Stability Analysis of Global BGP Routing Tables Using Time-Series and Correlation Models

O.L.M. Smith; K.N. Kantor

doi:10.31838/1mxxp805

Authors

O.L.M. Smith Departamento de Engenharia Elétrica, Universidade Federal de Pernambuco - UFPE Recife, Brazil Author
K.N. Kantor Departamento de Engenharia Elétrica, Universidade Federal de Pernambuco - UFPE Recife, Brazil Author

DOI:

https://doi.org/10.31838/1mxxp805

Keywords:

Routing stability, BGP dynamics, time-series modeling, route churn, prefix oscillation, Internet backbone, anomaly prediction.

Abstract

The paper examines the temporal stability of the global routing tables of Border Gateway Protocol (BGP) by studying periodic changes, fluctuations in the routing table, and instability events through time-series and correlation based statistical models. BGP data at large Internet Exchange Points (IXPs) and repositories such as RouteViews and RIPE RIS were analysed using a ten-year period to determine the number of prefix oscillations, session resets, and changes in AS-paths. The analysis of autocorrelation and spectral density showed that there were cyclic patterns of instability, which were usually associated with local peering updates and a new route redistribution periodically. Besides, cross-correlation between Autonomous Systems (AS) was an indicator of localised disruption that spread around the world via interconnected backbones. The statistical anomaly detection model to predict the possible instability events was worked out with the adaptive thresholding and regression-based residual tracking. Findings support the supposition that routing instability has demonstrably temporal periodicity and that proactive forecasting of routing instability can be highly advantageous in improving network resilience and fault-tolerance in large-scale Internet routing systems. The results can be useful to network operators and researchers in the development of predictive stability models of the dynamic BGP ecosystem.