【作者】 何俊峰；

【导师】 龚正虎；

【作者基本信息】 国防科学技术大学 ， 计算机科学与技术， 2012， 博士

【摘要】 大规模流媒体应用是Internet中极为重要一类应用，但由于当前的互联网的网络层只提供“尽力转发”的报文服务，而应用层的解决方案又存在与底层网络信息不相匹配的问题，导致大规模流媒体系统设计仍然是一项充满挑战性的研究。与此同时，当前互联网技术的发展呈现出一些新的动向，如数据传输和内容协商相分离、性能日益增强的用户边缘网络、统一网内存储的网络层功能、以内容为中心的网络设计等。这些新动向，为新一代大规模流媒体网络技术研究提供了许多机遇和挑战。通过对现有解决方案的分析，以及对互联网现有网络基础设施演变过程的思考，我们认为大规模的流媒体内容分发技术研究，应该从参与内容分发的边缘网络和网络利益群体需求分析着手，重点研究对边缘网络资源的有效合理利用。因此，本文的研究创新和贡献包括以下几个方面：（1）充分利用边缘网络用户流媒体需求的聚集效应和边缘网络潜在的存储资源，提出了边缘到边缘的流媒体服务模型（E2EMSM）边缘网络用户通常具有相似的网络利益和流媒体服务需求，可形成流媒体分发的聚集点。因此，本文的基本思想是将整个互联网络空间划分为核心传输网络空间和边缘服务网络空间，实现“核心高效传输，边缘多样服务”的设计理念。通过在边缘网络的流媒体分发聚集点部署相应的网络基础设施，如具备内容存储功能的新型网络交换设备（NCR），我们提出了边缘到边缘的流媒体服务模型（E2EMSM）。基于该模型设计的流媒体内容分发方案，可以充分利用边缘网络流媒体分发的聚集效应，优化流媒体分发过程，减少边缘网络流媒体分发的流量需求，提高边缘网络流媒体分发质量。（2）构建边缘网络存储资源模型（ESRM），并在此基础上提出边缘网络流媒体内容分发模型（Put/Get模型）和启发式流媒体内容部署算法（HEA算法）边缘网络中存在一系列的存储资源，如服务器上的存储资源，用户终端上的存储资源，新型网络交换设备（NCR）上的缓存资源。对这些潜在的可用存储资源进行描述，我们构建了边缘网络存储资源模型（ESRM）。在此基础上，就如何使用NCR上的存储资源，我们提出面向流媒体应用的边缘网络内容分发模型，即Put/Get模型。该模型通过将网络交换设备中的缓存资源作为数据分发的枢纽，能够将数据分发路径与底层网络拓扑结构进行较好地匹配，从而实现网络资源的最大化利用，提高流媒体分发的效率。随后，我们提出了一种面向Put/Get模型的启发式流媒体内容部署算法（HEA）。该算法基于对用户请求概率的假设，实现边缘网络范围内的流媒体内容部署优化，能够有效降低服务器工作负载，减少边缘网络流媒体分发的整体网络代价。（3）为了集中高效使用边缘网络用户终端的空闲存储资源，我们提出了一种流媒体分布存储系统框架（BufferBank）BufferBank系统主要由三个部分组成，即应用客户端（BBA）、缓存控制管理服务器（BBM）、缓存银行系统储户（BBD）。该系统采取类似于现实生活中的银行系统的激励机制，鼓励BBD将他们的空闲存储资源充分利用起来，存入BufferBank系统之中，由BBM将这些存储资源集中进行管理，并把它们借贷给上传的应用，从而在改善这些应用的客户体验的同时，增加网络运营商的收入。BBD用户能够通过将存储资源存入BufferBank的形式获得相应的奖励。该系统能够有效提高总体缓存服务带宽，降低数据访问延时，提高数据读写效率。（4）为了在边缘网络实现以内容为中心的网络设计思想，利用NetMagic硬件平台设计了具有双转发引擎的路由器NCR原型系统NCR基本原理是在支持常规的IP路由的基础上，增加内容转发处理，使得边缘网络能够更好地适应以内容的访问和获取为中心的应用模式。在常规NDN设计的基础上，增加NCR上的用户共享内容索引，即通过构造末端域索引表EDIT来减轻路由器上的内容存储和检索压力，从而提高NCR的报文处理速度。基于NetMagic网络实验平台详细的讨论了NCR转发引擎的设计与实现。更多还原

