Large computer networks typically require detailed configuration, applied at each node individually. This makes it hard to ensure that the total configuration of network nodes amounts to the intended behaviour of the network as a whole. To remedy this, one can separate the configuration of the network from the hardware implementing it. This technique, called Software Defined Networking (SDN) entails that network policy is decided centrally, after which relevant parts of a configuration implementing this policy are distributed among network devices.
As an added benefit of centralising network policy, it becomes possible to treat this policy as a single object, and reason about or even verify its properties. As an example, NetKAT [Anderson et al., 2014] is a framework which allows the user to specify global network policy in a domain-specific language. These expressions can be compiled [Smolka et al., 2015] to OpenFlow instructions and can run on actual hardware. Moreover, NetKAT allows the user to reason about the policy they wrote using straightforward algebraic laws, which accurately (i.e. soundly and completely) describe its semantics; as a matter of fact, this process can be automated [Foster et al., 2015] quite efficiently using coinductive techniques.
During the project have focussed on two lines of research:
1) Design and semantics of a probabilistic programming language that can be used for programming and verifying networks.
We have developed ProbNetKAT, a probabilistic programming language built on top of NetKAT that enables programming and verification of probabilistic features in SDN.
We develop McNetKAT, a scalable tool for verifying probabilistic network programs. McNetKAT is based on a new semantics for the history-free fragment of Probabilistic NetKAT in terms of nite state, absorbing Markov chains. This view allows the semantics of all programs to be computed exactly, enabling construction of an automatic verification tool. Domain-specific optimizations and a parallelizing backend enable McNetKAT to analyze networks with thousands of nodes. We used McNetKAT to automatically reason about general properties such as probabilistic program equivalence and refinement, as well as networking properties such as resilience to failures. We evaluated McNetKAT’s scalability using real-world topologies, compared its performance against state-of-the art tools, and developed an extended case study on a recently proposed data centre network design.
2) Development of a verification framework for properties with concurrent behaviour.
One feature missing from NetKAT is the ability to express concurrency, for instance when processing a packet using multiple tables simultaneously, or perhaps even multiple devices, such as a firewall and an intrusion detection system. We have worked extensively on developing tools for Concurrent Kleene Algebra: free models, complete axiomatisation, and decision procedures. We have then built on this foundation to develop a concurrent version of NetKAT, which enjoys the same benefits as NetKAT proper, including completeness of the algebraic theory, the possibility of running on network hardware and an efficient decision procedure.
