Skip to content
GDPR compliant. Hosted in the EU.
14-day free trial →
RFC 2182Name serversDNS resilience
Monitoring 17 min read 19 Nov 2026

Name server redundancy: when all servers share one network

Two name servers in one network fail together. What RFC 2182, IANA and DENIC require, and how to check the network and operator of each name server yourself.

X

XICTRAQ®

XICTRAQ®

DNS-MonitoringNameserverResilience
2
topologically separate networks required by IANA for a TLD's name servers (IANA)
1
differing IP address satisfies the .de pre-delegation check; a second network is not required (DENIC)
3
name servers recommended by RFC 2182 for most organization-level zones (RFC 2182)
Name server redundancy: one network or separate networksTwo NS records only protect you if a single failure cannot take out both serversBefore: one networkAfter: separate networksexample.comexample.comns1.example.com192.0.2.10ns2.example.com192.0.2.11AS 64500 · one data center, one operatorone uplinkIf the network or uplink fails, both servers are gonens1.example.com192.0.2.10ns2.example.net198.51.100.20AS 64501 · Operator AAS 64502 · Operator Bown uplinkown uplinkIf one network fails, the other server still answersScan view: name servers with IP address and operatorfirst IPv4 address per name serverName serverIP addressOperatorns1.example.com192.0.2.10Operator Ans2.example.net198.51.100.20Operator BSource: RFC 2182, section 3.1 · Addresses and AS numbers from the documentation ranges in RFC 5737 and RFC 5398
A domain needs at least two name servers, because the DNS protocol and the registries require it. Whether those servers survive a single failure, however, depends not on their number but on where they sit in the network: if both run in the same data center, on the same uplink or with the same operator, name resolution disappears entirely after one incident, and the shop, email and order confirmations go with it. This post explains what the rules on name server redundancy require and check, how to spot a distribution that only looks redundant, and how to determine the network and operator of each name server yourself in four steps.

Two name servers are not yet two paths

The basic rule is old and unambiguous: every zone must be available on at least two servers so that it stays reachable when a host or a link fails (RFC 1034). The idea behind it is redundancy. If one server goes down, the other answers the queries, and visitors' resolvers switch to it on their own. RFC 2182, published as Best Current Practice 16, spells out what matters: servers should be placed so that, for any likely failure, at least one of them is likely to remain reachable from all significant parts of the internet (RFC 2182).

This is exactly where many configurations fall short. Two NS records with different names and addresses look redundant. If both addresses sit in the same data center, behind the same router or in the same operator's network, the duplication protects against only one failure mode: the loss of exactly one server. It does not help against a power outage in the data center, a broken uplink, a routing error at the provider or an overload of its network. RFC 2182 puts it plainly: the servers of a zone should certainly not all be placed on the same LAN segment in the same room of the same building, because such a configuration almost defeats the purpose of having multiple servers (RFC 2182).

Secondary servers must be placed at both topologically and geographically dispersed locations on the Internet, to minimise the likelihood of a single failure disabling all of them.

— RFC 2182 (BCP 16), section 3.1 “Selecting Secondary Servers”

Topologically dispersed means connected over different network paths, so that the failure of one link or of routing at one provider does not make every server unreachable. Geographically dispersed means far enough apart that an event such as a power loss is unlikely to hit all of them at once (RFC 2182). Both are a question of infrastructure, not of the number of records.

What the rules require and what they check

IANA is the strictest, but only for delegations in the zones it manages itself: the root zone plus .INT and .ARPA. There, the name servers must be in at least two topologically separate networks (IANA). A network is defined as an origin AS in the BGP routing table, that is, the autonomous system that announces an address (IANA). Two addresses from different ranges announced by the same autonomous system therefore count as one network. IANA also requires at least two NS records that do not resolve to the same IP address (IANA).

For a .de domain, the DENIC pre-delegation check applies. It requires at least two name servers, at least one of which must be reachable over IPv4 (DENIC). For distribution, it is enough that at least one name server has an IP address that differs from the addresses of all the others (DENIC). DENIC explicitly justifies the rule with redundancy and states that it places particular weight on diversity in the network topology (DENIC); its specification, however, describes no check of the network operator or the autonomous system. Two name servers with neighboring addresses in the same data center pass the check.