【Abstract】 Large-scale streaming media distribution has gradually become the most importantand potential Internet application. Since the “best-effort”packet forwarding feature ofcurrent Internet’s network layer and mismatching between application layer solutionsand underlying network information, the design of large-scale streaming videoapplication over Internet is still a challenging study. Meanwhile, the evolution ofInternet shows some arresting trends that bring new opportunities. For example, theseparation between data transmission and content negotiation, the increasing user edgenetwork, uniform In-Network Storage capability and content-centric network design.After deep analysis of current mainstream solutions and the evolution of networkinfrastructure, this dissertation points out that in order to seek a sound solution of thelarge-scale streaming media distribution problem, researchers shoud study requirementsof different network stakeholders involved in the content distribution process firstly, andthen put their focus on effective utilization of all possible network resources, especiallythe networkwide storage resources. Therefore, contributions of this dissertation can beconcluded as follows:(1) Given the aggregation effect of edge network users’ media streamingrequirements and the potential storage resources within edge network, we propose anEdge to Edge Media streaming Service Model, named E2EMSM. Generally, it is tosome extent similar that network interest and streaming media service requirements ofusers from the same edge network. Considering this fact, we can achieve the designapproach,“high-efficiency transmission in the core, and diversified services in theedge”, by dividing the whole network space of Internet into two parts, the coretransmission network space and the edge service network space. To deploy somerelative network infrastructure, such as the new content storage enabled networkswitching equipment which is named NCR, at the aggregate point of edge network, wepropose the Edge-to-Edge Media streaming Service Model, named E2EMSM. Thesolution which is based on this model, can take full advantage of the aggregation effectto optimize distribution process, reduce flow requirements, and improve service quality.(2) An edge network storage resource model, named ERSM, is constructed. On thisbasis, a novel streaming media content distribution model, named Put/Get, is presented.And then, a heuristic streaming media content deployment algorithm, named HEA, ispresented. There are kinds of storage resources within edge network area, such as theserver’s storage resources, storage resources on the user terminal, and the cacheresources on the new network switching equipment (NCR). To describe these potentialstorage resources, we construct the edge network storage resource model, named ERSM.On the basis of cache resource on NCRs, we propose a new edge network streaming media distribution model, the Put/Get model, which uses the cache resource on NCRs asthe aggregation point of data forwarding. Under this way, the reality data distributionpath can match well with the underlying network topology, to achieve the maximumutilization of network resources, and improve the efficiency of streaming mediadistribution. And then, we propose a heuristic streaming media content deploymentalgorithm, named HEA, which is based on the assumption that the probability of userrequest. It can reduce server workloads and the overall network cost of streaming mediadistribution within edge network.(3) An edge network streaming media distributed storage system, namedBufferBank, is presented, which is to concentrate the use of free storage resources onusers’treminal. It is mainly composed of three parts, the application client (BBA),management server (BBM), and the depositor (BBD). It takes incentive similar to reallife banking system, where BBDs are encouraged to deposite their free storageresources into BufferBank and these resources are concentrated to manage and belended to upper applications by BBM.(4) Based on the NetMagic hardware platform, a streaming media router with dualforwarding engine, named NCR, is presented. The basic idea of NCR is to increase thecontent-centric routing mechanism on conventional IP router, which will make edgenetwork service fit the dominant network usage. By manage a user shared content indexon NCR, we can mitigate the stress of read/write operation on content storeimplemented on NDN Router, and then improve the forwarding speed. The forwardingengine of NCR prototype implemented on NetMagic hardware platform has beendiscussed.更多还原