This paper examines the latency in Internet path failure, failover, and repair due to the convergence properties of inter-domain routing. Unlike the public telephony network, where failover can occur within milliseconds, inter-domain routers in the packet-switched Internet may take tens of minutes to reach a consistent view of the network topology after a fault. These delays stem from temporary routing table oscillations during BGP path selection. During these periods, end-to-end Internet paths experience intermittent loss of connectivity, increased packet loss, and latency. The authors present a two-year study of Internet routing convergence, using experimental instrumentation at major Internet exchange points. They find that the measured upper bound on inter-domain routing convergence delay is significantly slower than previously thought, and that the theoretical computational bound on the number of router states and control messages exchanged during BGP convergence is factorial with respect to the number of autonomous systems. The study also highlights that specific router vendor implementation decisions and ambiguity in the BGP specification contribute to the observed convergence delays.This paper examines the latency in Internet path failure, failover, and repair due to the convergence properties of inter-domain routing. Unlike the public telephony network, where failover can occur within milliseconds, inter-domain routers in the packet-switched Internet may take tens of minutes to reach a consistent view of the network topology after a fault. These delays stem from temporary routing table oscillations during BGP path selection. During these periods, end-to-end Internet paths experience intermittent loss of connectivity, increased packet loss, and latency. The authors present a two-year study of Internet routing convergence, using experimental instrumentation at major Internet exchange points. They find that the measured upper bound on inter-domain routing convergence delay is significantly slower than previously thought, and that the theoretical computational bound on the number of router states and control messages exchanged during BGP convergence is factorial with respect to the number of autonomous systems. The study also highlights that specific router vendor implementation decisions and ambiguity in the BGP specification contribute to the observed convergence delays.