Measuring BGP pass-through times

Abstract

Fast routing convergence is a key requirement for services that rely on stringent QoS. Yet experience has shown that the standard inter-domain routing protocol, BGP4, takes, at times, more than one hour to converge. Previous work has focused on exploring if this stems from protocol interactions, timers, etc. In comparison only marginal attention has been payed to quantify the impact of individual router delays on the overall delay. Salient factors, such as CPU load, number of BGP peers, etc., may help explain unusually high delays and as a consequence BGP convergence times. This paper presents a methodology for studying the relationship between BGP pass-through times and a number of operationally important variables, along with some initial results. Our results suggest that while pass-through delays under normal conditions are rather small, under certain conditions, they can be a major contributing factor to slow convergence.