70 likes | 86 Views
This research explores the limitations of relying solely on measurements to understand interdomain routing behavior and proposes building models to go beyond these limitations. It discusses the causes of routing dynamics and slow BGP convergence, examines AS-wide BGP behavior and its impact on traffic, and highlights the challenges of studying interactions between ASes. Additionally, it explores the potential of utilizing all available BGP data and considering new routing protocols to better understand and improve interdomain routing.
E N D
Next steps in interdomain routing research(why measurements are not enough to decide about it) Steve UhligUniversité catholique de Louvain, Belgium suh@info.ucl.ac.be
Internet-wide BGP behavior • Claim: BGP convergence can be slow [Labovitz et al. SIGCOMM’98&00] • Causes of routing dynamics: network engineering, flapping links, traffic engineering, misconfigurations,... • Causes of “slow” BGP convergence: Internet size, CPU load, BGP timers, route flap damping [Agarwal et al. PAM’04, Feldman et al. PAM’04] We don’t really know how well BGP performs
AS-wide BGP behavior • Impact of eBGP on traffic in tier-1 [Agrawal et al. SIGMETRICS’05]: not much dynamics inside AS due to external dynamics • Dynamics of hot-potato routing [Teixeira et al. SIGMETRICS’04] and impact on traffic [Teixeira et al. PAM’04]: high sensitivity of inside AS due to routing AS routing behavior depends on chosen AS and measured period [Teixeira et al. CCR 2005]
Inherent limitations of measurements • Even multiple vantage points give poor view of what’s going on [Feldmann et al. SIGCOMM’04] • Locating instabilities is hard, finding out what happened even harder • Studying individual ASes does not help much quantifying AS-wide routing dynamics • Studying interactions between ASes is one next step towards understanding Internet-wide BGP behavior (mentioned by Christophe already)
Inherent limitations of measurements • Can measurements alone help us decide whether interdomain routing should be changed? I don’t think so ! But... • Trying to make sense of all available BGP data might give some better idea if what is right or wrong with BGP • Thinking about new routing protocols (HLP, metarouting) also helps understand what are the alternatives to BGP
Building models to go beyond measurements limitations • Building models of Internet route propagation can help understand what matters to explain what we see (next step) • But simply reproducing AS paths is hard [Mühlbauer et al. SIGCOMM’06] • How do you deal with: • What is a policy? • What is a routing domain?
Building models to go beyond measurements limitations • Building models of Internet requires reverse-engineering the Internet! • What is the right granularity for: • Topology: ASes, routing domains, and routers • Policies: multiple peerings, filtering performed on peerings • Granularity of Internet model will depend on what you want to do with it.