The recommendations go further. NIST advises running at least two of an organization's authoritative name servers on different network segments (NIST SP 800-81r3) and dispersing them across different sites as well. In its IT-Grundschutz baseline, the German Federal Office for Information Security requires at least one additional secondary for every advertising DNS server (BSI) and recommends, for elevated protection needs, connecting externally reachable DNS servers through different providers (BSI). Even the older memo on common DNS errors advises briefly to run secondaries outside your own network (RFC 1912).

Rule setScopeRequirement on numberRequirement on distribution
RFC 1034every DNS zoneat least two serversnone stated explicitly
RFC 2182 (BCP 16)operational guidance for every zonethree for most organization-level zonestopologically and geographically dispersed
IANAroot zone, .INT, .ARPAat least two NS records with different IP addressesat least two origin autonomous systems
DENICdelegation of a .de domainat least two name servers, one of them over IPv4at least one differing IP address
NIST SP 800-81r3guidance for organizationsone primary and at least one secondarytwo on different network segments, plus different sites
BSI IT-Grundschutz, APP.3.6organization's information domainone additional secondary per advertising serverdifferent providers for elevated protection needs

The table shows the core of the problem: the checks an ordinary domain actually has to pass at registration measure the number and the IP addresses of its name servers. Whether the servers hang off separate networks is checked by none of the delegation checks for a .de domain. The question stays with the domain owner.

How to spot a distribution that only looks redundant

Concentration is rarely intentional; usually it is the default of a hosting package or a leftover from a migration. Common patterns:

  • Neighboring addresses: ns1 and ns2 use addresses from the same /24 range, often consecutive ones. Such servers frequently sit in the same data center.
  • Different names, one operator: ns1.dns-provider.example and ns2.provider-net.example sound like two services but belong to the same provider and are announced from the same autonomous system.
  • Name servers under your own domain: ns1.example.com and ns2.example.com point to the same server as the website. If that server fails, the website and name resolution are gone at the same time.
  • Primary and secondaries on one platform: the hidden primary and all visible secondaries run with the same provider. A misconfiguration, a suspended customer account or a platform incident hits all of them together.
  • Stale entries after a migration: a name server is still listed in the delegation but no longer serves the zone. RFC 1912 calls this a lame delegation; in the mildest case it causes extra traffic, in the worst unresolved hosts and bounced email (RFC 1912).
  • One provider for everything: registrar, DNS, web hosting and mail all sit with the same provider, and a single incident there hits all four functions.

All of these patterns pass the delegation check and share the same weakness: a single event outside your control shuts name resolution down completely. Similar dependencies on third-party infrastructure are covered in the post on cloud outages and third-party dependencies.

What happens when that one network fails

If the web server is down, the browser gets no connection. If name resolution is down, it does not find the address in the first place, and that affects every service under the domain. RFC 2182 points out that clients react differently to an inability to resolve than to an inability to connect, and not in every case as desirably (RFC 2182). For a shop, this means:

  • Shop and website: visitors see a browser error instead of a maintenance page, because without an address the request does not reach any server.
  • Incoming email: sending mail servers cannot find the MX record. They usually queue the message and retry later; if the disruption lasts long, messages bounce back to the sender.
  • Order confirmations: receiving servers check SPF, DKIM and DMARC through the DNS of the sending domain. If it is unreachable, the check cannot complete, and confirmations arrive late, in spam or not at all, depending on the recipient. How a missing email gets noticed is shown in the post on the double opt-in flow.
  • Interfaces: payment providers, marketplaces and inventory systems call webhooks under your domain. If resolution fails, callbacks go nowhere.

The only buffer is resolver caches. Records a resolver looked up shortly before stay valid for the duration of their TTL. RFC 9199 notes that the effects of a DDoS attack on the DNS can be greatly reduced if caches last longer than the attack (RFC 9199). That argues for TTL values that are not too short in normal operation, but it does not replace distribution: an outage that lasts longer than the TTL becomes visible to everyone whose cache has expired.

