CAREER: Adaptive Network Processing

该方案的研究目标是设计、实现和评估一个可扩展的、自适应的网络服务平台。通过互联网支持复杂的新应用，如多方会议和流媒体传输，给底层基础设施带来了严重的技术挑战。然而，在网络中添加一些处理可以实现新的基于网络的服务，从而显著提高这些应用程序的端到端性能。处理成本的持续指数下降使网络服务变得可行，而应用程序复杂性的增加使其成为必要。然而，链路容量的增长速度甚至快于处理能力的增长，并且当前的网络服务平台不能以线速维持对多个流的重要服务。在这项提案中，研究人员概述了其目标是开发和评估自适应处理技术以扩展网络服务平台支持的流的数量的研究。当总需求超过可用资源时，自适应技术可以控制服务质量的下降，并且可以在单个节点内或跨多个协作节点应用。这项研究的重点是开发适用于广泛节点架构和应用的通用适配技术。拟议的工作分为三个主要方面：设计适应机制本身以及在线适应处理所需的基本控制和资源管理机制和接口；开发触发适应的单节点和多节点政策框架；以及开发能够对可能的适应进行在线评估的适应网络处理的分析模型。研究人员将使用分析、模拟和实验相结合的方法来评估这些技术，这些服务和策略是在奥德赛网络服务平台上实施的许多示例服务和策略。互联网的成功来自于它所支持的各种应用。随着新一代互联网应用的出现，一个关键的挑战是设计一个可扩展的、灵活的网络服务平台，以支持对多个端应用的高速处理；否则，端到端的性能将受到网络施加的处理约束的限制。本研究通过使用自适应技术明确地解决了网络处理可伸缩性问题。通过减少甚至在某些情况下消除网络处理瓶颈，这项工作将使新的应用程序类别能够在互联网上得到大规模支持。教育目标是设计一套核心网络课程，向学生介绍该领域的最新发展，同时强调基本概念，并培训学生理解如何通过组成相对较小、简单和模块化的组件来构建大型网络、复杂程序和现实系统。核心网络课程包括本科生课程、研究生课程和研究课程。本科生课程讲授网络基础知识，重点是基础知识。学生通过构建不同网络层的重要部分以及通过试验他们已经熟悉的真实应用程序来学习。研究生课程的目的是详细了解核心网络协议，了解当前Internet协议背后的原理，并向学生介绍网络领域的新研究思想。研究课程的目标是试图弥合理论和实施之间的差距：了解实施复杂协议所产生的管理费用，并设计机制以减轻其影响。互联网的无与伦比的成功使得每个CS学生都必须了解网络的基础知识。本课程将满足普通CS学生的这一需求，并将激励对网络感兴趣的学生在该领域进行进一步研究。

The research objective of this proposal is to design, implement, and evaluate a scalable and adaptive network service platform. Supporting sophisticated new applications, such as multi-party conferencing and streaming media transmission, over the Internet poses serious technical challenges for the underlying infrastructure. However, adding some processing inside the network enables new network-based services that substantially improves the end-to-end performance of these applications. The continual exponential decrease in the cost of processing makes network services viable, while the increases in application complexity makes it necessary. Link capacities, however, have been increasing at rates even faster than increases in processing capacity, and current network service platforms are incapable of sustaining non-trivial service to multiple flows at line speeds. In this proposal, the researcher outlines research whose goal is to develop and evaluate adaptive processing techniques to scale the number of flows supported by network service platforms. Adaptive techniques control degradation in service quality when aggregate demand exceeds available resources, and can be applied within a single node or across multiple cooperating nodes. The research focuses on the development of general adaptation techniques that are applicable across a wide range of node architectures and applications. The proposed work is grouped into three major thrusts: design of the adaptation mechanisms themselves and underlying control and resource management mechanisms and interfaces required to adapt processing on-line; development of a single- and multi-node policy framework to trigger adaptation; and the development of an analytic model of adaptive network processing that will allow on-line evaluations of possible adaptations. The researcher will evaluate these techniques using a combination of analysis, simulation, and experimentation with a number of example services and policies implemented over the Odyssey network service platform. The success of the Internet derives from the spectrum of applications it supports. As a new generation of Internet applications emerge, a pivotal challenge is to design a scalable and flexible network service platform that can sustain processing for multiple end-applications at high speeds; otherwise, end-to-end performance will be limited by the processing constraints imposed by the network. This research explicitly addresses the issues of network processing scalability by using adaptive techniques. By reducing, and in some cases eliminating, network processing bottlenecks, the work will enable new classes of applications to be supported on a large scale over the Internet. The educational objective is to design a set of core networking courses that introduce students to current developments in the field while emphasizing the fundamental concepts and to train students to understand how large networks, complex programs, and realistic systems can be built by composing relatively small, simple, and modular components. The core networking curriculum consists of an undergraduate, a graduate, and a research course. The undergraduate course teaches the basics of networking with an emphasis on the fundamentals. Students learn by building substantial parts of different network layers, and by experimenting with real-world applications with which they are already familiar. The aim of the graduate course is to understand core network protocols in detail, to appreciate the philosophy behind the current Internet protocols, and to introduce students to new research ideas in networking. The goal of the research course is to try to bridge the gap between theory and implementation: understand the overheads incurred in implementing complex protocols and devise mechanisms to mitigate their effects. The unparalleled success of the Internet has made it essential that every CS student understand the basics of networking. This curriculum will fill this need for general CS students and will motivate students interested networking to pursue further research in the field.