Towards systematic detection and resolution of network control conflicts

Abstract

The problem of detecting and resolving control conflicts has started to receive attention from the networking community. Corybantic is an example of recent work in this area. We argue that it is too coarse grain in that it does not model the combined operational objectives of multiple controller functions. This paper proposes a finer grain approach where a network control function is represented as a deterministic finite-state transducer. The machine runs on inputs provided by an SDN controller and outputs instructions that update the network as needed to meet objectives. Standard proof techniques and algorithms can be leveraged to analyze properties of these machines. Specifically, their intersection describes precisely the stable operating region of a network when the machines operate in parallel. The e region comprises conditions under which no control function is in the process of updating the network.