Geoff Huston finds two or more DNS listeners in 60% of sampled lookups

Geoff Huston Counted the DNS Listeners. The Fastest Answer Won the Credit.

Why it matters

When a browser asks for a name, the first resolver to answer gets the applause. Geoff Huston's APNIC Labs experiment found another audience: in its sample, the original question reached two or more recursive resolvers 60% of the time. Every one of those listeners saw the name.

Sofia Ren2 November 2026Reading Time 2 minLast update: 2 November 2026

Geoff Huston measured which recursive resolvers saw a DNS question, a different list from the first resolver to answer.

Summary

In Geoff Huston's APNIC Labs account, published 28 May 2026, the client stub passed the original DNS question to two or more recursive resolvers 60% of the time in the measured sample.
APNIC's resolver-use figures count resolvers that saw the original question. The first resolver to ask APNIC's authoritative server is a different measure: it is the likely first answer the user receives. Observer reach and response speed cannot share one scoreboard.

A question with more than one audience

APNIC Labs plants a unique DNS name in each browser-based test. Its authoritative server logs which recursive resolvers seek that name. The method ties the test user to the resolver addresses seen by APNIC, then groups those addresses into service categories. Huston's published analysis says millions of distinct users run the measurement each day. The unique name keeps an earlier cached answer from swallowing the question before it reaches the measurement server.

The result is easy to misread. An application may hand the original query to several recursive resolvers. One can win the race and be the resolver the user experiences. The others still receive the question. APNIC's 60% figure describes multiple recipients of the original query inside this experiment. It does not count 60% of all Internet users or promise that every resolver gives a separate answer.

Huston reports a second contrast. In APNIC's categories, the resolver in the user's own autonomous system appeared for 65.2% of measured users in 2026, compared with 65.0% in 2022. The group of known open resolvers appeared for 21.7%, compared with 20.0%. These are observed-use categories from the experiment, not a tidy sum of mutually exclusive customers. The stable same-AS share makes the forwarding path worth examining instead of assuming the public resolver won every household.

The business and privacy question sits behind the methodology. A speed test tells us who answered first. An observer test tells us who received the name, even when another service answered. A provider can gain visibility without appearing as the winner on a latency chart. Huston explicitly says his figures use the set of resolvers that saw the original query. A published per-client distribution of how many distinct operators saw each sampled query would sharpen that visibility picture.

Sources: APNIC Labs analysis by Geoff Huston; APNIC author profile.