How large such events can get is shown by a figure from the same source: the most powerful DDoS attack against DNS servers recorded up to that point reached 1.2 Tbps and used Internet of Things devices; the underlying report dates from October 2016 (RFC 9199). An attack of that size hits an operator's network as a whole, not a single server within it. How such attacks are amplified through open resolvers is described in the post on DNS amplification and open resolvers.

The risk is not spread evenly over time. During Black Week, with Black Friday on November 27 and Cyber Monday on November 30, 2026, campaigns, newsletters and ads run at full speed, and every click depends on working name resolution. If it fails, the ad budget keeps paying for clicks that do not reach the shop. Why ad spend and revenue deserve a careful comparison that week anyway is explained in the post on ROAS during Black Week.

Checking in four steps: from NS record to network operator

Whether your own name servers are separated can be answered with built-in tools in a few minutes. The steps use the dig tool, the examples the documentation domain example.com and addresses from the ranges reserved for documentation.

  1. Query the NS records in both places: in the zone itself and in the parent zone that holds the delegation. Both lists should match (RFC 1912); if the delegation differs, resolvers may query servers that no longer know the zone.
  2. Resolve the addresses of every name server: A and AAAA records separately. A server with an IPv4 and an IPv6 address may be attached to different networks over the two protocols.
  3. Determine network and operator per address: for every address, identify the autonomous system that announces it. The same AS number means the same network in the sense of the IANA definition. The free ASN lookup maps an IP address to provider, AS number, country and IP range.
  4. Query each server on its own: directly and without recursion for the SOA record, once over UDP and once over TCP. All authoritative servers should return the same SOA record (IANA), and for the name servers of top-level domains IANA requires answers over UDP and TCP on port 53 (IANA).
$ dig +short NS example.com
ns1.example.com. ns2.example.net.
$ dig +norec NS example.com @a.gtld-servers.net | grep -A2 'AUTHORITY SECTION'
;; AUTHORITY SECTION: example.com. 172800 IN NS ns1.example.com. example.com. 172800 IN NS ns2.example.net.
$ dig +short A ns1.example.com; dig +short A ns2.example.net
192.0.2.10 198.51.100.20
$ dig +short AAAA ns1.example.com
2001:db8:10::53
$ dig +norec +short SOA example.com @192.0.2.10
ns1.example.com. hostmaster.example.com. 2026111901 7200 900 1209600 3600
$ dig +norec +tcp +short SOA example.com @198.51.100.20
ns1.example.com. hostmaster.example.com. 2026111901 7200 900 1209600 3600

The result of the third step is decisive. If all addresses share the same AS number, the servers are in one network in the IANA sense, even if names and address ranges differ. If they are in different autonomous systems, the most important condition is met; whether the networks are also geographically separate cannot be read from the AS number alone. Identical SOA serial numbers show that every server holds the current version of the zone. The DNS lookup puts the answers of several public resolvers side by side.

The TCP test is part of it

A name server that only answers over UDP rarely stands out in daily operation. Resolvers fall back to TCP, however, as soon as an answer is too large for a UDP packet, for example with signed zones. DENIC checks TCP reachability at delegation time as a should-level requirement and issues a warning otherwise.

Anycast, one provider, two names: when distribution is only geographic

Many DNS providers run their name servers via anycast: the same address is announced from many locations, and routing sends each query to the nearest one. Multiple NS records are strongly recommended for all zones, and according to RFC 9199 anycast is used by the root zone, most top-level domains and many large organizations, among others (RFC 9199). The registry for the .nl ending moved its name servers to four anycast servers in early 2018 (RFC 9199).

Anycast spreads load and makes a service more robust against the loss of individual sites. It does not change the fact that all sites are run by one operator with one configuration and often from one autonomous system. A faulty update, a routing problem at the operator or a suspended customer account then hits all name servers of the zone at once. Whether an address is operated via anycast is shown by the bogon and anycast check. How routing errors cut individual networks off from a website is described in the post on BGP and RPKI.

