A System for Fast, Accurate, and Intuitive Querying of Large-Scale BGP Datasets#

Authors: Thomas Holterbach, Thomas Alfroy (bgproutes.io), Abbas Mohsenpour (UCLouvain), Thomas Krenc (IIJ Research Laboratory), kc Claffy (CAIDA), Cristel Pelsser (UCLouvain)

• Why: Not a paper about networking per se, but a strong signal of where infrastructure tooling is heading: LLM-synthesized code over purpose-built compressed data stores. The pipeline — measurement insight (BGP data is highly redundant and compressible) enabling a domain API that an LLM can drive — is a template worth stealing for any large operational dataset you own.
• What: ChatBGP, a domain-specific chatbot that turns plain-English BGP questions into Python code running against a compressed BGP database, returning answers in seconds rather than the hours that MRT-archive analysis takes. Four contributions: a measurement study showing BGP data is highly redundant and compressible; a database scheme exploiting that redundancy for fast retrieval; an expressive API over the database; and a prompt-engineering scheme guiding ChatGPT to synthesize optimized, minimal-retrieval Python against that API. (from abstract only)
• How: Redundancy-aware compression plus a query-shaped API, with the LLM constrained to synthesize code that retrieves only the data each query needs.
• Where applicable: RIPE RIS / RouteViews-class BGP data; aimed at operators diagnosing connectivity issues, researchers, and students. Note the accuracy depends on LLM code synthesis — attractive for triage and exploration, but verify before acting on results in a production incident.