【作者】 朱敏；

【导师】 徐恪；

【作者基本信息】 清华大学 ， 计算机科学与技术， 2014， 博士

【摘要】 随着互联网应用环境、类型及组成的多样化发展，构造符合应用变化需求的未来互联体系结构成为学术界和工业界的共识，如何评估提出的方案是否符合应用发展的需求也成为互联网体系结构研究中的一个重要课题。由于互联网体系结构的复杂性，现有互联网体系结构的评估工作相对分散且针对性较强，主要以功能和性能评估为主，缺乏对体系结构自身适应应用变化能力进行评估的框架及模型。本文总结体系结构提供应用服务过程中的扩展性、可部署性和经济性需求，提出了互联网体系结构应用适应性的概念，并围绕其评估方法及应用变化下的互联网体系结构发展问题展开了相关研究，主要的研究内容如下：1.多角度总结互联网体系结构发展及现有评估工作。总结分析了互联网应用变化的趋势，针对面向应用变化的高效数据分发体系结构研究进行了综述；从模型、机制和方法三个方面总结分析了互联网体系结构评估工作的研究现状、发展趋势及面临的问题，明确了研究方向。2.基于应用环境变化导致的挑战及应用发展需求，提出互联网体系结构应用适应性概念及其评估框架。该框架建立在体系结构应用服务扩展能力及经济适应能力基础上，结合效用机制及多学科融合来表示和分析技术及经济因素对体系结构应用适应性的影响，为从多角度建立应用适应性评估模型提供了理论指导。3.提出基于服务能力的互联网体系结构应用适应性评估模型。该模型从体系结构提供应用服务的本质出发，可支持不同应用组成及类型变化下体系结构适应性的评估。利用该模型对多种内容分发机制的应用适应能力进行了分析讨论，实验结果对于运营商部署和完善相关协议具有重要的参考价值。4.提出基于用户效用的三网融合评估模型。该模型通过用户效用将应用组成、初始市场份额、社会经济状况、投资及用户群体关系等影响三网合一的要素结合在一起，可用于分析多种体系结构在合作竞争下的发展趋势。基于该模型及微分方程对三网融合发展趋势的分析结果可为网络运营商的决策提供指导。5.提出一种应用效用最大化的自适应内置缓存系统。该系统从应用需求现状出发，结合现有内容分发机制的优势，综合考虑缓存位置、数据访问热点、数据量、经济开销对应用效用的影响，采用在边缘自治域部署支持组播的内置缓存策略，实现了特定经济开销限制下的高效数据分发。更多还原

【Abstract】 With the increasing complexity of environment, category, and composition of ap-plications on the Internet, it has become a consensus among academia and industry todevelop the next generation of Internet architecture to meet the diverse application needs.However, there is no systematic research evaluating the application adaptation capabilityof proposed architectures in current literature. Most of existing evaluation focused on thefunctionality and performance of the Internet architecture but lack of evaluation frame-works and models that accommodates the change of application. This is partly due tothe complexity of the Internet architecture. We fill the gap in the literature by (a) ana-lyzing the scalability, deployability, and economic demands of the Internet architecturein supporting application services,(b) proposing the concept of the Internet architectureapplication adaptability and its evaluation methods, and (c) studying the development ofthe Internet architecture with the changing of applications. Our main reserach works andconclusions are as follows:1. We provide a comprehensive review of existing research on the developmentof Internet architecture and its evaluation. We analyze the development trends of theInternet applications and their impact on future Internet data distribution architectures.We also review the related works and challenges for the Internet architecture evaluationresearch from the perspectives of models, mechanisms, and methods. Such review helpsus identify the direction for further study.2. We propose the concept of Internet architecture application adaptability and de-sign an evaluation framework according to the challenges and opportunities brought bythe application changes. This framework is based on the service scalability and economicadaptation of the architecture. It reflects the impact of technical and economic factorson Internet architecture application adaptability, measured by the utility and multidisci-plinary. It provides a theoretical basis to build the architecture application adaptabilityevaluation models from multiple perspectives.3. We propose an Internet architecture application adaptability evaluation modelbased on service capacity. This model can evaluate the application adaptability of theInternet architecture under different compositions of and types of applications. We apply this evaluation model to various types of content distribution mechanisms. Our resultsprovide valuable insights to ISPs.4. We propose a tri-network convergence evaluation model based on user utilitytheory. This model takes into consideration of both supply-side factors and demand-side factors including initial network market share, application composition, investment,economic cost and social relationships. It can be used to analyze the development trendsof architectures under the cooperation and competition among ISPs. The findings of thetri-network convergence evolution based on the model and differential equations provideguidance for ISPs’ decision.5. We design a utility-maximizing adaptive in-network caching system. This systemis based on the status quo of application demands and the existing content distributionmechanisms. It employs the multicast and in-network caching in the edge ASes afterconsidering the utility factors such as the cache placement, the data access hotspots, thedata size and the economic cost. This system achieves efficient data distribution undercertain economic cost.更多还原