ARPA was created by the Eisenhower administration in 1958 in reaction to the Soviet Union’s orbiting of the Sputnik satellitesperceived in the United States as a clear sign that the Soviets were outpacing the United States in science and technology. The motivation was to fund and manage research and development projectsprimarily through universitiesin hopes of avoiding another such national embarrassment. Famously wary of the military-industrial complex and the sometimes-intense competition among the branches of the military, Eisenhower made ARPA an independent agency under a civilian director with liberal funding and with wide leeway in the projects they undertook.

Much of the early research at ARPA dealt with satellite and missile technology, and such related technologies as command and control systems. Obviously, computers were a central part of the research both as tools and as objects of research. And because much of the research funded by ARPA was performed at universities around the United States, the computers involved were also scattered around the country. Bob Taylor’s own office was connected to three computers, at MIT, the University of California at Berkley, and System Development Corporation (SDC) in Santa Monica, California. Each computer required a separate terminal in Taylor’s office and separate login procedures. Taylor, inspired by Licklider’s ideas, wondered why there could not be a single terminal connecting him to all three computers. More important, Taylor was seeing a growing inefficiency as work was duplicated among the researchers using computers around the country. If UCLA developed a program that researchers at MIT found useful, for instance, MIT would have to write a version of it for their own, different computer. Wouldn’t it be better if the researchers at MIT could use the UCLA program right on UCLA’s computer? Further, if researchers could access remote computers the need for an expensive computer of their own could be reduced. Licklider’s ideas about time sharing were already being implemented; it was time for his ideas about networking communities of interest to be implemented.

So, in 1966, Bob Taylor proposed a project to develop such a network, and received funding. The ARPA NetworkARPANET, as it was eventually calledwas begun.

Taylor’s first step was to hire a manager and principal architect for the network. For this he chose Larry Roberts, a respected young computer scientist at MIT. In Taylor’s view, Roberts was the only candidate for the position: In addition to his computer science background, he had excellent management skills and had already run on a small networking project connecting MIT’s Lincoln Labs computer to the SDC computer in Santa Monica.

Among Roberts’s roles was bringing together key individuals and ideas for ARPANET implementation. Three people in particular developed the fundamental architectural concepts of the ARPANET: Leonard Kleinrock, Paul Baran, and Donald Davies. In the early 1960s, these three had developed similar ideas without knowledge of each other’s work.

Len Kleinrock, a close friend and gambling buddy of Roberts from MIT, had done his doctoral thesis, “Information Flow in Large Communication Nets,” in 1962. The focus of the research, later published in a book,was on queuing theory in store-and-forward networks. This research formed the foundation of packet switching. Eventually, Kleinrock would work directly on the ARPANET project, developing performance analysis methods for it.

Paul Baran joined the RAND Corporation in 1959 and spent the next five years researching network survivability.His focus was not on computer networks per se, but on the command and control networks for ballistic missile systems and how to ensure that if some nodes were destroyed in a nuclear attack, the other nodes would continue functioning. Baran understood that human neural networks are tremendously robust: If a part of the brain is damaged, the neural net can build new pathways to bypass the damaged cells. So once again, networking ideas sprung from the study of human cognition. A centralized communication network, such as shown in Figure 1.1, can be easily destroyed by eliminating the central node. A decentralized network, as shown in Figure 1.1such as a typical telephone systemprovides a bit more redundancy but not enough. Baran proposed a distributed network as shown in Figure 1.1in which there is no centralized switch and which therefore can build paths around any nodes destroyed by enemy attack.

Figure 1.1. Paul Baran introduced the concept of a distributed network that was far more resilient than centralized or decentralized networks.

Baran further proposed that the messages sent across the decentralized network should themselves be broken into segments, which he called “message blocks.” Because data communication is inherently bursty, Baran theorized that message blocks allowed more efficient use of the available bandwidth between nodes by allowing the interleaving of message blocks from multiple sources. The nodes themselves would be store-and-forward switches that determined the best route to the destination for each message block and forwarded it quicklyBaran called it “hot-potato routing”and if a node had been destroyed, the message block could be routed around the destruction. Baran’s ideas gave rise to dynamic routing.


Larry Roberts learned of Paul Baran’s work in 1967 and met with him in 1968. Afterward, Baran became informally involved in the ARPANET in an advisory capacity.

