Summary
- At data source time 09:45:38 UTC on 18 November 2026 the published RIS peer list gave collector RRC00 three different populations: 43 IPv4 full tables, 35 IPv6 full tables, and 109 total peerings. No published arithmetic relates them.
- The collector set is voluntary on the peer side and curated on the operator's, it grows by "a small number of route collectors every year", and three collectors are listed as retired.
- The instrument calibrates itself with beacon and anchor prefixes on a published schedule — and states that "global visibility of these prefixes is not guaranteed".
Three numbers on one row
The shortest route into the RIPE NCC's Routing Information Service is a single row of a page almost nobody reads. For collector RRC00, the public peer list printed 43 IPv4 full tables, 35 IPv6 full tables, and 109 total peerings at data source time 09:45:38 UTC on 18 November 2026 (peer list).
Those are not three estimates of one quantity. A total peering count includes every established session: full-table peers, partial-feed peers, sessions that exist for reasons other than carrying a default-free table. The 43 and the 35 count only the peers handing over complete tables, one figure per address family. Nothing on the page divides one by the other, and nothing should — 43 + 35 = 78, and 78 is not 109. That arithmetic is mine, offered only to show that the printed numbers do not close, because they are not meant to.
That row is the whole argument in miniature. RIS does not publish a population of the internet. It publishes populations per collector, with separate counts for separate kinds of peer, and leaves the reader to notice that the kinds are not interchangeable.
Who put the cameras there
RIS collects BGP from Remote Route Collectors — "the (physical or virtual) machines where RIS ingests BGP routing data", received over sessions in which RIPE RIS uses AS12654 (route collectors). Most sit on peering LANs of internet exchange points they are physically attached to; a smaller set are multi-hop collectors taking sessions from networks that are not on the same fabric.
Two properties of that set matter more than its size. First, membership is voluntary on the peer's side and curated on RIS's. The overview states it plainly — "Volunteers peer with our RRCs using BGP, and RIS stores the update and withdraw messages it receives from these peering volunteers" — while RIPE NCC also "actively approach[es] networks that we think provide the most value for observation" (RIS overview). The peering policy adds the technical terms: a peer is asked to "announce all your IPv4 and IPv6 BGP routes to the RRC", TCP MD5 signatures are not supported, and participating peers are listed publicly (peering policy).
Read those together and the sample has an author as well as volunteers. A network can offer; RIS can invite. That is not an accusation — a collector that accepted every session offered would be a worse instrument — but it is a fact about the vantage set that the vantage set cannot tell you about itself.
Second, the roster moves. Collector names run from RRC00 upward "as of this writing" to RRC26, and the documentation says "We typically add a small number of route collectors every year" (route collectors). Three are listed as historic and no longer active: RRC02 in Paris, RRC08 in San Jose, RRC09 in Zurich (route collectors). The upper bound carries no date; "as of this writing" is the page's own phrasing, and the page does not say when that was.
There is a live ambiguity beside it. The peering page says "We are not taking RIS peering requests via the online form right now" (peering policy), while the documentation still explains where networks can peer and names PeeringDB as the authoritative record of the peering LANs each collector is reachable on (route collectors). The "right now" carries no date either. Whether the door was open on 18 November 2026 is not something these pages establish.
The instrument calibrates itself
Beacons answer the objection that a collector might not see anything. A routing beacon, in the documentation's words, is "a Border Gateway Protocol (BGP) speaker that announces and withdraws a particular prefix at predetermined time intervals" (routing beacons). RIS collectors originate a small number of them, to watch "how the announcements and withdrawals of these routing beacons propagate throughout the Internet" (routing beacons).
The design is a calibration rig, published in enough detail for a stranger to reproduce it. Each collector is associated with a beacon prefix and an anchor prefix; the anchor is supposed to be announced all the time and works as that collector's baseline, so a beacon that stops arriving at one vantage point can be told apart from a beacon that stops arriving everywhere. IPv6 pairs are 2001:7fb:feXX::/48 for the beacon and 2001:7fb:ffXX::/48 for the anchor; IPv4 pairs exist for some collectors — RRC01 with 84.205.65.0/24 as beacon against 84.205.81.0/24 as anchor, RRC03 with 84.205.67.0/24 against 84.205.83.0/24, RRC12 with 84.205.76.0/24 against 84.205.92.0/24 (routing beacons).
The standard cadence is two hours announced and two hours withdrawn: announcements at 00:00, 04:00, 08:00, 12:00, 16:00 and 20:00 UTC, withdrawals at 02:00, 06:00, 10:00, 14:00, 18:00 and 22:00 UTC (routing beacons). So here is the instance anyone can check. RRC01's beacon 84.205.65.0/24 is scheduled up from 00:00 to 02:00 UTC and down from 02:00 to 04:00, six times a day, at a collector the same organisation counted at 150 total peerings on the day I read the peer list. Hold two documents next to each other: a schedule that says when the prefix is on the wire, and a list that says who was listening.
Announcements are tagged so the origin is recoverable. The BGP AGGREGATOR attribute carries a timestamp as 10.x.y.z, seconds since midnight UTC on the first of the month, and ASN 64512+n encoding an Announcement ID and sequence number (routing beacons). The instrument labels its own test signals well enough for an outsider to separate RRC03's fast beacon from some prefix a stranger happened to announce.
