NECO: Architectural Support For Fault Management

今天的互联网是一个由所有人使用的全球基础设施。它也是许多企业的生命线。然而，互联网仍然缺乏我们从关键基础设施中期望的那种可靠性和健壮性。虽然网络具有内置的容错能力，但它通常仅限于硬故障；网络既不能检测软故障，也不能对软故障做出反应，例如可能由于实施错误或硬件故障或与拥塞相关的高延迟或丢失。要支持对延迟和丢失敏感的应用程序，必须设计有效的机制来快速检测此类故障并以自动方式从故障中恢复。网络设备对调试性能问题提供的支持非常少，更不用说自动响应它们了。运营商今天在提供商边缘路由器对之间注入主动探测器，以检测其网络中的任何转发问题，如高延迟和丢失。运营商通常依赖于应用间接推理算法，例如断层扫描方法，这些算法可以通过将活动探测器与网络拓扑快照相结合来推断根本原因。由于问题从根本上是欠约束的，推理充其量只能是近似的，可能相当慢。自动响应机制需要快速准确的定位才能有效。本研究的主要研究重点是创建新的网络内故障管理机制，以自动检测、定位、报告和响应故障和其他性能降级。研究围绕三个基本思想：1)为路由器配备专门的高速低复杂度测量原语来测量延迟和丢失；2)利用这种反馈使路由协议能够自动响应拥塞或慢性故障条件；3)将这些机制集成到NetFlow中，以提供每流的延迟和丢失。这些机制共同为网络运营商和管理员提供了一套丰富的工具，以有效地监控和管理他们的网络，并提高这些网络的整体容错性能。智力上的优点：这项研究将通过允许网络自动检测和响应软故障，为已建立的网络容错研究领域做出贡献；它将开辟用于高保真测量的可扩展路由器原语的新研究领域；它将通过纳入与互联网最新创新和进步相称的流的新定义，显著增强路由器的能力。这一研究方向结合了三个不同的研究领域--可扩展的路由器原语、容错路由协议和测量--以显著提高关键基础设施Internet的健壮性。这项工作有可能提供互联网所需的强大故障管理平台，以支持对延迟敏感的下一代交互应用程序。广泛影响：该项目将有助于培养下一代网络工程师和设计师。研究人员将通过传统的学术渠道传播研究成果，并将积极参与行业论坛，利用他们与美国电话电报公司和思科的合作。他们还将设计和修改研究生和本科生课程中的网络和路由器架构课程，并通过荣誉计划监督博士、硕士和本科生项目。

The Internet today is a global infrastructure used by one and all. It is also the life-line of many businesses. The Internet, however, still lacks the kind of reliability and robustness we expect from critical infrastructure. While the network has in-built fault-tolerance, it is typically restricted to hard failures alone; the network is equipped to neither detect nor react to soft failures such as high delays or losses that may be due to faulty implementations or hardware failures or congestion related. To support delay- and loss-sensitive applications, it is important to devise efficient mechanisms to quickly detect such failures and recover from them in an automated fashion. Network devices provide very little support for debugging performance problems?let alone automatically respond to them. Operators today inject active probes between pairs of provider edge routers to detect any forwarding problems such as high delays and losses, in their network. Operators typically rely on applying indirect inference algorithms, such as tomographic approaches, that can infer root causes by joining active probes with network topology snapshots. Given the problem is fundamentally under-constrained; inference is only approximate at best and can be quite slow. Automated response mechanisms require fast and accurate localization in order to be effective. The main research focus of this research is to create novel in-network fault management mechanisms to automatically detect, localize, report and respond to failures and other performance degradations.The research revolves around three basic ideas: 1) Equipping routers with specialized high-speed low-complexity measurement primitives to measure delay and loss; 2) Using this feedback to allow routing protocols to automatically respond to congestion or chronic failure conditions; and 3) Integrating these mechanisms into NetFlow to provide per-flow delay and loss. Together, these mechanisms provide a rich set of tools for network operators and administrators to monitor and manage their networks efficiently and improve the overall fault-tolerance properties of these networks.Intellectual Merit: This research will contribute to the established research area of network fault-tolerance by allowing the network to automatically detect and respond to soft failures; it will carve new areas of research on scalable router primitives for high-fidelity measurements; and; it will significantly enhance the capabilities of routers by incorporating new definitions of flows commensurate with latest innovations and advancements in the Internet. This research directions combine three disparate areas of research?scalable router primitives, fault-tolerant routing protocols, and measurement?to significantly improve the robustness of critical infrastructure, the Internet. The work has the potential to provide a powerful fault management platform the Internet requires to sustain the next generation of delay-sensitive and interactive applications.Broad Impact: The project will contribute to the education of the next generation of networking engineers and designers. The researchers will disseminate the results through the traditional academic channels, and will actively participate in industry forums leveraging their collaborations with AT&T and Cisco. They will also design and revise courses in Networking and Router Architectures in the graduate and undergraduate curriculum, and supervise Ph.D, Master, and undergraduate projects through the Honors program.