An idea rather than a machine
Most histories tell the story of the internet in dates. Another way to understand it is by one design decision: instead of building a single large network, its designers chose to link existing networks together. The word internet is a shortening of internetwork, and it describes this arrangement exactly. It is a network of networks, in which each part is run by a different organization and connected to the others by agreement.
Why not just build one network?
By the early 1970s, researchers had built several packet networks for different purposes. One used leased telephone lines, another radio, another satellites. Replacing them all with a single standard was unrealistic: each had been built for a particular medium and funded by a particular group. The more practical question was how to make them cooperate without changing them much.
The answer, developed by Vinton Cerf and Robert Kahn in the mid-1970s and refined over the next several years, was to define a thin layer of common rules that sat above all of the networks. Each network could keep its own internal technology. It only had to be able to carry a packet in the shared format. Machines called gateways, now usually called routers, passed packets from one network to the next.
Layering
Engineers describe the arrangement as layers. At the bottom are the physical links and local network technologies, such as Ethernet cables, Wi-Fi radio or optical fiber. Above them sits the Internet Protocol, which supplies addresses and moves packets one hop at a time. Above that are protocols such as TCP, which makes delivery reliable for the ends, and then the applications people use: email, the web, file transfer, video calls.
Each layer needs to know little about the others. A web browser does not care whether a packet traveled over cable or radio. A new link technology can be added at the bottom without changing the web at the top. This is sometimes drawn as an hourglass, with a narrow waist at the Internet Protocol and wide, varied technology above and below. Because everyone agrees on the waist, the layers on either side can change independently.
Simple in the middle
A related principle is often called the end-to-end argument, set out in a well-known paper by Jerome Saltzer, David Reed and David Clark, first presented in 1981. Put simply, functions such as checking that data arrived intact are best done by the computers at the ends of a connection, since only they can know whether the job was done properly. The network in the middle is kept as plain as possible and does best-effort delivery.
This has consequences. New applications can be invented at the edges without asking permission of the network’s operators. That is how the web, streaming video and many other services could appear later without redesigning the core. It also means the network does not by itself know or control what its users do, a fact with both benefits and drawbacks.
Rules written in the open
The protocols were documented in a series of public notes called Requests for Comments (RFCs), a practice that began with Steve Crocker in 1969 on the ARPANET project. Anyone could read them, and from the mid-1980s, the Internet Engineering Task Force provided an open forum where engineers could discuss and agree on technical standards. Participation did not depend on being a member of a large organization. The approach favored working implementations over long formal processes, and it helped to spread the protocols widely.
Joining without permission
To connect to the internet, a network needs an address range and an arrangement with at least one network already connected. It then exchanges routing information with its neighbors using a routing protocol; the one used between large networks is the Border Gateway Protocol, which was developed in the late 1980s. No central authority has to approve the connection technically. Some coordination is needed, because addresses and domain names have to be unique, and organizations handle that role: ICANN, formed in 1998, oversees the domain name system and address allocation.
The result is that the internet today consists of tens of thousands of independent networks, run by companies, universities, governments and others. Reports differ on the exact number, which changes constantly.
Strengths and costs of the design
The design explains the internet’s growth. It was cheap to join, tolerant of different technologies and open to new uses. It also explains weaknesses. Because the middle is simple, features such as security, identity and quality guarantees had to be added later at the ends, and it took years to do so. Early protocols assumed that users could be trusted, and that assumption has been a source of problems with spam, forged addresses and attacks, which led to the development of network firewalls and more secure protocols.
None of this was inevitable. Other designs, which put more control inside the network, were backed by telephone companies and standards bodies and were serious competitors. The internet design won out partly on its merits and partly because it was available, free to use and already working.