Network verification — When Clarke meets Cerf

摘要

Surveys reveal that network outages are prevalent, and that many outages take hours to resolve, resulting in significant lost revenue. Many bugs are caused by errors in configuration files which are programmed using arcane, low-level languages, akin to machine code. Taking our cue from program and hardware verification, we suggest fresh approaches. I will first describe a geometric model of network forwarding called Header Space. While header space analysis is similar to finite state machine verification, we exploit domain-specific structure to scale better than off-the shelf model checkers. Next, I show how to exploit physical symmetry to scale network verification for large data centers. While Emerson and Sistla showed how to exploit symmetry for model checking in 1996, they exploited symmetry on the logical Kripke structure. While header space models allow us to verify the forwarding tables in routers, there are also routing protocols such as BGP that build the forwarding tables. We show to go from header space verification to what we call control space verification to proactively catch latent bugs in BGP configurations. I will end with a vision for what we call Network Design Automation to build a suite of tools for networks inspired by the Electronic Design Automation Industry. (With collaborators at CMU, Edinburgh, MSR, Stanford, and UCLA.)