Spectrum
Every MRT update file from RouteViews and RIPE RIS, parsed in a single stateless pass into file-level metrics: announcements and withdrawals, BGP path attribute usage, RPKI route-origin validation verdicts, bogon counts, compressed and uncompressed sizes, and parse cost. This page reads those metrics live and shows what the corpus looks like at file granularity.
loading…
Index coverage
Files present in the broker index (expected) versus files Spectrum has parsed, per UTC day. The gap is ingestion debt visible at file granularity, not routing activity.
Corpus volume
Bytes processed per day, as served (compressed) and fully expanded (uncompressed).
Per-collector coverage
Share of indexed update files parsed by Spectrum, per collector and day. Blank cells mean no indexed files for that day. Rows are ordered by total files in the window.
Attribute mix
BGP path attribute instances per day, top types across all collectors (last 56 days).
Attribute share
Share of all attribute instances per day. Rising bands show attributes spreading through the update stream.
Per-attribute volume
Instances per day for each attribute type seen in the update stream (last 56 days), sorted by total volume. Each tile is one attribute, so formation-specific behaviour stays visible.
Parse cost
Per-file parse time distribution per day, milliseconds (p50 and p95 across the day's files).
Method & limits
Spectrum is the deep extension of the BGPKIT broker MRT file index: each update file is downloaded, parsed once, and reduced to file-level metrics with no cross-file state. Rows are keyed by collector, data type and interval start, so every metric joins cleanly to the file index used by the broker.
- Coverage: expected counts come from the broker file index for the same UTC day bucket; parsed counts come from Spectrum's parse table (successful files only). Both are epoch-aligned UTC days; file intervals differ per project (RIPE RIS publishes every 5 minutes, RouteViews every 15).
- History: parsing starts 2026-08-11. Coverage gaps before mid-October 2026 are ingestion gaps from the pre-reconciliation discovery; they are not routing events. Recent days show the reconciled steady state.
- RPKI: verdicts use the global VRP snapshot (Cloudflare aggregate) at parse time, not at BGP event time; per-file valid / invalid / unknown element counts.
- Attributes: instance counts over the attribute types carried in the update stream (ORIGIN, AS_PATH, COMMUNITIES, and similar); counts are instances, not unique prefixes or paths.
- Parse cost: wall-clock parse duration per file on the worker that processed it; not a benchmark, and it varies with hardware and file size.
- Not covered yet: RIB dumps; per-session analysis lives in separate experiments.
Data sources: RouteViews and RIPE RIS. BGPKIT is not affiliated with the collectors; report an issue or a correction to contact@bgpkit.com.