In the early 1960s every computer was an island. Major research centers owned expensive, mutually incompatible machines, and anyone who needed a program or data set from another institution relied on magnetic tapes sent by mail or on travel. The networks that existed were telephone and telegraph systems, designed for voice and short messages, not for the irregular, bursty traffic of computers: long silences followed by quick exchanges of a lot of data. Renting a dedicated telephone line and paying for it every minute was costly and wasteful for that pattern. A new way of organizing communication was needed.
How it works
The central idea of data networks is packet switching. Instead of reserving an exclusive circuit between two points for a whole conversation, as traditional telephony does (circuit switching), the message is split into small blocks called packets. Each packet carries a header with the destination address and sequence information, and is forwarded from node to node by routers or switching nodes. Packets from different conversations share the same lines and are reassembled in order at the destination. If one is lost, it can be sent again. The result is much better use of cable capacity.
For machines from different makers to understand one another, they need shared rules called protocols. These define data formats, addressing, acknowledgment of receipt and error handling. A network protocol is, in essence, a technical agreement about how to converse, and much of the early history of networking is the history of getting such agreements to work.
Antecedents: time-sharing and a vision of networking
In the 1960s time-sharing, demonstrated in systems such as MIT's CTSS, let many users work on one computer simultaneously from remote terminals. It was a first step toward thinking of the computer as a service reached from a distance. J. C. R. Licklider, a psychologist and computer scientist, headed the computing research office of the Pentagon's Advanced Research Projects Agency (ARPA) from late 1962. In a 1963 memorandum addressed to colleagues, he described an "Intergalactic" computer network, a vision of an interconnected community of machines and people that influenced his successors.
Packet switching: several inventors
Independent research efforts arrived at packet switching. In the United States, Paul Baran of the RAND Corporation studied distributed communication networks in the early 1960s, designed so that no single center would be a vulnerable point, and divided messages into blocks sent along multiple paths. His work was motivated by concerns about the survival of communications during the Cold War. In Britain, Donald Davies of the National Physical Laboratory (NPL) reached a similar idea in 1965 and 1966, aimed at public computing services, and he chose the term "packet." Leonard Kleinrock developed mathematical queuing theory for message networks, which provided a basis for analyzing their performance. Priority among these pioneers is debated, and it is more accurate to see convergent contributions.
ARPANET
In 1966 and 1967 Larry Roberts at ARPA pulled the ideas together and led the design of the ARPANET, a network to link computers at research centers the agency funded. The stated goal was to share expensive resources and programs among those centers. The firm Bolt Beranek and Newman (BBN) built the IMPs (Interface Message Processors), small dedicated computers that served as routers, sitting between each host computer and the network.
On October 29, 1969, the first message passed between a computer at the University of California, Los Angeles (UCLA), in Kleinrock's group, where the student programmer Charley Kline typed it, and one at the Stanford Research Institute (SRI). The attempt to send "LOGIN" failed after the first two letters, and a full login succeeded about an hour later. By the end of 1969 the network had four nodes: UCLA, SRI, the University of California at Santa Barbara and the University of Utah. The initial host-to-host protocol, NCP, was put in place from 1970. In 1971 Ray Tomlinson of BBN wrote a program for electronic mail between machines on the network and chose the @ sign to separate user from computer; email soon became the ARPANET's most popular use.
A widespread version of the story holds that the ARPANET was built to survive a nuclear attack. The reality is more complicated. Baran's work had that motivation, but the ARPA network was conceived for resource sharing among researchers. Participants and historians disagree about how much military concern shaped the design, so the simple explanation should be treated with caution.
Other networks and internetworking
The ARPANET was not alone. At NPL, Davies's team began building a local network in 1968; parts of it were running by 1969, and it was fully operational in early 1970. In France, Louis Pouzin led the Cyclades project, launched in 1971, with design and staffing under way in 1972 and a first demonstration in November 1973. Cyclades favored the datagram model: each packet travels independently, and responsibility for reliability lies at the endpoints rather than inside the network. That idea influenced the design of the internet. At the University of Hawaii, Norman Abramson created ALOHAnet, a radio network whose development began in 1968 and which became operational in June 1971, linking terminals on several Hawaiian islands to a central computer. It demonstrated a simple random-access method for sharing one radio channel. Drawing on ideas of this kind, Robert Metcalfe and colleagues at Xerox PARC described Ethernet in 1973, a local-area cabling technology that became standard in businesses and schools.
Other kinds of networks appeared as well. Usenet, a system of discussion groups linking computers running Unix, was conceived in 1979 by the Duke University graduate students Tom Truscott and Jim Ellis. Steve Bellovin, a graduate student at the University of North Carolina, wrote the first version of the software as shell scripts, and the first links, between Duke and UNC over telephone lines, began operating in 1980. BITNET, a mail-and-file network among academic institutions, began in May 1981 with a link between the City University of New York and Yale, created by Ira Fuchs and Greydon Freeman.
The challenge then became joining different networks together. In 1974 Vint Cerf and Robert Kahn published a paper proposing a protocol to interconnect packet networks, which evolved into TCP/IP: TCP handles dividing and reassembling data reliably, and IP handles addressing and forwarding. On January 1, 1983, the ARPANET officially switched from NCP to TCP/IP on a coordinated transition day. The set of networks joined by this protocol came to be called the internet.
Around the same time the Domain Name System (DNS) replaced manually maintained lists of computer names, a file kept at SRI, with a distributed lookup system. Paul Mockapetris designed it, and his specifications, RFC 882 and RFC 883, were published in November 1983; Jon Postel had helped shape the naming plans beforehand and later managed the registry of names and numbers, so credit is shared. In the United States, the NSFNET, built by the National Science Foundation from the mid-1980s, widened academic access and served as a backbone until the commercial opening of the 1990s.
Impact and limitations
The first networks created the practice of sharing data and software, communicating by email and standardizing protocols through open technical discussion documents known as RFCs. They also established the technical model on which the Web would be built.
Access, however, was restricted. Through the late 1980s, networks served mainly universities, laboratories and government agencies, chiefly in the United States and some European countries, and most people had never heard of them. There was also competition among models: telecommunications companies and international standards bodies backed designs such as X.25 and later the OSI model, while TCP/IP prevailed in part because it worked and was openly implemented. Concerns that seem central today, such as security and privacy, were not at the forefront in a small community of mutual trust and had to be addressed later.
Connections to other technologies
The idea of carrying data over shared infrastructure descends from the telegraph, which also transmitted coded messages as pulses, and the telephone, whose physical network was reused by modems to carry data between computers. The connected machines descend from the first computers, and opening the network to the public led to the development of the Web.