Improving Network Understanding

One cannot imagine the world today without the Internet, as it has become an integral part of our daily lives. However, with all the benefits and opportunities it brings also come enormous availability and reliability requirements, which put immense pressure on the operators running the individual networks of the Internet. They need to avoid disruptions at all costs and resolve outages as fast as possible. Unfortunately, this is a highly challenging task due to the sheer complexity of these networks. This dissertation focuses on assisting network operators in one aspect of their daily work: network understanding. To this end, we built three systems that automate and improve network understanding to allow the network operators to direct their full attention to the mission-critical tasks requiring their expert insights. First, we developed Net2Text, a system which assists network operators in understanding their network’s forwarding behavior. Based on the operators’ queries, it automatically produces succinct summaries of the raw forwarding state. The key insight behind Net2Text is to pose the problem of summarizing the network-wide forwarding state as an optimization problem that aims to balance coverage, by explaining as many paths as possible, and explainability, by maximizing the provided information. Second, we developed Config2Spec, a system which assists network operators in understanding their network’s configuration. It automatically mines the network’s specification, which consists of all the policies that the configuration enforces. The key insight behind Config2Spec is to combine two well-known techniques: data-plane analysis and control-plane verification. This combination allows to prune the large space of candidate policies efficiently and then validate the remaining ones. And third, we developed Metha, a system which helps network operators understand the capabilities of their network validation tools. It finds inaccuracies in the underlying network models using differential testing. The key insight behind Metha is to leverage grammar-based fuzzing together with combinatorial testing to ensure thorough coverage of the search space using syntactically- and semantically-valid configurations