CAREER: An Infrastructure for Multi-Technology Network Fault Management

计算机通信网络正在从提供“最大努力”服务的便利通信机制演变为我们的国家基础设施的关键要素，我们依赖于商业和个人使用。因此，对改进的故障管理技术的需求正在增加，并且网络故障管理的公开问题的复杂性继续增长。 网络的异构性是造成这一问题日益复杂的一个主要原因，当今的网络由多种技术（如无线、ATM、以太网）组成，这些技术需要支持多种网络应用（WWW、电子商务、视频会议）。 故障管理问题已经在不同程度上针对不同的技术进行了研究，但仍有许多悬而未决的问题，并且研究人员还没有意识到任何可以有效地跨技术应用的方法。目前的方法在很大程度上是临时的，依赖于人类网络管理员的专业知识。即使在单一技术的网络（如以太网局域网），为一个网络开发的策略不能很容易地推广到其他网络。此外，网络内的变化率使得网络管理员难以保持足够高的专业水平来开发和维护有效的故障管理策略。需要能够在各种技术中推广和应用的新技术，以有效的方式确保网络的可靠性。 本研究的长期目标是通过尽可能自动化网络管理来实现自我管理网络。在网络管理中，该项目将首先关注故障管理。故障管理包含了大量复杂的问题，这些问题尚未得到很好的研究或定义。在研究自动化方法之前，必须更好地理解故障管理问题。该研究提出了一系列研究，旨在发现问题的各个方面，这些方面在许多不同类型的网络中是共同的。 这些研究主要是实验性的。本研究计划使用网路故障插入法来研究各种故障对网路的影响。故障将被插入到模拟网络，网络测试床，和校园网。这些实验将提供对故障传播的深入了解，并允许研究人员研究故障管理问题的各个方面如何从故障到故障、网络动态以及不同网络类型的变化。从两个运营网络（CS部门网络和摩托罗拉无线网络）收集的数据将进行分析，并用于验证模拟的准确性。将建立一套用于测试和比较故障管理结果的基准，并向公众提供。 在短期内，研究结果可以用于开发故障管理方法，这些方法（1）可以应用于不同的技术，（2）可以从网络推广到网络，（3）可以适应网络的变化。通过实验发现的共同特征可以用来制定理论基础。这个基础加上基准的可用性，使研究人员能够系统地分解和展示广泛适用的故障管理问题的进展。从长远来看，这些结果可以帮助我们了解如何设计可编程逻辑器件的组件。

Computer communication networks are evolving from convenient communication mechanismsoffering "best-effort" service into key elements of our national infrastructure that we depend on for both business and personal use. As such, the need for improved fault management techniques is increasing and the open problem of network fault management continues to grow in complexity. A major cause of the increase in complexity of this problem is the heterogeneity of networks.Today's networks are comprised of a variety of technologies (e.g. Wireless, ATM, Ethernet), thatare required to support a diverse set of networked applications (WWW, electronic commerce, video conferencing). The fault management problem has been studied in varying degrees for different technologies, butthere are many open issues and the research is unaware of any methods that can effectively beapplied across technologies. Current methods are largely ad hoc and rely on the expertise of ahuman network manager. Even within single-technology networks (e.g. Ethernet LANs); strategies developed for one network cannot easily be generalized to other networks. In addition, the rate of change within a network makes it difficult for the network manager to maintain a high enough level of expertise to develop and maintain effective fault management strategies. New techniques that can be generalized and applied across technologies are needed to ensure network reliability in an efficient manner. The long-term goal of this research is to move toward self-managing networks by automating asmuch of network management as possible. Within network management, the project will first focus onfault management. Fault management encompasses a large, complex set of problems that have not been well studied or defined. Before methods for automation can be studied, the fault management problem must be better understood. The research proposes a set of studies aimed at discovering aspects of the problem that are common across many different types of networks. The proposed studies are primarily experimental. The research plans to use network fault insertion to study the impact of various faults on a network. Faults will be inserted into simulated networks, anetwork testbed, and a campus network. The experiments will provide insight into fault propagation and allow the researcher to study how various aspects of the fault management problem change from fault to fault, with network dynamics, and with different network types. Data collected from two operational networks (CS Department network and Motorola wireless networks) will be analyzed and used to validate simulation accuracy. A set of benchmarks for testing and comparing fault management results will be created and made publicly available. In the short-term, results can be used to develop fault management methods that (1) can be applied across different technologies, (2) can generalize from network to network, and (3) can adapt to changes in the network. Common traits discovered through experimentation can be used to formulate a theoretical foundation. This foundation coupled with the availability of benchmarks allows researchers to systematically decompose and demonstrate progress on widely applicable fault management problems. In the long-term, the results can help us understand how to design components for manageability.