Measuring I-BGP Updates and Their Impact on Traffic

Snapshots of BGP tables and updates have been studied in the past to understand the convergence time and global routing stability at the protocol level. However, very little has been done to carefully analyze the causes behind these BGP updates and their impact on traffic. We bridge the gap by presenting a systematic approach for correlating Internal BGP(I-BGP) updates with packet traces in a large-scale operational backbone. The I-BGP updates were collected from a route reflector in the backbone while the packet traces were collected from several OC-12 links at a Point-of-Presence(POP). We observe that continuous IBGP noise (100-250 updates/minute) is present at all times but there are periods of high ”churn” affecting a significant portion of the BGP table. We focus on the subset of updates that can potentially affect the traffic trajectory (e.g., Nexthop change), and find that during churn periods, as few as 6% of these updates affect network prefixes that carry as much as 80% of the traffic. Preliminary analysis shows no discernible correlation between I-BGP churn and packet loss/reordering. We also discuss our initial effort to identify the possible sources of I-BGP updates including changes in network reachability, internal versus external BGP policies.