Stabilizing BGP routing without harming convergence

摘要

RFD and MRAI are the only two built-in mechanisms in BGP router against unstable routes, they can however negatively impact the convergence. In this paper, we propose a churn aggregation approach CAGG to stabilize BGP routing without harming convergence. CAGG is based on the observation that AS PATH change is the dominant cause for BGP updates and only a small number of AS PATHs are explored by each highly active prefix. A CAGG equipped router converts the multiple AS PATHs explored by a highly active prefix into an aggregated path, and propagates the aggregated path instead to reduce the number of resulted BGP updates from AS PATH changes. Our experiments with real BGP data show that CAGG can reduce as much as 50% of BGP updates, and 60% of BGP path exploration duration in its best case, while on average 28.1% and 32% respectively across 36 RouteViews monitors. Furthermore, CAGG is shown to perform better than both RFD and PED[1] in reducing BGP updates, path exploration duration and accelerating BGP convergence, at the cost of buffering around 5,000 AS PATHs.