The thorough solution is name servers with at least two independent operators. NIST gives the example of locating some authoritative servers on the organization's own premises and others in a hosting service or with a partner organization (NIST SP 800-81r3). Two providers mean two places where the zone is maintained. If one takes it over from the other via zone transfer, the transfer belongs on the checklist so that it only goes to the intended servers. How to check that is shown in the post on the open zone transfer.

If the zone is signed with DNSSEC, both providers have to coordinate their keys and signatures. Otherwise validation errors occur, and validating resolvers treat the zone as unresolvable. Background is covered in the post on DNSSEC validation.

No switchover right before peak season

Anyone who moves name servers or adds a second provider changes the delegation at the registrar. Until the old NS records have aged out of all caches, resolvers keep querying the previous servers. RFC 1912 therefore advises keeping the old name servers running for a while so that no lame delegation arises (RFC 1912). Such a switchover belongs in a quiet week with time to verify, not in the days before Black Friday.

How many name servers and where: a sober rule of thumb

More servers are not automatically better. RFC 2182 considers two carefully placed servers often sufficient but advises more than two, because cases where two are not enough occur often enough (RFC 2182). If one server is unavailable for an extended period, a zone with only two listed servers is actually running with just one (RFC 2182). Three servers are recommended for most organization-level zones, with at least one well removed from the others (RFC 2182); for zones with higher demands, four or even five may be desirable (RFC 2182).

There is an upper limit as well. Every additional server increases the likelihood that one of them is misconfigured or malfunctions without being detected (RFC 2182). RFC 1912 names seven as the recommended maximum and notes that more entries mainly make things harder to manage (RFC 1912). For a shop, this results in a sober rule of thumb: three to four name servers, spread across at least two autonomous systems, and across two independent operators for elevated protection needs.

A second, often overlooked value sits in the zone's SOA record: expire. It defines how long a secondary keeps serving its copy of the zone when it can no longer reach the primary. RFC 1912 suggests two to four weeks (RFC 1912). For .de domains, DENIC checks whether the value lies between 604,800 and 3,600,000 seconds, that is, between one week and 1,000 hours, and issues a warning otherwise (DENIC). It rates values below one week as very critical, because a zone can then lose all of its authoritative name servers within a short time (DENIC).

Separate networks only help while synchronization works

If the hidden primary fails unnoticed, the secondaries keep serving the zone only until the expire value runs out, after which it goes down despite distributed servers. Regularly querying the SOA serial number on every server shows whether synchronization still works.

Monitor the state instead of checking it once

A one-time check holds for the day it takes place. Name servers change, however, without anyone explicitly deciding it: a host migrates its platform, a package change resets the defaults. That is why it pays to treat name servers as an ongoing measurement.

DNS and domain monitoring by XICTRAQ® supports this in three places. The scan lists the name servers of a domain with IP address and operator; it uses the first IPv4 address of each name server, and the operator comes from mapping that address to its network operator. A name server change appears in the changelog with old and new values and is flagged as a warning, because it can also indicate a takeover. And in service monitoring, each name server host can be checked individually with a TCP port check on port 53 or a ping; if one fails, XICTRAQ® reports it through the configured notification channels.

Name server overview

Host, IP address and operator for every name server at a glance. If all entries share the same operator, that is the cue for the four-step check.

Changes in the changelog

New or removed name servers appear with old and new values. A change without a matching ticket deserves the same attention as described in the post on the registry lock.

Reachability per server

A TCP port check on port 53 for each name server host reports when one of the servers stops accepting connections, even while the domain still resolves through the others.

What XICTRAQ® does not assess here

The name server overview shows the first IPv4 address of each name server; IPv6 addresses and further IPv4 addresses do not appear there. XICTRAQ® does not automatically assess whether the servers are in separate networks, and a missing distribution does not trigger an alert. Which autonomous systems are behind the addresses is something you determine with the steps above. A TCP check on port 53 also shows that a server accepts connections, not that it answers for the zone correctly and authoritatively.

The website itself is watched by uptime monitoring from several nodes, and the timing waterfall of every check reports the time for DNS resolution separately. If it rises markedly or resolution fails, that is an early signal that something is wrong with the name servers. The overview for your own domain is available in the scan after registration.

