Two different problems
Computer networking developed along two paths that later met. One was connecting machines within a single building or campus. The other was connecting sites that were far apart, which meant relying on telephone companies and other carriers. Engineers use the terms local area network (LAN) and wide area network (WAN) for the two cases, and the technology that grew up around each was quite different.
The local side: Ethernet and its rivals
By the early 1970s, research centers were filling with personal workstations and shared printers, and they needed a cheap way to connect them. At Xerox’s Palo Alto Research Center, Robert Metcalfe and David Boggs developed Ethernet. Metcalfe wrote a memo describing the idea in 1973, and a published paper by Metcalfe and Boggs followed in 1976. It drew on earlier work on ALOHAnet, a radio network in Hawaii in which stations sent data whenever they wished and resent it if a collision occurred.
The original Ethernet shared a single coaxial cable among all machines and ran at 10 megabits per second. A station listened before sending and backed off if another was transmitting. In 1980, Digital Equipment Corporation, Intel and Xerox published a joint specification, and the Institute of Electrical and Electronics Engineers approved a related standard, IEEE 802.3, in the early 1980s. Other designs competed, including IBM’s Token Ring, but Ethernet’s low cost helped it spread. Later versions moved to twisted-pair cable and switches, and speeds rose from 10 to 100 megabits and beyond.
The wide area side: borrowed telephone lines
Long-distance data links began with the telephone network. Early systems used modems to send digital signals as audio tones over dial-up lines. For networks that needed constant connections, organizations leased dedicated lines from telephone carriers. The ARPANET used leased lines at 50 kilobits per second, a speed that seems very small now. Later services included T1 lines at about 1.5 megabits per second and T3 lines at about 45 megabits per second, which were common for universities and businesses in the late 1980s and early 1990s.
Other approaches were also tried. X.25, standardized in the 1970s, offered packet-switched data services run by telephone administrations and was widely used in Europe and for banking links. Frame relay and ATM were later carrier services. Each one was a way to let a customer treat part of a carrier’s network as a private data line.
Joining the pieces
The hardest step was connecting networks that used different rules. A Token Ring network and an Ethernet network could not simply be wired together, and many vendors had their own proprietary systems, such as IBM’s SNA or Digital’s DECnet. A device that forwards data between networks became essential. Routers read the address on each packet of data and pass it toward its destination, one hop at a time. In the early 1980s, researchers at several universities, including Stanford, built multiprotocol routers so campus networks could talk to the ARPANET.
What made global linking possible was a common protocol. TCP/IP, adopted across ARPANET on January 1, 1983, could run over Ethernet, over leased lines, over radio and over almost anything else. A campus could build its own LAN, attach it to a regional network and connect that region to the national backbone, all using the same addressing system.
Regional networks and the backbone
In the United States, the National Science Foundation encouraged this layered structure in the mid-1980s. Universities joined regional networks, and these connected to the NSFNET backbone, which began in 1986 and was upgraded to faster lines by the end of the decade. Europe built comparable academic networks, and international links grew from the 1970s onward, including connections from the ARPANET to University College London and a Norwegian research site in 1973.
As commercial service providers began to operate their own backbones, they needed to exchange traffic with each other. Interconnection points where providers met and swapped traffic appeared in the early 1990s. The global network has no single owner. It is a collection of thousands of independent networks that agree to pass each other’s traffic, a feature that the wide area designs of the 1980s made possible.
Fiber and the physical network
Wide area links took a different form once optical fiber became practical in the late 1970s and 1980s. Fiber carries far more data over long distances than copper telephone wire, and it is now the main medium for long-haul and undersea routes. Earlier generations of undersea cable carried telegraph and telephone signals, and the same routes that once carried submarine cables for messages now carry internet traffic as well.
A pattern of layers
The development shows a repeated pattern. Local networks grew first because they were cheap and under one organization’s control. Wide area links were slower and more expensive because they depended on carriers. Shared protocols joined the two. Later, falling costs and higher speeds narrowed the gap, so that a home connection today may be faster than the backbone of the early internet.