Career: Toward a QoS Provision Architecture in Noncooperative Networks: Theory and Implementation.

随着承载从电子邮件到海量数据再到语音和视频等大量信息的高速局域网和广域网部署的增加，为不同的应用基础提供足够的服务质量（QoS）已成为一个重要问题。目前的技术——有些源于内在的限制——很难在不诉诸过度供应的情况下提供差异化的QoS和保证，而过度供应会使这些服务变得令人望而却步。考虑到人们对互联网作为日常、休闲、经济、教育、娱乐和其他社会功能不可或缺的组成部分的预期使用，需要一种新的架构来将构成当今最努力的带宽的同质集群转化为具有分级QoS属性的分层服务，从而更好地满足应用程序、多媒体或其他方面的不同QoS需求。这次冒险成功的关键是，用户和网络都不需要进行复杂的计算，因为复杂的计算会使解决方案完全不可行的，或者使其与过度配置一样昂贵。本提案基于微观经济学和非合作博弈论的思想，描述了一种适合于最佳努力环境的QoS提供体系结构。我们构建了一个非合作的QoS提供模型，其中假设用户是自私的，并且数据包通过交换机路由，其中作为其指定优先级的功能，差异化服务被交付。服务质量是一种诱导现象，是由底层网络基质介导的自私应用程序之间相互作用的结果。目前的方法与以前的工作有三个重要的不同之处——第一，它是专门为模拟QoS提供而制定的最全面的非合作资源分配模型之一，第二，该方法描述了适合于多交换机广域网环境的特定网络架构，第三，它解决了理论和实践方面的问题，强调了可实现性。PI不仅从博弈论的角度研究了非合作QoS提供模型，显示了纳什均衡与Pareto和系统最优之间存在丰富的结构，而且还研究了当模型扩展到包含可能存在冲突需求的多维QoS向量时，促进足够服务的问题，系统方差对QoS呈现的影响，利用可编程通用处理器共享交换机对网络资源进行自适应重新定位，以及许多交换机系统中出现的无数分布式控制问题。本提案描述了PI以前的研究——包括单开关和多开关——并概述了未来研究的计划，强调对理论和实践问题的综合方法，并对可行的实施有强烈的看法。本提案的双重目标是描述在普渡大学计算机科学系开发网络课程和教育的策略。主要目标有三个。首先，扩展和重组现有的网络课程，赋予学生更多的结构、深度和指导。这是通过加强研究生课程的两个新的网络课程，网络高级主题（已经实施）和实验网络来实现的。在本科课程中，PI最近引入了一门新的网络课程，还有几门课程正在规划阶段。第二，通过将最先进的网络技术纳入课程（PI已经为ATM交换机采购了设备资金，将被纳入1998年秋季的网络课程），并设计一门新的实验网络课程，让学生积极参与网络游戏，研究非合作和合作协议及其行为后果，从而增强研究生研究项目的实力。最后，随着网络技术和继续教育需求的增长，远程学习迅速成为现实，PI希望开发一门可以在网络上有效教授的本科水平的网络课程。作为普渡大学CEE项目的一部分，PI的研究生网络课程将于1998年秋季学期向校外站点进行电视转播，这些站点将接收实时视频/音频馈送和语音通道，以供提问。

With the increased deployment of high-speed local- and wide-area networks carrying a multitude of information from email to bulk data to voice and video, provisioning adequate quality of service (QoS) to the diverse application base has become an important problem. Current techniques---some stemming from intrinsic limitations---make it difficult to provide differentiated QoS and guarantees without resorting to over-provisioning which renders these services prohibitive.Given the expected use of the Internet as an everyday, casual, yet evermore indispensable component of economic, educational, entertainment, and other societal functions by the population at large, a new architecture is needed to transform the homogeneous clump that constitutes today's best-effort bandwidth into stratified services with graded QoS properties that better matches the diverse QoS needs of a compendium of applications, multimedia or otherwise. Crucial to the success of this venture is that neither the user nor the network be required to engage in complex computations which would thence make the solution outright infeasible or render it equally as expensive as over-provisioning.This proposal describes a QoS provision architecture suited for best-effort environments based on ideas from microeconomics and noncooperative game theory. We construct a noncooperative QoS provision model where users are assumed to be selfish, and packets are routed over switches where---as a function of their enscribed priority---differentiated service is delivered. Quality of service is an induced phenomenon, achieved as a result of interaction between selfish applications mediated by the underlying networksubstrate.The present approach differs from previous works in three significant ways---one, it is one of the most comprehensive noncooperative resource allocation models specifically formulated to model QoS provision, two, the approach describes a specific network architecture suited to many-switch wide area network environments, and three, it addresses both theoretical and practical aspects, with an emphasis on realizability. Not only does the PI study the noncooperative QoS provision modelfrom a game-theoreticperspective showing the existence of a rich structure relating Nash equilibria to Pareto and system optima, he also investigate the problem of facilitating adequate services when the model is extended to incorporatemulti-dimensional QoS vectors with possibly conflicting requirements, the effect of system variance on QoS rendered, the adaptive relocation of network resources to hot spots using programmable general processor sharing switches, and the myriad of distributed control problems arising in many-switch systems.This proposal describes the PI's previous research---both single-switch and many-switch---and outlines a program for future research that emphasizes an integrated approach to theoretical and practical issues with a strong view toward feasible implementations.The dual goal of this proposal is to describe a strategy for developing the networking curriculum and education in the Computer Sciences Department at Purdue University. The main goals are threefold. First, expand and reorganize the existing networking curriculum to impart more structure, depth, and guidance to the student. This is achieved by enhancing the graduate program with two new networking courses, Advanced Topics in Networking (already implemented) and Experimental Networks. In the undergraduate curriculum, a new networking course was recently introduced by the PI with several more in the planning stage. Second, enhance the strength of the graduate research program by incorporating state-of-the-art networking technology into the curriculum (the PI has procured equipment funding for ATM switches which will be incorporated into the fall 98 networking curriculum) and devising a new course in Experimental Networksinvolving active student participation as players in network games where noncooperative and cooperative protocols and their behavioral consequences are studied. Finally, with distance learning fast becoming a reality spurned by the growth in enabling networking technology and continuing education demand, the PI hopes to develop an undergraduate-level networking course that can be effectively taught over the Web. This will be facilitated by the PI's graduate networking course being televised in the fall 1998 semester as part of Purdue's CEE program to off-campus sites who will receive live video/audio feed and a voice channel for questioning.