What counts as a network?
It is tempting to name one machine or one date as the first computer network, but the answer depends on how the term is defined. If a network means any set of computers linked by communication lines, the earliest examples date to the 1950s. If it means a system in which separate computers exchange data as equals, the first working experiments came in the mid-1960s. If it means a packet-switched network that grew into the internet, the answer is generally ARPANET in 1969. This article covers what came before.
Remote terminals: SAGE
The earliest large-scale connections used computers as central hubs, with distant terminals attached. The best-known early example was SAGE, the Semi-Automatic Ground Environment, a United States air defense system built in the 1950s. Radar stations sent tracking data over telephone lines to large computers, which helped operators follow aircraft. To send digital data over telephone lines, the system needed modems, devices that convert digital signals into audio tones and back. Bell Laboratories developed modems for this purpose in the late 1950s, and they later became ordinary commercial products.
Reservations: SABRE
A civilian parallel was SABRE, built by IBM for American Airlines and in operation from about 1960. Reservation agents at terminals around the country could check and book seats in a central computer. It showed that online systems could handle commercial work reliably, and similar designs appeared for banks and other companies. These were not networks of equals, since every terminal depended on one central machine, but they demonstrated that real-time data communication over telephone lines was practical.
Time-sharing
At universities, time-sharing changed how people thought about computers. Instead of submitting a deck of punched cards and waiting for printed results, several users could connect to one computer through terminals and work interactively. MIT’s Compatible Time-Sharing System ran in 1961 and the Dartmouth Time Sharing System followed in 1964. Soon, terminals were being used from other buildings and cities over leased or dial-up lines. A community of users formed around each machine, sharing programs and sending messages to one another. The early mail programs on time-sharing systems were a forerunner of email.
Linking computers together
J.C.R. Licklider, who had worked on time-sharing, wrote in 1962 about a future in which computers around the country would be connected so that researchers could draw on programs and data from many places. As head of computing research at ARPA, he funded work that moved toward that goal. His successors, including Ivan Sutherland, who later built a head-mounted display known as the Sword of Damocles, and Robert Taylor, continued the effort.
A first experiment came in 1965. Lawrence Roberts and Thomas Marill linked a TX-2 computer at MIT’s Lincoln Laboratory in Massachusetts to a Q-32 computer at the System Development Corporation in Santa Monica, California, using a telephone line. They showed that two quite different machines could exchange data and run programs remotely. The experiment also showed a drawback: a dial-up telephone connection was slow and expensive, and kept occupied even when no data moved. The result helped persuade Roberts, who joined ARPA soon afterward, that something better than circuit switching was needed.
Ideas on the way
Several other ideas were developing at this time. Paul Baran in the United States and Donald Davies in Britain worked out how to send data in packets. Davies’s group at the National Physical Laboratory in England built a local network in the late 1960s, and it operated for years, serving laboratory staff, which means it has a claim to be among the first working packet networks. Because the system was small and local, it is less well known than ARPANET.
Others in France, such as the Cyclades project of Louis Pouzin, followed in the 1970s. Each group contributed pieces: ways to divide messages, to route them and to deal with errors.
Why there is no single first
Putting these threads together, there is no single first network. SAGE and SABRE linked computers and terminals over long distances. The 1965 experiment tied two computers together as peers. The National Physical Laboratory network was an early working packet system. ARPANET combined these ideas at a larger scale and was the first to be widely documented and to expand. Anyone who claims a single first is selecting one definition among several.
Why it mattered
The pre-ARPANET years set up the problems the later network had to solve: how to share expensive computers, how to send data over lines built for voice, and how to make unlike machines understand each other. These problems appeared at the same time as the first computers were becoming more widely available in universities and business. Solutions found in this period, from modems to terminal protocols, remained in use well after packet networks arrived.