AitF: Collaborative Research: Foundations of Intent-based Networking

University Of Wisconsin-Madison, Madison WI

Investigators

Abstract

Computer networks play an essential role in the day-to-day operations of businesses, organizations, and governments: they facilitate access to services and information as well as help protect against some types of cyberattacks. Unfortunately, current networks require highly-skilled network operators to provide detailed specifications of how the network should behave. This is a tedious and error prone process that limits how easily a network can evolve to meet emerging business needs and opens the door for subtle errors that can have a drastic impact on network availability, performance, and security. The goal of this project is to automatically produce the detailed specifications required by networking hardware from a set of high-level security and performance objectives specified by individuals who may have limited networking background. In other words, this project aims to allow administrators to focus on what the network should do rather than how it should be achieved. The broader impact of this project is to pave the way for increased network stability and security, and also to aid in training the next generation of network professionals. Automatically producing network configurations that satisfy a set of high-level policies and objectives (collectively referred to as "intent") requires both a language for network administrators to formally specify their intents and a mechanism for generating optimal and correct configurations for various types of networking hardware. To satisfy these requirements, the PIs plan to explore how program synthesis techniques can be applied and extended to network configurations. The project will lead to the design of synthesis techniques for generating specific types of intent implementations (e.g., traditional control plane configurations), as well as introduce domain-specific refinements to the chosen synthesis algorithms to ensure the time required for synthesis is practical and the resulting data and control planes are optimal (e.g., the configurations have minimal complexity). The algorithms produced by this research will advance the state of the art of program synthesis and provide new insights into how to apply program synthesis to other domains.