The practical problem
By the mid-1960s, research computers were scattered across universities and laboratories, and almost every one was different. They ran incompatible software, stored data in different formats and often sat on separate budgets. Researchers who needed a program or data set on another campus had few choices: travel, mail a tape, or dial in with a terminal. The question facing network builders was a practical one. How could dissimilar machines exchange data over long distances without every site rewriting its software for every other?
Telephone lines and small gateway computers
The answers adopted in the late 1960s relied on a few ingredients. The first was the telephone system. Rather than build new cables, early networks leased lines from telephone companies and ran digital data across them using modems. Speeds were slow by modern standards, so efficiency mattered.
The second ingredient was a small dedicated computer at each site that handled communications. On the ARPANET, these were called Interface Message Processors, built by the firm BBN. Each one connected a local host machine to the network and to its neighbors, receiving data, checking it and passing it along. This arrangement meant that a university needed to teach its computer to talk only to its own interface machine, not to every other host on the network. Later internet routers descended from the same idea, with each device doing a single job well.
The third ingredient was a shared way of cutting data into small units and forwarding them, the packet approach described in the history of packet switching. Because the lines were shared among many conversations, the network could stay busy without reserving a circuit for every user.
Host rules and early services
Joining a network is more than wiring. The attached computers needed agreed rules, called protocols, for starting a conversation, sending data and closing it. The early ARPANET used a host-to-host protocol known as the Network Control Protocol. On top of it, programmers built simple services: remote login so that a user at one site could work on a computer at another, file transfer, and by the early 1970s, electronic mail. The sense that people could use the network as a place to meet came as a surprise to some of its planners, who had expected resource sharing to dominate.
Other networks emerged in parallel
The ARPANET was one of several efforts. In the United Kingdom, the National Physical Laboratory built a local packet network in the late 1960s. In France, the research network CYCLADES, led by Louis Pouzin, tested a design in which the network moved packets and the end computers took responsibility for reliability, a philosophy that later influenced the internet. Commercial and public networks also began to appear in the 1970s, as did computer-company networks, such as those built for IBM and Digital Equipment Corporation customers, which tended to be proprietary.
Local networks also appeared. The most famous, Ethernet, began at Xerox PARC in the early 1970s and is covered in the account of local area networks.
The internet idea: connecting networks
By the early 1970s, the challenge shifted. The ARPANET carried data over leased telephone lines. Other experiments used packet radio, which sent data through the air in the San Francisco Bay area, and packet satellite links across oceans. Each used different speeds, packet sizes and error rates. Linking them seemed hard.
Researchers including Vint Cerf and Bob Kahn worked out a plan for an “internetwork,” in which a common protocol and special gateway computers would join separate networks. Their design was published in 1974. The concept evolved over the next several years into the pair of rules known as TCP and IP, with one breaking the job into reliable delivery between end computers and the other handling addressing and forwarding. A related idea, explored by several groups, was to keep the middle of the network simple and place complexity at the edges.
This is why historians treat the internet less as a single network than as an agreement about how networks join each other. It is a point that matters when reading about the technical rules known as internet protocols, covered in a separate article.
Moving from experiment to infrastructure
The ARPANET switched its hosts to TCP/IP on January 1, 1983, a date often cited as the moment the internet in its modern sense began. Around the same time, the military portion of the network was separated from the research portion. The domain name system, which replaced a single shared list of host names with a distributed directory, followed in the mid-1980s. In the late 1980s, the National Science Foundation built a backbone network for academic use, and regional networks linked to it. Gradually the connected community widened from a few dozen research sites to universities across the country and then to other countries.
What was different about these networks
Compared with today’s internet, the early networks were small, slow and managed by people who largely knew one another. Capacity was limited, and users often had to ask permission or live in an eligible institution. There were no web pages, no search engines and no commercial advertising. Conversation happened through email, mailing lists, and discussion systems such as Usenet, which spread among Unix computers beginning around 1979 and 1980.
Those constraints shaped lasting habits, including openness about technical specifications, the use of informal public documents called Requests for Comments, and a preference for working code over formal approval. The people who shaped these choices are the subject of the separate history of ARPANET.