• BGP topics to be discussed in the next few weeks: –Excessive route update –Routing instability –BGP policy issues –BGP route slow convergence problem –Interaction between BGP/IGP and among BGP components –Anti-IP-spoofing with BGP –New EGP proposals BGP Routing Stability of Popular Destinations Jennifer Rexford, Jia Wang, Zhen Xiao, and Yin Zhang • Some causes of BGP route changes: – Equipment failures. – Policy changes – Intra-domain topology changes • Potential problems caused BGP route changes: • One “event” triggers a long sequence of updates – CPU – Changing paths with traffic can cause congestion • Transient loops • Make it hard to direct (engineer) the traffic • What is the current situation: – A large fraction of prefixes have stable BGP routes – A small fraction of prefixes are responsible for the majority of Internet traffic – Are prefixes receiving a large volume of traffic more or less stable than prefixes receiving a lower volume of traffic? • Intuitively, more traffic can cause more changes • Popular sites have well managed multiple connections to the Internet. • How the study is done? – BGP routes and updates in RouteViews and RIPE NCC are publicly available – This study adds one monitor in the ATT backbone – The anomalies are removed: • Burst updates due to router failure • Redundant advertisements: – Multiple updates for the same route – Withdraw before announce – Updates or events • An event can cause a lot of updates • Routing stability is better reflected by events • How to get events from updates? – Updates spaced close together in time are counted as one event – This may not be accurate. • Grouping events: 45seconds/75seconds • Event duration: mostly < 5 mins • A small number of prefixes are responsible for most updates events • Update event vs. traffic volume – Most traffic goes to a small number of prefixes • Update event vs. traffic volume – Prefixes responsible for most update events do not receive a lot of traffic • Explanations: – Unstable prefixes tend to be unpopular • Unstable BGP routes make it different for other hosts to reach the destinations. They cannot be popular. • Popular prefixes do not experience many events • Top websites cause very few update events • Conclusion: – The majority of the update events are concentrated in a few prefixes that do not receive much traffic – Popular sites almost have no updates – Implications: suppressing updates mostly likely will not cause disruption of the Internet. • Who are the prefixes that cause most of the updates? • How long does the instability last? • Can we do something about it?