CHASING THE UNKNOWN: A PREDICTIVE MODEL TO DEMYSTIFY BGP COMMUNITY SEMANTICS

Abstract

The Border Gateway Protocol (BGP) specifies an optional communities attribute for traffic engineering, route manipulation, remotely-triggered blackholing, and other services. However, communities have neither unifying semantics nor cryptographic protections and often propagate much farther than intended. Consequently, Autonomous System (AS) operators are free to define their own community values. This research is a proof-of-concept for a machine learning approach to prediction of community semantics; it attempts a quantitative measurement of semantic predictability between different AS semantic schemata. Ground-truth community semantics data were collated and manually labeled according to a unified taxonomy of community services. Various classification algorithms, including a feed-forward Multi-Layer Perceptron and a Random Forest, were used as the estimator for a One-vs-All multi-class model and trained according to a feature set engineered from this data. The best model's performance on the test set indicates as much as 89.15% of these semantics can be accurately predicted according to a proposed standard taxonomy of community services. This model was additionally applied to historical BGP data from various route collectors to estimate the taxonomic distribution of communities transiting the control plane.