Checklist before Black Week

Working through these items before the end of November leaves time to fix a finding calmly before the sale week starts.

  • Query and compare the NS records in the zone and in the delegation of the parent zone.
  • Determine the autonomous system for every IPv4 and IPv6 address of the name servers.
  • Check that at least two different autonomous systems are represented, and two independent operators for elevated protection needs.
  • Check name servers under your own domain for their dependency on the web server.
  • Query and compare the SOA serial number on every server, and compare the expire value against the recommendation.
  • Query each name server individually over UDP and TCP.
  • Plan name server and provider changes outside peak season and verify them afterward.
  • Have reachability per name server and changes to the NS list monitored continuously.

Separate networks belong in operational planning

Name server redundancy is rarely a deliberate decision. It follows from the default of a hosting package, and the registries' checks confirm it as long as two different IP addresses are entered. The recommendation in RFC 2182 still applies unchanged: dispersed by topology and geography, so that a single failure is unlikely to hit all servers at once.

Putting it into practice does not take large infrastructure, but a second, independent operator or at least a second autonomous system, coordinated zone maintenance and a measurement showing that both stay that way. That pays off on the day a single data center, uplink or provider fails.

Sources and standards

This post is based on: RFC 1034 “Domain Names – Concepts and Facilities”, section 4.1; RFC 2182 (BCP 16) “Selection and Operation of Secondary DNS Servers”, sections 3 to 5; RFC 1912 “Common DNS Operational and Configuration Errors”, sections 2.2 and 2.8; RFC 9199 “Considerations for Large Authoritative DNS Server Operators”, sections 3.1, 3.4 and 3.5; IANA, “Technical requirements for authoritative name servers”, last revised November 14, 2024; DENIC, “Nameserver Predelegation Check – Kommentiert” (German), sections 2.1.2, 2.1.4.3 and 2.1.6.3; NIST SP 800-81r3 “Secure Domain Name System (DNS) Deployment Guide”, March 2026, section 2.3.2; BSI (German Federal Office for Information Security), IT-Grundschutz Compendium Edition 2023, module APP.3.6 DNS Servers, requirements on redundant DNS servers and on connection through different providers. Example addresses and AS numbers come from the documentation ranges in RFC 5737, RFC 3849 and RFC 5398.

Are two name servers enough for a domain?
Formally, yes: RFC 1034 and the DENIC pre-delegation check require at least two. Whether they are enough when it matters depends on their distribution. If both are in the same network, they usually only protect against the failure of a single server. RFC 2182 recommends three servers for most organization-level zones, with at least one well removed from the others.
How can I tell whether my name servers are in the same network?
Resolve the addresses of all name servers and determine for each one the autonomous system that announces it. The same AS numbers mean the same network under the IANA definition. Addresses from the same /24 range are a strong hint, but they do not replace checking the AS number.
Are a provider's anycast name servers distributed enough?
Geographically, usually yes; organizationally, no. Anycast spreads one address across many sites, but these are typically run by the same operator with the same configuration. In practice that helps well against the loss of individual sites, but hardly against an incident affecting the whole platform; that takes a second, independent operator.
Does DENIC check whether my name servers are in separate networks?
Not according to its published specification. The pre-delegation check requires at least two name servers and at least one IP address that differs from the others. A check of the network operator or the autonomous system is not described there. IANA applies the stricter rule of at least two separate networks to delegations in the root zone.
What happens to email when all name servers fail?
Sending mail servers cannot find the domain's MX record and usually queue the message to retry later. If the outage lasts longer, it may bounce back to the sender. Outgoing mail can also fail the recipients' SPF or DMARC checks.
Can XICTRAQ® detect that my name servers are not distributed?
XICTRAQ® shows the name servers with IP address and operator in the scan, so identical operators stand out at a glance. There is no automatic assessment of the distribution and no alert for a missing distribution. What does get reported are changes to the name servers in the changelog and, if you set up a service check per name server, the failure of an individual server.

Related in XICTRAQ®

Start in 60 seconds

Let XICTRAQ® watch your website.

Free plan with no fixed term available. No credit card, no installation, no hidden fees.