2026 Volume 34 Pages 898-912
Network troubleshooting requires operators to identify the causes and impacts of network failures. Operators investigate packet forwarding related to failures through continually probing networks with testing and monitoring tools. While these tools depend on concrete packet headers at a probe, operators must consider the prohibitive number or combination of parameters such as packet header constraints and forwarding paths. We propose Lupe, a system that enables operators to (1) specify symbolic packet headers and forwarding paths and (2) obtain all packet forwarding behaviors matching the symbolic values for identifying the failure causes and impacts. For instance, operators can obtain the answers to the following questions from Lupe by single queries: (1) “Show forwarding paths destined for a prefix related to a failure” and (2) “Show all types of packet headers forwarded through a flapped link.” We design efficient data models, algorithms, and a query language for the extraction. Our evaluation shows that Lupe extracts forwarding behaviors matching symbolic headers or paths in a production datacenter network (>2000 network devices) in 15 seconds. Furthermore, we confirm the usefulness of Lupe through two case studies to help operators identify the causes and impacts of real failures in campus and production datacenter networks.
