CAREER: Intent-Based Network Management

基于意图的网络管理本研究正在开发一个框架，用于更好地管理IP网络，使用基于意图的网络管理方法来解决与管理复杂IP网络相关的性能和鲁棒性问题。我们的目标是自动生成所需的网络配置，只需要高层次的目标作为输入，同时也确保网络级的目标，如性能和可靠性。该研究开发了适用于管理当前和未来网络的设计原则，并为设计可管理性网络提供了见解。该框架使用基于抽象的表示，使高层次的规范自动化低层次的任务，使人们可以直接管理网络。在今天的网络管理中的三个主要障碍是解决：不够的可见性，无法预测的结果网络配置的变化，和隐藏的错误，由于汇编语言的接口配置。 本研究的关键技术包括：（1）基于测量的网络协议运行模型推导，以引出未记录的限制行为。（2）量化测量方法局限性的方法，以实现更准确的结果解释。(3)隐私保护、激励兼容的跨网络数据共享。(4)通过开发量化网络属性的指标和封装设备细节的抽象来系统评估权衡。（5）实时决策支持假设分析和自动化基于意向的配置生成。(6)联合控制和数据平面管理。该研究将通过解决实现自动化、可演进、健壮的网络管理方面的关键挑战，推进IP网络管理的最新技术水平。任何开发的软件都将公开提供。研究成果将被整合到本科生和研究生课程中。

Intent-based Network ManagementThis research is developing a framework for better managing IPnetworks using an intent-based network management approach to addressthe performance and robustness issues associated with managing complexIP networks. The goal is to automatically generate needed networkconfigurations by requiring only high-level objectives as input whilealso ensuring network-level objectives such as performance andreliability. The research develops design principles applicable tomanaging both current and future networks and provides insight intodesigning networks for manageability. Using abstraction-basedrepresentations, the framework enables high-level specifications toautomate low-level tasks, so that one can directly manage the network.Three main obstacles in today's network management are addressed:insufficient visibility, inability to predict the outcome of networkconfiguration changes, and hidden errors in configurations due toassembly-language-like interfaces. The research develops these keytechniques: (1) Measurement-based derivation of network protocols'operational models to elicit undocumented limiting behavior. (2)Methods to quantify limitations of measurement methodologies to enablemore accurate result interpretation. (3) Privacy-preserving,incentive-compatible data sharing across networks. (4) Systematicevaluation of trade-offs by developing metrics for quantifying networkproperties and abstractions encapsulating device details. (5)Real-time decision support for what-if analysis and automatedintent-based configuration generation. (6) Joint control and dataplane management.The research will advance the state of the art in managing IP networksby addressing key challenges in achieving automated, evolvable, androbust network management. Any developed software will be publiclyavailable. The research results will be integrated into undergraduateand graduate curriculum.