Explore a 17-minute conference talk from NSDI '22 that introduces Differential Network Analysis (DNA), a novel approach to analyzing network control plane changes. Learn how DNA efficiently computes differences in end-to-end forwarding behavior resulting from control plane modifications, addressing limitations of traditional control plane verifiers. Discover the three-stage incremental computation process involving control plane simulation, data plane modeling, and property checking. Understand how DNA leverages differential dataflow programming, incremental data plane verification, and customized graph algorithms to achieve significant performance improvements. Examine evaluation results demonstrating DNA's ability to compute reachability differences up to 3 orders of magnitude faster than state-of-the-art control plane verifiers, using both real and synthetic control plane changes.
Overview
Syllabus
Intro
Existing control plane verifiers
Changes are often small Analyzing from scratch is wasteful
Hard to select properties to reverify
Make differences first class citizens!
Differential control plane simulation
Modular, three-stage framework Incrementally
Evaluation: end-to-end
Evaluation: three stages
Evaluation: generality & scalability
Taught by
USENIX
