Noisy Routers: Investigating the Make-Up of Route Collector Data

Key Takeaways

• Top 5% of peers produce 56% of all BGP updates.
• A single peer (AS140627) contributed up to 2.93 billion updates in one day.
• 19 sessions (≈2% of total) accounted for two‑thirds of December 2021 updates.
• 17% of prefixes generated 90% of update volume, indicating path‑specific noise.
• Perth collector saw 3.3 M updates per noisy prefix, far above other collectors.

Pulse Analysis

BGP’s role as the Internet’s routing backbone makes its data a critical resource for operators and researchers. Yet the sheer volume of updates—now exceeding a billion daily—has become a storage and processing burden, largely because a small set of peers repeatedly flood collectors with redundant announcements. This "noise" inflates MRT archives, complicates convergence analysis, and raises operational costs for projects like RIPE RIS and RouteViews, which must retain massive datasets for historical diagnostics.

The study’s granular breakdown reveals a stark skew: the top 5 % of collector‑peer pairs generate more than half of all updates, and a handful of sessions dominate monthly traffic. Such concentration means that any measurement relying on raw BGP feeds risks bias, over‑representing the behavior of a few misbehaving routers. For network operators, the findings underscore the need to pinpoint noisy sessions rather than bluntly disabling peers, preserving valuable connectivity while mitigating storm‑induced overloads. Researchers can now design pruning strategies that retain essential routing dynamics without the clutter of repetitive, low‑value updates.

Looking ahead, extending this analysis to other collector platforms will test whether the observed patterns are universal. Collaboration with operators to remediate misconfigurations could dramatically reduce noise, improving the fidelity of routing telemetry. Moreover, selective pruning—removing high‑frequency, low‑information updates—offers a pragmatic path to slimmer archives without sacrificing diagnostic capability, benefiting both academia and industry as they strive for more accurate, cost‑effective Internet measurement.