Anycast prefixes by RIR: ARIN, RIPE NCC, APNIC, LACNIC and AFRINIC

The LACeS anycast census relies on two primary measurement platforms to identify and geolocate anycast deployments:

• TANGLED Anycast Testbed: Used for the Anycast-Based (AB) component. LACeS utilizes 32 globally distributed nodes from TANGLED to send synchronized probes (ICMP, TCP, DNS) to each target, identifying unique responding sites. See: research paper
• CAIDA Archipelago (Ark): Used for the Great Circle Distance (GCD) component. We leverage latency measurements from a large number of geographically distributed Ark vantage points to geolocate anycast sites by applying the "speed-of-light" constraint. See: research paper

Quick summary:

• Use (AB > 3) || (GCD > 1) when high confidence is needed.
• Use (AB > 1) || (GCD > 1) (i.e. all prefixes) when completeness is needed.

False positives (AB). The anycast-based approach (AB) can produce false positives — especially when AB reports fewer than 3 sites. This is a known limitation of the underlying methodology; see MAnycast2.

False negatives (GCD). The latency-based approach (GCD) is highly accurate, but may miss prefixes deployed only in small geographic regions (regional anycast).

Partial anycast. Because probing runs at /24 granularity, some prefixes contain a mix of unicast and anycast addresses. Around 1,000 /24s are partially anycast (identified via bi-annual /32 scans). These are flagged with the partial column, but the flag may not reflect short-term changes.

Each row in the dataset represents one /24 (IPv4) or /48 (IPv6) prefix detected as anycast by at least one method. Column meanings:

Both AB and GCD consider anycast confirmed when the number of detected sites is greater than 1. The backing_prefix and ASN columns are sourced from CAIDA's prefix2as.

IPv4. Each protocol has its own hitlist with up to five ranked targets per /24, probed in order until one responds. Every hitlist starts with anycast targets from earlier censuses, addresses that responded to the same protocol in the last 7 daily censuses, and bgp.tools' anycast-prefixes. The ICMP hitlist adds the USC/ISI ANT IPv4 hitlist and OpenINTEL A record addresses; the TCP hitlist adds OpenINTEL A record addresses. The DNS hitlist adds name servers from OpenINTEL infra:ns and the DNS Dependency Graph, open resolvers from ODNS, and public DNS nameservers. All hitlists are rebuilt weekly.

IPv6. The same structure at /48 granularity. ICMP and TCP use AAAA record addresses from OpenINTEL, TUM's IPv6Hitlist (ICMP, TCP/80 and TCP/443 responsive addresses), and TU Dresden & HAW Hamburg's IPv6-SRA hitlist. DNS uses name servers from OpenINTEL infra:ns and the DNS Dependency Graph, public DNS nameservers, and TUM's UDP/53 responsive addresses. To keep probing times manageable, only the first /48 of aliased prefixes is scanned. For a more complete view, join with TUM's aliased-prefix set.

The census runs daily and is updated in this repository automatically.

If you use this dataset in academic work, please cite:

Hendriks, Luckie, Jonker, Sommese, van Rijswijk-Deij. LACeS: An Open, Fast, Responsible and Efficient Longitudinal Anycast Census System. IMC 2025. doi:10.1145/3730567.3764484 · PDF

@inproceedings{10.1145/3730567.3764484,
  author    = {Hendriks, Remi and Luckie, Matthew and Jonker, Mattijs
               and Sommese, Raffaele and van Rijswijk-Deij, Roland},
  title     = {LACeS: An Open, Fast, Responsible and Efficient
               Longitudinal Anycast Census System},
  booktitle = {Proceedings of the 2025 ACM Internet Measurement Conference},
  year      = {2025},
  pages     = {445--461},
  doi       = {10.1145/3730567.3764484},
  series    = {IMC '25}
}

All tooling is publicly available under MPL 2.0:

• MAnycastR — measurement tooling for AB and GCD measurements.
• MiGreedy — optimised implementation of the iGreedy latency-based geolocation algorithm.

The RIPE Atlas integration excludes probes with inaccurate reported locations using the list published by K. Izhikevich et al., “Rooting Out Incorrect RIPE Atlas Probe Geolocations” (ACM IMC ’26, doi:10.1145/3777912.3839804), available at kizhikevich/violating_ripe_probes.

The LACeS census data is available programmatically via a REST API hosted on this server. All endpoints return JSON and require no authentication.

Full documentation with parameter descriptions, response schemas, and interactive examples: Swagger UI · ReDoc

This census depends on the following publicly available datasets and resources: