One lookup sent to a recursive resolver can come back out of it as up to about 1,620 times as many packets. That asymmetry is the whole of NXNSAttack, a DNS denial-of-service attack disclosed in May 2020 by academic researchers, and one of the resolution-path vectors a DDoS test has to account for. The multiplication happens inside the resolver itself: a single referral to many nameserver names that do not exist makes it send a burst of queries at whatever those names point to.
How NXNSAttack works
A referral is the ordinary DNS answer that says "I do not know, ask these servers". It names the nameservers for a zone, and when it leaves out their addresses (the glue), the resolver has to look each name up before it can continue.
That is the step NXNSAttack abuses. The attacker controls the authoritative server for a domain of its own, and makes resolvers ask about names under it. The answer is a referral listing many nameserver names and no addresses. The resolver dutifully tries to resolve every one of them.
The attacker chooses where those names live. Put them under the target's domain and each one becomes a fresh query to the target's authoritative servers, for a name that does not exist. That is the subdomain target, and the disclosure reported packet amplification of about 74x for it.
Aim the work at the resolver instead, so that it is the resolver's own fetching that piles up, and the figure reaches about 1,620x. Measured in bytes rather than packets, the reported amplification ran from 21x to 163x.
Why it matters under attack
The queries come from real resolvers. Against the subdomain target, the traffic reaching your authoritative servers is sent by legitimate recursive resolvers, often the large public ones your own users depend on. Blocking them blocks your customers too.
Every name is new. The fake nameserver names never repeat, so no cache absorbs them. The load on the authoritative tier looks the same as DNS water torture: a stream of one-time NXDOMAIN lookups. What NXNSAttack adds is the multiplier, so far fewer client queries are needed to reach the same rate.
Encrypted DNS does not help. DNS over TLS or HTTPS protects the path between the client and the resolver. The amplified traffic is the resolver's own outbound recursion, which is legitimate protocol behaviour, so encryption changes nothing about it.
Your resolver can be the instrument. An operator running recursive resolvers is exposed twice: as the server whose CPU and outbound capacity are burned, and as a source of the flood landing on someone else's name servers.
Hardening, and what a DDoS test measures
The fix sits in the resolver. Patched resolver software limits how many referral names it will chase for one client query, which cuts the multiplier at its source. Running an unpatched validating or caching resolver anywhere reachable from untrusted clients is the exposure to close first, and an open resolver widens it to anyone on the internet.
For the authoritative side, the defences are the ones that hold against any unique-name flood: anycast spread, NXDOMAIN rate limiting, and enough headroom per zone.
A test never borrows real third-party resolvers to reproduce this. For your own authoritative servers, it drives the query shape NXNSAttack produces, unique nonexistent names arriving from many resolver addresses, from the test's own generators, and reports the per-zone rate at which resolution for real names starts to slow. For resolvers you operate, the reading is simpler and more useful: whether a single client query can make the resolver fetch an unbounded number of names, checked in an authorised staging copy rather than against production.
The broader resolution-path picture, including how a DNS flood differs from these cache-defeating variants and how to test authoritative and recursive tiers separately, is laid out in DNS DDoS testing.