ITR: Collaborative Rsearch: Scalable Quality-of-Service Control for the Next Generation Internet

今天的互联网的巨大成功要归功于25年前制定的简单的“沙漏”IP网络协议架构。近年来，随着网络技术的飞速发展和丰富多媒体内容的爆炸式增长，网络社区发现自己处于一个重要的十字路口：下一代互联网架构应该是什么，以控制网络资源并提供新兴多媒体应用所需的服务质量（QoS） ？提供QoS保证的可能方法有多种。下一代互联网的QoS解决方案的选择将对互联网本身的发展及其功能产生重大影响。做出“正确”的选择需要对QoS控制的可伸缩性以及这些控制对QoS供应效率的影响有一个基本的理解。所提议的研究的目标是对提供可扩展QoS保证的各种方法所涉及的基本权衡进行全面、定量的理解。为此，我们将发展连贯的理论来系统地解决QoS控制中的可伸缩性问题。研究计划大致分为四个领域：保证流的聚合网络演算；获得对精细时间尺度(例如；研究人员将开发一个聚合网络演算来研究聚合QoS控制机制对数据平面操作的性能和复杂性的影响。这个理论是为保证流开发的，保证流要求网络在其生命周期内提交一定数量的资源（例如，带宽和缓冲区），无论网络拥塞状态如何。聚合网络演算将提供一个数学框架来量化聚合QoS控制对QoS供应中基本权衡的影响。它还将对可伸缩数据平面QoS控制机制的设计产生见解。响应流的端到端QoS控制：研究人员将开发流体模型来研究聚合QoS控制机制对响应流端到端性能的影响。响应流通过调整其传输速率来响应网络拥塞的迹象，例如丢失。这些模型将使我们能够更好地理解响应流的行为，如asTCP与不同的聚合QoS机制相结合，并为响应流设计端到端QoS服务。QoS控制规律和控制平面聚合规则。我们将开发用于捕获慢时间尺度的QoS控制律，网络的系统范围行为和聚合规则，以解决聚合QoS控制下控制平面操作的性能和复杂性。这些QoS控制定律和聚合规则将引导我们为可扩展的控制平面操作设计分布式和集中式算法。作为发展这些理论的一个组成部分，研究人员还将设计有效的和可扩展的QoS机制，以及量化和评估各种QoS解决方案的权衡的工具和技术。基于这些努力的结果，研究人员将研究如何将各种QoS解决方案组合起来，以构建有意义的端到端服务。该研究将融合正式建模/分析、实验/实现和评估。作为我们研究的结果所获得的理解和见解将导致为未来互联网构建可扩展QoS控制的理论、设计原则和指导方针的建立。随着下一代互联网的形成，这将使人们能够做出理性和明智的选择。

Today's Internet owes its great success to the simple, "hour-glass" IP network protocol architecture laid out twenty-five years ago. With rapid advances in networking technologies and explosive growth of rich multimedia content in recent years, the networking community finds itself at an important crossroads: what should be the next generation Internet architecture for controlling network resources and provide the quality of service (QoS) needed by emerging multimedia applications? There is a multidimensional spectrum of possible approaches to providing QoS guarantees. The choice of a QoS solution for the next generation Internet will have a substantial impact on both the evolution of the Internet itself, and on what it enables. Making the "right" choices requires the development of a fundamental understanding of the scalability of QoS controls and the impact of these controls on the efficacy of QoS provisioning. The goal of the proposed research is to develop a comprehensive, quantitative understanding of the fundamental trade-offs involved in various approaches toward providing scalable QoS guarantees. To this end, we will develop coherent theories to systematically address the issue of scalability in QoS controls. The research program divides broadly into four areas: Aggregate network calculus for guaranteed flows: To gain a thorough understanding of the fine time-scale(e.g., packet-level) behavior of a network system in providing QoS performance guarantees, the researchers will develop an aggregate network calculus to study the impact of aggregate QoS control mechanisms on the performance and complexity of data plane operations. This theory is developed for guaranteed flows - flows which require the network to commit, either at a per-flow or an aggregate level, a certain amount of resources (e.g., bandwidth and buffer) throughout their life time, regardless of the network congestion status. The aggregate network calculus will provide a mathematical framework to quantify the impact of aggregate QoS controls on the fundamental trade-offs in QoS provisioning. It will also yield insights into the design of scalable data plane QoS control mechanisms. End-to-end QoS controls for responsive flows: The researchers will develop fluid models to study the impact of aggregate QoS control mechanisms on the end-to-end performance of responsive flows. A responsive flow responds to signs of network congestion, such as loss, by adapting its transmission rate. These models will enable us to develop a better understanding of the behavior of responsive flows such asTCP coupled with different aggregate QoS mechanisms and to design end-to-end QoS services for responsive flows. QoS control laws and control plane aggregation rules. We will develop QoS control laws for capturing the slow time-scale, system-wide behavior of a network and aggregation rules that address the perfor-mance and complexity of control plane operations under aggregate QoS controls. These QoS control laws and aggregation rules will lead us to the design of distributed and centralized algorithms for scalable control plane operations. Scalable QoS mechanisms and service architectures As an integral part in developing these theories, the researchers will also design effective and scalable QoS mechanisms, and tools and techniques for quantifying and evaluating the trade-offs of various QoS solutions. Based on the results from these efforts, the researchers will study how various QoS solutions can be combined to construct meaningful end-to-end services. The research will blend formal modeling/analysis, experimentation/implementation, and evaluation. Theunderstanding and insights gained as a result of our research will lead to the establishment of the theory,design principles, and guidelines for building scalable QoS controls for the future Internet. This, in turn,will allow reasoned and informed choices to be made as the next generation Internet takes shape.