At the same time that Roberts learned of Baran’s work, he also learned of the work being done by Donald Davies, a physicist at the British National Physical Laboratory (NPL) in London. Without knowing of Baran’s work, Davies proposed a similar concept of routing segmented messages dynamically, although his proposals did not have the level of redundancy of Baran’s distributed network. However, Davies was interested not in military command and control systems but in a new form of public communication. He saw the routing of message segments through the network as analogous to the postal service routing small packages through its system; for this reason, he called the message segments not message blocks, as had Baran, but “packets.” And the routing of the packets through the network he called “packet switching.”

Another significant contribution from Davies came when he realized that different machines were likely to speak different computer languages, and therefore have difficulty communicating directly. Davies proposed using smaller, dedicated “interface computers,” speaking a common language across the network, between the host systems and the network.

When Larry Roberts proposed the ARPANET concept to the researchers at the various host sites, the concept was well received but not the practical implementation. Computer resources were always at a premium, and few wanted to have a part of their time-sharing computers’ resources used for the routing and processing of packets. Wesley Clark, an engineer at Washington University in St. Louis who, in the mid-1950s while at MIT had given J. C. R. Licklider his first serious exposure to computers, suggested placing small computers between the hosts and the network. These small computers would perform the dynamic routing. At the time, Clark did not know of Donald Davies’s almost identical idea.

Roberts adopted this suggestion and called the small computers “Interface Message Processors” (IMPs). These were the precursors to modern routers.

Roberts issued a Request For Proposal for building the IMPs, and the contract was awarded to BBN in Cambridge, Massachusetts, where Licklider had earlier served as a vice president. Licklider had been responsible for purchasing the first computer for BBN, and as a result the company evolved from being just an acoustics engineering consultancy to being a leading computer research firm.

Frank Heart, BBN’s director of the IMP project, selected a “hardened” version of the Honeywell DDP-516 minicomputeron which to build the IMP. Although the 516 was built to military specifications to withstand battlefield and naval deployment conditions, it was not enemy actions that Heart feared. Rather, it was the unwanted attention of graduate students. In the spirit of a highly reliable network, the IMP developers counted on the hardened 516 to be sufficiently resistant to tinkering.

By October 1969, the first two IMPs were installed at UCLA and at Stanford Research Institute (SRI), connecting the mainframe hosts at those sites over a 50kbps link, and the first packets were exchanged. By early December, two more host sites at the University of California Santa Barbara and the University of Utah were added to the network. Early the following year, a cross-country 50kbps link connecting BBN in Cambridge to UCLA was added. By April 1971, there were 15 sites

  1. UCLA
  2. SRI
  3. UCSB
  4. University of Utah
  5. BBN
  6. MIT
  7. RAND Corporation
  8. System Development Corporation
  9. Harvard University
  10. MIT Lincoln Labs
  11. Stanford University
  12. University of Illinois at Urbana
  13. Case Western Reserve University
  14. Carnegie Mellon University

  15. NASA Ames Research Center

The ARPANET was not only up and running, it was growing steadily.

The first public demonstration of ARPANET was held at the International Conference on Computer Communication (ICCC) at the Washington Hilton Hotel in October 1972. Robert Kahn, one of the BBN IMP researchers who had been instrumental in developing the IMP-to-host protocol, architecting the ARPANET, and improving its reliability, had spent more than a year organizing the event. Approximately 40 terminals of different makes and models were connected to a Terminal Interface Processor, or TIP, which in turn was connected to the ARPANET over two 50kbps lines. Conference attendees were invited to come in and play with applications of all sorts, running on computers all over the country. The ICCC event was a huge success, demonstrating to the computer and telecommunications industry that packet switching networks are viable and convincing many that the industry was about to change significantly.


ARPANET and the Bomb

One of the most enduring myths about the origins of the ARPANET is that its was designed to withstand a nuclear attack. The myth is undoubtedly rooted in Paul Baran’s work at the RAND Corporation. But nuclear attack was the furthest thing from the minds of Larry Roberts and his scientists. Baran’s network survivability ideas influenced the construction of the ARPANET for the much more immediate reason that switching nodes and the links connecting them were at the time notoriously unreliable. (According to a famous story, when Charley Kline sent the first packets on the ARPANET in 1969, from UCLA to Stanford Research Institute, he got as far as the g in “login” before the system crashed.) Redundant packet switches and links, and a dynamic routing protocol, were for routing packets around failed links and nodes, not radioactive craters.


Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *


Please type the text above: