供者,其稳定性将直接影响用户的体验。本文从边缘服务子系统的整体性以及其运行实际意义出发,定义了边缘服务子系统的稳定性状态,并在边缘服务子系统的动力学模型的基础上,分析得到了边缘服务子系统状态的稳定性判据。第四,提出了一种基于边缘服务子系统状态稳定性判据的接入控制算法。接入控制是边缘服务子系统的重要组成成分,控制着用户行为对于系统的影响。本文在上述稳定性分析的基础上,对流媒体服务系统提出了一种接入控制算法。当服务请求到达时,首先判断边缘服务子系统当前状态的稳定性,若当前状态为稳定状态,则接入请求；否则拒绝接入直到当前状态变为稳定状态。更多还原

【Abstract】 The P2P (peer-to peer) technology has been influencing the development of the Internet with its good extensibility and effective utilization of resources since its emer-gence in the beginning of the21st century. The development of the Internet began to shift focus from the centralized system to the distributed system. The network service systems, such as file-sharing, instant messaging, social network and streaming service system, have changed the way people live, work, and study. One example is the P2P file-sharing system, with the BitTorrent software as a typical one, which makes full use of users’resources including bandwidth and storage space to share files. Another one is the streaming service system represented by the video-on-demand systems such as Youku, Tudou and iQiYi. The system distributes services to the edge of the Internet by adopting the CDN (content delivery network) technology. Furthermore, by strengthen-ing the interaction between proxy servers using the P2P technology, the stream service system can improve the quality of services (QoS), expand the throughput of the system and reduce the running cost of the network service providers. This dissertation focuses on the two typical network service systems, which are BitTorrent and network stream-ing service system. Structures, main working process and influencing factors of the two systems are analyzed in the dissertation. The influencing factors are mathematically ex-pressed by a series of differential equations. By setting up a dynamic model comprising of the differential equations, the dissertation analyses the evolution of the systems and influence of the factors on the systems as well on each other. The model can work as theoretical support for the study, design, and application of the network service systems.In the dissertation, the structure, main processes and the influence factors of the BitTorrent and streaming service system are analyzed, then the dynamic models are set up to describe the evolution of the BitTorrent and streaming service system. Further-more, we discuss the stability of streaming service system based on the dynamic model. The contribution of this dissertation are listed in the following part.Firstly, a novel dynamic model of BitTorrent P2P file-sharing system is proposed in this thesis. Users of the BitTorrent system, which is a typical application of P2P system, not only download the file but also upload the file to others. This reflects the character of parity, interaction, good extensibility of P2P. The past research put their attention on the evolution of the downloaders’and seeds’number in the BitTorrent sys-tem, and assumed that the main factor influence the evolution is the users’bandwidth, while other influencing factors have been ignored. Instead of studying the macroscopic characteristics of the whole system, the author models one single node in the BitTorrent. The factors that influence the interaction between nodes in the BitTorrent are expressed in mathematical forms, and a dynamic model is set up to describe the interaction be-tween the users of the BitTorrent system in an accurate manner, and also to analyze the influence of the factors on one another.Secondly, a dynamic model of the streaming service system is proposed. The tra-ditional research for the streaming service systems placed emphasis on the algorithms and ignored the interaction between the algorithms. The streaming service system can be seen as a logical network, which is composed of virtual service nodes that repre-sent the streaming service subsystem. Then the factors, such as hardware, user be-havior, algorithms, etc. are analyzed and expressed in mathematical forms. Then the dynamic model of the streaming service system is set up to describe the interaction between clients and the virtual service nodes and also between virtual service nodes. Inter-relations of the influencing factors can also be revealed by the analysis.Thirdly, a criterion for the stability of the streaming service system is proposed in this thesis. The service subsystem is at the edge the Internet, which serve the users directly. The stability of the subsystem influence the user experience directly. We con-sider the subsystem as a whole and analyze the actual conditions, and then define the stability of the streaming service system. Furthermore, based on the model already set up, the criterion for stability of the streaming service system can be gained through analysis.Lastly, an admission control policy based on the stability criterion of the streaming service system is proposed. The admission control policy is a dominant factor in decid-ing whether the users’requests can be admitted by the system. Based on the stability criterion, the author designs an admission control policy under which users’requests will be admitted when the system is stable and rejected when the system is unstable.更多还原