We present DBNet, a data-driven network automation framework built on top of a DBMS. DBNet utilizes key primitives of a DBMS including tables, procedures, transactions, logging, and access control to serve the functions of a data-centric network control plane. DBNet accomplishes this functionality by storing mirrored network device states, executing automation programs on these mirror states within the DBMS, and proxying state updates out to the physical devices upon changes to mirror/local state. The framework also stores network telemetry data, performs analytics on the data, uses the analytics to motivate control plane actions, and provides provenance logging features on the actions taken. We apply DBNet to motivating cloud network infrastructure examples and show how developers can use DBNet's interface to express rich user-defined policies. Our preliminary case studies show that the overhead to run DBNet is trivial in the timescales generally relevant for network automation.

翻译：我们提出DBNet——一个构建于数据库管理系统（DBMS）之上的数据驱动网络自动化框架。DBNet利用DBMS的核心原语（包括表、存储过程、事务、日志记录和访问控制）来实现以数据为中心的网络控制平面功能。该框架通过存储镜像的网络设备状态、在DBMS内对这些镜像状态执行自动化程序，以及在镜像/本地状态发生变化时将状态更新代理到物理设备来实现这一功能。该框架还存储网络遥测数据，对数据进行分析，利用分析结果驱动控制平面操作，并为所执行的操作提供溯源日志记录功能。我们将DBNet应用于云网络基础设施的激励性实例，并展示开发者如何利用DBNet的接口表达丰富的用户自定义策略。初步案例研究表明，在与网络自动化通常相关的操作时间尺度上，运行DBNet的开销微不足道。