In this talk, we provide analysis and data from a sixteen-month study of Internet BGP route convergence latencies. We first describe our experimental instrumentation of the Internet, which included a number of geographically and topologically diverse BGP fault injection and route collection probe machines. We then describe the measured response of BGP after several types of routing events, including single route failures, multi-homed fail-over, and route restoral. We provide analysis and a brief theoretical description of expected and worst-case BGP convergence behaviors. Finally, we dispel several tenets of conventional networking wisdom about BGP and routing convergence.