Appearance
Data sources
Atlas is a composite of about a dozen upstreams. Knowing which one a number came from tells you how fresh it is, how authoritative it is, and who to talk to when it is wrong.
RIPE Atlas
| Data set | Cadence | Where it shows up |
|---|---|---|
| Probe and anchor lookups | live | Probe lookup, anchor detail |
| Measurement metadata | read-through cache | Measurements |
| Measurement results | live (ongoing) / cached (terminal) | Measurement detail |
| Connection events (msm 7000) | every 15 min | Statistics, Trends |
| Probe uptime (msm 7001) | continuous, change-compressed | Trends, probe history |
| Probe archive snapshots | daily, one per day | Graphs, Trends |
| Anchor mesh matrix | daily | Mesh views |
| Well-known measurements | hourly | Well-known |
| DNSMON roster | daily | DNSMON anchors |
| Ingestion delay metric | every minute | Data delay |
| Your credits, probes, measurements | live, via your own key | Dashboard |
Trust: authoritative for what it covers. Probe locations are the exception — those are host-declared and unverified.
Routing and network data
| Source | Provides | Cadence |
|---|---|---|
| RIS / RIS Live | Live BGP updates, routing history | live |
| PeeringDB | Networks, IXPs, facilities, presence | daily |
| APNIC | AS population estimates (users behind a network) | daily |
| manycast.net | Anycast prefixes, confidence scores, catchments | weekly |
| RIR / whois | Allocation and registry records | on lookup |
| RPKI | Route-origin validation status | on lookup |
Trust: RIS is observation and is as good as its vantage points. PeeringDB is self-reported — presence claimed, not verified. APNIC population is a modelled estimate, fine for ranking and wrong for precise counts. Anycast detection is probabilistic; see Anycast.
Geolocation
| Source | Provides | Cadence |
|---|---|---|
| Geofeeds (RFC 8805) | Operator-published netblock → location | daily; discovery weekly |
| MaxMind | Commercial IP geolocation | on lookup |
| IPmap | RIPE's geolocation | on lookup |
| Reverse DNS | Hostname hints | on lookup |
| Measurement latency | Upper bound on distance | derived |
Trust, strongest first: geofeeds (the operator saying where their own addresses are), then measurement constraints (which can disprove but never confirm), then commercial databases. Discovered geofeeds are held disabled until reviewed, so an unvetted feed never silently influences results. See Geofeeds and geolocation.
Reference data
| Source | Provides | Cadence |
|---|---|---|
| GeoNames | Cities, coordinates, place names | daily |
| UN/LOCODE | Unambiguous location codes | periodic |
| ISO country codes | The coarse geographic key | static |
| Weather | Hourly per-country temperature and alerts | no longer collected — stopped 4 Oct 2026; what was collected stays readable |
Derived by Atlas
Everything below is computed here from the sources above — no additional upstream load, and the methodology is ours:
- Probe and anchor health — rule evaluation, with an append-only transition log
- Mesh coverage and mesh quality — daily derivation over the mesh matrix
- Country coverage and AS population gaps — Deployments
- Weather ↔ outage correlation — de-trended, lagged; no longer recomputed since 4 Oct 2026; see Deployments
- Missed disconnects — Statistics
- Activity feed — a merge across the change logs
Live versus cached
The rule of thumb: lookups are live, aggregates are archived. A single probe, anchor or measurement is fetched from RIPE when you ask for it. Anything counting, ranking or trending across the fleet reads Atlas's own store and carries the cadence in the tables above.
Which means "today" is often absent from a daily series, and that is normal. See Data delay and freshness.