Improving network agility with seamless BGP reconfigurations
In this talk, we address the problem of deploying a new BGP configuration in a running ISP with no impact on the data-plane traffic. First, we show that the current best practices to reconfigure BGP (eBGP and iBGP) do not provide guarantees with respect to packet loss. In particular, we show that long-lasting routing and forwarding anomalies can occur even when the initial and the final BGP configurations are anomaly-free. Then, we study the problem of finding an operational ordering of the reconfiguration steps which guarantees no packet loss. Unfortunately, such an operational ordering, when it exists, is computationally hard to find. Finally, to enable disruption-free reconfigurations, we propose a framework which extends current carrier-grade routers to run two BGP control-planes in parallel. We present a prototype implementation and show its effectiveness through a case study.