High-Quality Network Specifications via Coverage Analysis and Relational Abstractions

Abstract

Network misconfigurations frequently lead to outages that can impose severe social and economic consequences. Preventing these misconfigurations via automated testing or verification requires complete and precise specification of expected network behaviors. However, creating such specifications is challenging for network engineers due to the enormous scale and rapid evolution of modern networks. This dissertation develops two approaches to address this challenge. The first approach quantifies which network components are covered vs uncovered (and potentially untested) relative to given specifications. Results of this assessment help network engineers refine the specifications to make them more comprehensive. The second approach enables compact specification of network changes, which are the primary source of outages. These specifications capture the similarities and differences between two network states, which makes them complete and precise but compact. Network engineers need only specify the changing parts of the network, which tend to be small, and then simply mandate that all other behaviors “stay the same” without needing to enumerate them. Both approaches have been deployed and shown to be effective in real-world production networks.