Practical Automation for Management Planes of Service Provider Infrastructure

Abstract

Managing service provider infrastructures (SPI) is ever more challenging with increasing scale and complexity. Network and container orchestration systems alleviate some manual tasks, but they are generally narrow solutions, with controllers for specific subsystems that do not coordinate on high-level goals, and fall far short of automating the full range of tasks that engineers face day to day. We seek to highlight the need for "practical automation" to manage SPIs. Via realistic examples, we argue that practical automation should provide cross-controller coordination, and should work within the reality that many tasks will involve humans. We describe a proof-of-concept system that leverages AI planning to synthesize management steps to move the system towards a goal state. A preliminary implementation shows that our approach can accurately generate plans for complex management tasks, while scalability and modeling diverse controllers remain as future challenges.