1. Why DNS is a critical dependency
The Domain Name System (DNS) translates names such as mail and login addresses into the numeric addresses that networks use. Most people never think about it until it fails. When it does, websites stop loading, email cannot be delivered, VPN and single sign-on logins time out and cloud applications look unavailable, even when the servers and connections behind them are healthy. DNS is quiet when it works, so it is often missing from risk registers.
It helps to separate two roles. Authoritative DNS answers questions about your own domains and is published to the whole internet. Recursive DNS, also called a resolver, looks up names on behalf of your users and devices, from the office network or a remote worker's laptop. Both can be targeted, and they fail in different ways. An attack on the authoritative side makes your services unreachable from outside. An attack on the recursive side stops your own staff from reaching anything, including internal applications.
2. How the main attacks work
A volumetric distributed denial-of-service (DDoS) attack floods the network link or the DNS servers with more requests than they can process. The goal is saturation: legitimate queries are dropped for as long as the flood lasts. Because the traffic arrives at the name servers themselves, the available bandwidth, the number of locations and the way traffic is routed matter as much as software settings. A well-tuned server on a small connection still fails when the link is full.
Amplification attacks abuse DNS servers that answer anyone on the internet. The attacker sends small requests with a forged source address, and the server returns larger responses to the victim. Random subdomain attacks work differently. They send queries for names that do not exist, which bypass caches and force the authoritative servers to do work on every request. Both patterns show why open answering is a risk on any server that is not meant to serve the public.
3. Architecture: anycast and separation of roles
Anycast is a routing technique in which the same address is announced from several locations. Internet routing sends each request to the nearest or best-connected location, so an attack is spread across sites instead of concentrated on one. A site can also be withdrawn from routing during an incident without changing the address that clients use. Anycast is powerful, but it needs careful capacity planning, because a surviving site that absorbs the traffic of a failed neighbour can overload quickly.
Separation of roles matters as much. Authoritative and recursive functions should run on different systems, with different access rules. Authoritative servers should answer only for your own zones, and recursive resolvers should answer only for your own networks. Hidden primary servers, which are not reachable from the internet, protect the master copy of your zones from direct attack. Capacity should be planned for realistic failure scenarios, such as losing one location at peak hours, rather than from a single average figure.
4. Comparing the mitigation controls
No single control is enough. Each one addresses a different part of the problem, and each has a cost, usually in complexity, latency or the risk of blocking legitimate users. Read the last two columns carefully: a control that protects the service may leave a gap elsewhere, and responsibility for it may sit with a provider rather than with your team.
DNS Security Extensions (DNSSEC) are related but serve a different purpose. They sign DNS data so that resolvers can detect forged or altered answers. DNSSEC protects integrity rather than availability, so it does not stop a flood of legitimate-looking queries. Enable it for the integrity benefit, and plan for key management and for checks that prevent a broken signature from making a domain unreachable for validating resolvers.
| Control | What it helps with | What it does not solve | Usually operated by |
|---|---|---|---|
| Anycast distribution | Spreading traffic across locations and absorbing volume | Shared upstream links or a single provider that fail together | Your team or a DNS provider with its own network |
| Response rate limiting (RRL) | Reducing repeated amplified answers sent to the same target | Attacks from many sources, or legitimate bursts from shared addresses | Configuration of your authoritative servers |
| Response policy zones (RPZ) | Blocking queries to known malicious domains for your own users | Attacks on your public authoritative servers; feeds need maintenance | Configuration of your recursive resolvers |
| Upstream DDoS scrubbing | Filtering very large volumes before they reach your network | Application-level faults inside your own systems | A network or security provider in front of your links |
5. Rate limiting, filtering and the mistakes behind outages
Rate limiting must be tuned carefully. Response rate limiting (RRL) restricts how many identical answers an authoritative server sends to the same client network in a short period. Set too tight, it slows down legitimate users who share an address with many others, such as an office behind one public address or a mobile carrier network. Start with conservative limits, monitor how many responses are dropped, and adjust after observing normal traffic for a representative period.
Response policy zones (RPZ) let a recursive resolver block or redirect queries to known malicious domains for your own users. They are useful, but a stale or incorrect block list can cut off legitimate sites for your staff, so each feed needs an owner and a review process. The most common mistake is an open recursive resolver that answers queries from anyone on the internet, which is a ready-made amplifier. Other frequent causes of outages are running both roles on one host, having no monitoring of query rates, relying on one site or one upstream link, and never testing failover.
6. How to start: a checklist for your current setup
Before buying anything, check what you already run. The list below is a practical starting point for a review with your IT team or your provider. Write each answer down so the next review can compare them.A setup where nobody can answer several of these questions is not ready for a volumetric attack, whatever the contract says. Bring the answers to your first meeting with a provider, so the discussion starts from facts.
Start with the first item. List every domain and every resolver your organisation uses, then test whether the resolvers answer queries from outside your own networks. That single check often reveals the most urgent exposure in a short time. The remaining answers show which controls above are missing and in what order to add them. Record the result, and repeat the check after every change to a resolver or a zone.
- Which domains are public, and which name servers answer for each of them?
- Can our recursive resolvers be queried from the internet? They should not be.
- Do authoritative and recursive functions run on separate systems?
- Have we tested failover between locations in a planned exercise, and what did we learn?
- Do we have dashboards for query rates, error rates and dropped responses?
- Who receives alerts, and who can change rate limits or block lists during an incident?
- What exactly does the upstream DDoS contract filter, and who decides when it is activated?
Frequently Asked Questions
Is DNS protection the same as a DDoS protection service?
Not exactly. DDoS protection is the broader category and filters traffic across the network and applications. DNS protection focuses on name resolution: anycast distribution, rate limits, query filtering and the hardening of resolvers. Many organisations use both. A DDoS service can absorb a volumetric flood, but it does not decide which resolvers are open or how your zones are hidden. Ask which layer each control covers.
Should we run our own DNS servers or use a managed provider?
It depends on your size, skills and the uptime you need. A specialised managed provider usually brings distributed capacity and operational experience that are hard to build internally. Running your own servers gives control, but it requires skills and capacity in several locations. Whichever you choose, keep a written view of who operates what, how changes are approved and how incidents are reported.
What should a DNS runbook for an active attack contain?
It should name who watches the dashboards, who can enable stricter rate limits, who contacts the upstream provider and how staff are informed. Avoid improvised changes under pressure, because they often break legitimate resolution. Record every change with its time, so that temporary rules can be reviewed and relaxed after the incident.
Does DNSSEC protect us from floods?
No. DNSSEC lets resolvers verify that answers were not forged or altered. It protects integrity against spoofing and tampering, not against large volumes of queries. It also adds operational work, such as key rotation, and a mistake can make a domain unreachable for validating resolvers. Plan it as an integrity project with its own checks and its own owner.
What reporting obligations might apply under NIS2?
For organisations in scope of the NIS2 Directive, as implemented in Italy by D.Lgs. 138/2024, a significant incident is reported in stages. An early warning is due within 24 hours of becoming aware of it, a notification within 72 hours, and a final report later. Whether an incident is significant depends on its impact. Check with your legal or compliance function which obligations apply to you.