The internet is a network of networks. Each large network that joins it is a set of routers run under one technical administration, and these networks choose to connect to one another. No single network is the whole internet, and no network is in charge of all the others.

A resolver is a program that asks name servers for the answer to a client's question about a domain name. A root name server sits at the top of the DNS hierarchy. It does not hold every website's address, but it answers for the root zone and refers the question to the name server for the right top-level domain, such as the one for .com. A top-level domain name server then directs the question to the authoritative name server for the domain being asked about, and that server holds the definitive records, so it can give the final answer that the resolver has been looking for.

To save time, a recursive resolver remembers recent DNS answers for a set period called the time-to-live, or TTL. If someone asks the same question again before time runs out, the resolver can answer from memory instead of having to ask the root, the top-level domain and the authoritative servers again. Name servers hold information about parts of the domain tree and point to other name servers, so no one server needs every answer.

On a secure web page the steps come in a set order: first a DNS lookup, then the TCP handshake, then the TLS handshake. TLS needs a reliable, in-order connection underneath it, which is why it starts only after TCP has set one up. These early steps are part of why the first request to a site takes longer than later ones.

TLS protects a connection in three ways. Encryption keeps outsiders from reading the data. Integrity means nobody can change it in secret. Authentication lets each side prove who it is. On the web, usually only the server proves who it is to the browser.

An IPv4 address is 32 bits long, which allows about 4 billion addresses. On 3 February 2011 the central pool handed out its last blocks to the five regional registries. IPv6 was made to fix the shortage: it raises the address size from 32 bits to 128 bits.

A router can keep one forwarding instruction for a whole group of addresses that start the same way, instead of remembering every computer. One big network of switches cannot grow to billions of computers, because each switch would have to remember every device's address. So the internet splits computers into separate local networks joined by routers.

A content delivery network, or CDN, is a group of servers spread out over many places. Using a content delivery network has a price. A website now depends on another service, so if the CDN goes down or is blocked in a region, the website can stop working. Sometimes a CDN even makes pages slower, because the browser must make extra DNS lookups and connections to reach it.

The Internet Engineering Task Force, founded in 1986, is the main group that writes standards for the internet. Nobody has to join anything to help write them. The IETF has no membership: anyone can take part through a working group mailing list or a meeting, and participants are volunteers who act as individuals. It is up to network operators and software writers whether to use a newly published standard. A few parts of the internet need one shared list so every computer agrees. The Internet Assigned Numbers Authority keeps those registries openly published and provides its services without charge.

The ARPANET, the internet's direct forerunner, began with four computers in late 1969. In 1983 the network was split into a public ARPANET and a military MILNET, and only 45 hosts stayed on the ARPANET. The word "packet" was adopted for the ARPANET from the work of Donald Davies.

In September 1990, Tim Berners-Lee's boss at CERN agreed to buy a NeXT computer so he could write a global hypertext system. By the end of 1990 he had the first web server and browser running at CERN. In 1991 the web spread step by step. On 12 December 1991 Paul Kunz installed the first web server outside Europe, at the Stanford Linear Accelerator Center in California.

To repair an undersea cable, a repair ship drags a grapnel, a hook-like tool, across the seabed until it catches the cable, then pulls it up on board. The damaged part is cut out, a spare piece is spliced in, and the repaired cable is lowered back to the sea floor. As of November 2026, almost all the data that travels between countries goes through cables laid on the sea floor.

Words to know

• resolver: DNS asker
• authoritative: giving final answers
• encryption: scrambling data
• grapnel: hook-like tool
• registries: public lists