You type a shop’s web address and press Enter. You know the name, but your computer needs to find the network address of the service before it can connect. The system that supplies this naming information is the Domain Name System, usually called DNS. It does not contain the shop’s products or run its checkout. It helps your device find where to ask for them. 1

A browser is the program you use to display pages. A server is a computer system running a service that answers requests. In this case there are two different kinds of request. First, the device needs directions to a name. Later, the browser needs the actual page from the web service. A server answering the first question does not thereby supply the second answer. We will use shop.example.com to follow an ordinary lookup for which no usable answer is already saved. 1

Think of a large address directory maintained in separate sections. No single office has to maintain every entry in the world. Instead, one part of the system can tell a helper which other part to ask. The helper follows those directions until it gets the information needed for your request. This division of work is what makes DNS more than a list of website addresses. 1

The helper and the source have different jobs

Your device normally asks a lookup service to find the answer for it. If that service does not have a usable answer saved, it asks other DNS systems. This helper is called a resolver. It does the searching so that you do not have to. 1

The source information comes from the name servers appointed to publish it. A name server is a computer system that answers DNS questions. These source servers are called authoritative: they publish the information for their assigned names, rather than merely passing on a saved answer. The word does not mean they have certified the business. 1

Follow the questions in order

The device asks its lookup helper, in effect: “What network address can I use for shop.example.com?” The helper is normally already selected in the device’s or browser’s network arrangements. The user does not have to locate the shop’s DNS operator first. Nor is the helper necessarily run by the shop. It may be a service used by the office network or another configured provider. 1

If the helper has no useful saved directions, it can start at the top of the naming system, called the root. It asks a root server about shop.example.com. The root server does not have every shop’s web address. It points the helper to the servers responsible for .com: “Ask one of these servers next.”

The helper then asks a .com server about shop.example.com. That server can point it to the name servers responsible for example.com: “Ask these servers about this domain.” Neither reply supplies the shop’s page or its final network address. 1

The helper then asks a name server appointed for example.com about shop.example.com. For our simple example, that server publishes the requested network address. The helper returns the address to the device. The root, the .com service and the domain’s own name servers did different jobs: they narrowed down who should answer, then supplied the information at the source. The helper may remember these directions, so later lookups need not repeat the entire journey. 1

Now the browser can contact the web service at the returned address and request a page. The web service supplies the page content, such as products and opening hours. If its application is broken, a perfectly successful DNS lookup will not repair it. If the device cannot find an address, on the other hand, the browser may never get as far as asking that web service. This order is why “the website will not open” is a symptom, not yet an explanation of which part failed. 1

Why yesterday’s answer may still be used

The difference is easier to see when something changes. The shop’s administrator updates the published destination. The resolver your office uses may still have a saved copy of an earlier answer and use it again. Updating the source does not replace every saved copy at once.

The temporary store of saved answers is called a cache. Each published piece of DNS information is called a record. A record carries a time-to-live value, or TTL: it tells systems how long they may ordinarily reuse that information before refreshing it. The saved answer can therefore remain usable even after the administrator changes the source. 1 2

This is one reason two people can briefly reach different destinations after a move. It is not, however, an explanation for every outage. Waiting cannot correct a destination that was entered wrongly at the source. Nor does DNS promise that every migration will finish after one universal number of hours. 1

Where NS records fit

We now have a second question: how does the lookup service discover which name servers to ask? DNS publishes directions for that too. A record is simply a named piece of DNS information. An NS record, short for “name server,” identifies a server designated to answer for the relevant part of the directory. It does not identify the machine that displays the website. 1 2

For example, a shop can use one company’s name servers while another company runs the web pages. Seeing the DNS company’s name in an NS answer tells you something about the naming service, not where the shop keeps its orders. Changing NS records changes where DNS information is obtained; it is not a substitute for copying the website or its data. 1 2

Suppose the business changes its appointed name servers. The new DNS provider needs the information that the website, email and other services still require. If the new servers publish only the web destination and omit the mail instructions, moving the pages successfully does not make that omission harmless. The helper now asks the new source, and the new source must be able to answer the relevant questions. Changing who answers is different from changing one answer. 1 2

The company should therefore ask its administrator to prepare the information on the new name servers before directing DNS questions to them. The administrator should also check the services that depend on that information. This is a recommendation about planning a move, not an instruction to change NS records whenever a website changes. A website can move while the same name servers continue to publish updated directions. 1

Reading a DNS result without overreading it

Several name servers allow DNS information to be offered from more than one server. But two different names are not proof of two independent companies, networks or failure protections. The NS list alone cannot show how the operator has arranged those dependencies. Likewise, a correct DNS answer does not show that the checkout behind the resulting address works. 1 2

An unsuccessful lookup also needs a little care. “The name does not exist,” “the name has no record of the requested kind,” and “the request failed” are different outcomes. If a phone call is unanswered, you have not established that the person has no address; the same caution helps when interpreting a failed DNS request. DNS distinguishes missing names from missing kinds of information. 14

When moving your website, the useful starting point is simply to separate the jobs: who runs the website, who publishes its DNS information and who can change the appointed name servers? With that distinction clear, “the domain is with another supplier” becomes something you can unpack rather than a dead end. You can leave the technical editing to a specialist and still understand which part of the move they are handling.

Sources

1 — RFC 1034 — Domain names: concepts and facilities

2 — RFC 1035 — Domain names: implementation and specification

14 — RFC 2308 — Negative Caching of DNS Queries