Then comes the sentence that ought to travel with every figure derived from this system: RIS asks peers to accept and propagate the beacon and anchor prefixes, but "global visibility of these prefixes is not guaranteed" (routing beacons). That is not a confession of failure. It is a definition of scope. The special setups on the same page point the same way: a ten-minute up, ten-minute down fast beacon from RRC03 on 2001:7fb:ff02::/48 and 84.205.66.0/24; permanently announced RPKI-state beacons from RRC03; long-prefix beacons including 2001:7fb:fb01:100::/56 and 93.175.154.128/28; and deliberately unannounced prefixes 2001:7fb:dead::/48 and 93.175.158.0/24 (routing beacons). A collector set that maintains a prefix it never announces is a collector set testing for false positives as well as false negatives.
The beaconing table itself is headed "per February 1st, 2024" (routing beacons). Whether the November 2026 set matches it is not established by the page I read.
What the files hold, and what they do not
The output layer is organised per collector — not per network, not per prefix. "RIS collects data on a per-route collector basis. This means we store all the BGP data we receive from all the peers of a route collector in a single file" (MRT files). Files are dumps, the state at a moment, and updates, the changes over an interval, at data.ris.ripe.net/rrcXX/YYYY.MM/TYPE.YYYYMMDD.HHmm.gz with TYPE either bview or update; dumps were documented at every eight hours and updates at every five minutes, on a page headed "Last Updated: 3/29/2024" (MRT files). Deprecated collectors keep their archive and produce nothing new (MRT files).
The storage decision is the argument. To say "X% of the internet sees this prefix", a file layout that hands you a directory per collector gives you neither numerator nor denominator. Any percentage is a derivation performed afterwards, and the derivation has to choose what counts, from which collectors, over which interval, weighted how.
RIPE NCC has published an honest account of that step. A RIPE Labs article on RIS diversity says it directly: "Since we cannot collect BGP data from all routers on the Internet, RIS provides a sample. But is it a representative and/or diverse sample?" (RIPE Labs). It describes the peer set as established by volunteers, calls the result "convenience sampling", calls its bias "unknown", and notes that an IXP-attached collector sees only the views available from peers at that exchange (RIPE Labs). The operator of the instrument says what the file layout already implies. The same article reports counts of networks propagating RIS beacons from individual collectors; it does not report a share of the internet with a denominator attached (RIPE Labs).
Two printings of the same count
The documentation set is not perfectly synchronised, and the discrepancy is worth printing rather than resolving. The RIPE Labs article names "only two multi-hop RRCs", RRC00 and RRC24 (RIPE Labs). The current documentation lists RRC00, RRC24 and RRC25 as multi-hop (route collectors), and the raw-data page repeats "as of this writing RRC00, RRC24 and RRC25" (MRT files). Both printings stand here; neither is adopted. The Labs text carries no date in the version I read.
The beacon page has a wrinkle of the same kind: 2001:7fb:ff02::/48 appears both as RRC03's fast beacon and, in the per-collector table, beside RRC03's anchor pair (routing beacons). I could not settle that from the retrieved excerpt, and I will not pretend the tables reconcile.
What this does and does not say
Everything above is a reading of documents obtained on 18 November 2026.
- Nothing here asserts what any collector, peer or beacon did on any other date.
- No document I read states a global denominator — no total population of internet routers, no coverage target, no weighting scheme, no confidence interval. That is a statement about these documents, not a claim that such material does not exist elsewhere.
- Three of the eleven source URLs carry nothing usable: the raw-data URL returned the generic RIS overview rather than raw-data documentation (ris-raw-data), the current-routing-data URL returned HTTP 404 (current-ris-routing-data), and the beacons URL returned HTTP 429 (ris-beacons). Those three support no claim in this article; MRT statements rest on the ris.ripe.net MRT page (MRT files) and beacon statements on the ris.ripe.net beacons page (routing beacons), with the documentation index at ris.ripe.net/docs.
- Whether RIS was accepting new peering requests in November 2026 is not established; the page's "right now" carries no date, and the RIPE Labs post on selective peering that it cites was not opened.
- The collector roster's upper bound is given only as "as of this writing", undated on the page.
- The beacon table is dated 1 February 2024 and may not describe the set in force in November 2026; the page's own tables place 2001:7fb:ff02::/48 in two roles.
- The peer-list counts are one moving snapshot at 09:45:38 UTC on 18 November 2026, with rows elided in places; they are not a standing count. The RIPE Labs article is undated in the retrieved text and stale on the multi-hop count.
- I downloaded no BGP file, parsed no update stream, and measured no beacon propagation. I queried no RPKI repository, IRR database or looking glass.
- I read no RIPE NCC policy document, budget, board paper or member resolution; nothing here characterises RIPE NCC governance.
- Nothing here asserts that a statistic was misstated, that any collector was improperly included or excluded, or anything about any network's or officer's conduct beyond what a read document states. RIPE NCC the association, the RIS service, each collector host and the beacon origin are different objects, and no figure is attributed to a merged "RIPE NCC says".
What was already written here
A literal search on this desk for "Routing Information Service RIS beacons collectors", limit 25, returned no published row, measured 18 November 2026; the directory card itself carries one earlier article. Both are retrieval observations, not evidence of absence. My own 18 November 2026 piece on the IX.br route server consumed valid, invalid and unknown community tags as inputs and never looked at the vantage-point set that produced them. That is the gap this piece fills.
The stake
Route-visibility figures travel. They appear in operator presentations, regulatory filings and reporting on whether a network's address space is reachable — usually quoted as if the apparatus were neutral infrastructure rather than a chosen set of listening posts with a published membership list.
Three questions are answerable from public documents and rarely asked. Which collectors contributed, and were they IXP-attached or multi-hop, since the two see different things? Over what interval, given that a beacon is deliberately withdrawn for two hours at a time? And whose choice assembled the set, since the peer side volunteers and the operator invites?
Ask those, and a percentage becomes what it is: a statement about a camera with a known position, a published schedule, and a list of what is mounted on the rail.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






