An analysis of convergence delay in path vector routing protocols

Path vector routing protocols such as BGP are known to suffer from slow convergence. In response a number of convergence enhancements have been proposed, but convergence dynamics have yet to be well understood and there has been no general framework to assess and compare the various improvement algorithms. In this paper we present a general framework to analyze the convergence delay bounds of path vector routing protocols, under the assumption of shortest path policy and single link failure. Our framework takes into account important factors including network connectivity, failure location, and message processing delay. It applies to all path vector protocol variants(standard path vector protocol and convergence improvement algorithms) and allows us to develop analytical bounds that were not previously available, such as the delay bounds for path fail-over for BGP and most of the convergence enhancements. Our analysis shows that BGP fail-over delay bounds are determined largely by a combination of two factors: 1) the distance between the failure and the prefix origin and, 2) the length of the longest alternate path used to reach the destination after the failure. These factors are captured formally and can explain why existing convergence enhancements often provide only limited improvements in fail-over events. Furthermore, explicitly modeling message processing delay reveals insights into the impacts of topology structure (e.g. richness in connectivity) and different effectiveness of different enhancements. These new results allow one to better understand the behavior of various path vector protocols under given topology structure, network size, and message delays.