Optimizing MRAI on large scale BGP networks: An emulation-based approach

Modifying protocols that pertain to global Internet control is extremely challenging, because experimentation is almost impossible and both analytic and simulation models are not detailed and accurate enough to guarantee that changes will not affect negatively the Internet. Federated testbeds like the ones offered by the Fed4FIRE+ project offer a different solution: off-line Internet-scale experiments with thousands of Autonomous Systems (ASes). This work exploits Fed4FIRE+ for a large-scale experimental analysis of Border Gateway Protocol (BGP) convergence time under different hypotheses of Minimum Route Advertisement Interval (MRAI) setting, including an original proposal to improve its management by dynamically setting MRAI based on the topological position of the ASes in relation to the specific route being advertised with the UPDATE messages. MRAI is a timer that regulates the frequency of successive UPDATE messages sent by a BGP router to a specific peer for a given destination. Its large default value significantly slows down convergence after path changes, but its uncoordinated reduction can trigger storms of UPDATE messages, and set off unstable behaviors known as route flapping. The work is based on standard-compliant modifications of the BIRD BGP daemon and shows the tradeoffs between convergence time and signaling overhead with different management techniques.