Evaluating Global-scale Distributed Systems using Scalable Network Emulation

在过去十年中，国家信息基础设施的力量得到了极大的扩展。商业互联网服务和政府资助的用于研究和教育的先进网络支持了一系列国家优先事项。不幸的是，健壮的自适应网络软件的快速开发和部署受到一个基本障碍的阻碍：由于其主机环境（互联网）的规模和复杂性，原型系统难以评估。本提案旨在通过构建一个软件环境来评估原型网络软件系统，通过可扩展的互联网在不现实的、可控的、可重复的条件下进行评估，从而消除这一障碍。所提出的系统ModelNet在高速集群上模拟广域网，使研究人员能够在可配置的类似互联网的环境中部署修改后的软件原型，并使其适应故障和不同的网络条件。本研究将首先研究将网络仿真扩展到数千个总通信带宽超过10gb /s的未修改应用程序的技术。其次，它将使社区能够利用大规模网络仿真作为快速开发和评估下一代互联网服务和应用的主要技术。最后，该研究将通过提供一个受控环境，使网络服务适应一系列实际部署场景，从而寻求网络服务鲁棒性和性能的根本改进。

The power of the national information infrastructure has expandedenormously over the last decade. Commercial Internet service andgovernment-sponsored advanced networks for research and educationsupport a range of national priorities. Unfortunately, rapiddevelopment and deployment of robust, adaptive network software ishampered by a fundamental obstacle: prototype systems are difficult toevaluate due to scale and complexity of their host environment---theInternet. This proposal seeks to remove this obstacle by constructing a softwareenvironment for evaluating prototype network software systems underrealistic, controlled, repeatable conditions through scalable Internetemulation. The proposed system, ModelNet, emulates a wide-areanetwork on a high-speed cluster, enabling researchers to deployunmodified software prototypes in a configurable Internet-likeenvironment and subject them to faults and varying network conditions. This research will first investigate techniques for scaling networkemulation to thousands of unmodified applications with aggregatecommunication bandwidth of over 10 Gb/s. Second, it will enable thecommunity to leverage large-scale network emulation as a primarytechnique for rapidly developing and evaluating next-generationInternet services and applications. Finally, the research will seekfundamental improvements in network service robustness and performanceby providing a controlled environment for subjecting network servicesto a range of realistic deployment scenarios.