In the Beginning, There Was Arpanet

Jan. 1, 1997

One midsummer day in 1968, computer scientist Severo M. Ornstein was sitting in his office at the small Cambridge, Mass., firm Bolt Beranek and Newman, Inc., when his boss walked in and handed him a sheaf of papers. It was a request from the Defense Department’s Advanced Research Projects Agency (ARPA) for companies to bid on a project that would link together computers at different sites, enabling them to swap data and files. Such a network had never really been built before. Mr. Ornstein’s boss wanted to know if BBN could do it.

Ornstein took the papers home and studied them overnight. A day or two later, he slapped the proposal back down on his superior’s desk.

“I suppose we could build that, if you wanted to,” he said. “But I can’t see what one would want such a thing for.”

Seldom in modern history has a technological prophecy been more wrong, as Ornstein himself has since laughingly admitted. Despite his reservations, BBN bid for and won the contract to construct ARPANET—a rudimentary, four-computer experiment that has since exploded into today’s world-girdling, multipurpose Internet.

Maybe the Internet would have come into existence without ARPA funding. The utility of such a network is obvious, and numerous scientists, including some in other nations, were tinkering with similar technology at the time.

But US military funding gave computer networks a great boost and ensured American leadership in a scientific area that has become crucial to the economy as well as national security. In fact, some of the researchers involved in ARPANET’s development feel that history will judge their efforts as having been as important as the Manhattan Project. Flexible leadership from ARPA (today, the Defense Advanced Research Projects Agency, or DARPA) helped make it a reality.

“Great” and “Heroic”

In a 1990 oral history interview taped by the University of Minnesota’s Charles Babbage Institute, Leonard Kleinrock, a UCLA researcher and ARPANET pioneer, said, “It was one of the great experiments in science, I think. . . . It was a heroic kind of thing.”

However, it was the Air Force—not ARPA—that probably provided the first US military seed money for investigation of computer networks and their possible implications for national defense.

The time was the early 1960s and the place was the Rand Corp. in California. Using general research funds provided by the Air Force, researcher Paul Baran was investigating methods of making strategic military communications more robust.

Specifically, Mr. Baran was concerned about ensuring that national command authorities could order a retaliatory nuclear strike in the event of widespread war. At the time, the nuclear arsenals of both superpowers seemed vulnerable to decapitation by a first strike. The result was a hair-trigger balance, with each side eyeing the other nervously for signs of attack.

If leaders knew that they would be able to hit back under almost any circumstances, Baran reasoned, they’d be less likely to rush into nuclear war. His solution: communication via a distributed network.

“The basic network configuration was simple,” Baran said in his Babbage Institute interview. “Avoid any central node. Build a distributed network of nodes, each connected to its neighbor.”

Such a spiderweb design would enable a message to get through even if a number of strands were broken. It was a relatively obvious solution to the problem. But Baran’s second recommendation was perhaps not so obvious: Break down messages into components.

“Message blocks,” he called them. Computers would chop up communications as they were sent, sending numerous message bits speeding over various paths toward the target. Computers at the receiving end would retrieve the bits and reassemble them in the correct order. The approach would increase the system’s efficiency as well as its reliability.

Eventually, Baran produced 11 volumes explaining the technical details of his proposed system. The vast majority of the information contained in the volumes was unclassified. Baran did not mind if the Soviets picked up the distributed-node idea. After all, if they felt more secure, the nuclear balance might be more stable, and thus the United States would be safer.

Finally, in 1965, Rand formally proposed construction of Baran’s system to the Air Force. Service officials vigorously backed the idea—though the nation’s communications behemoth, AT&T, felt it could not be done.

The problem was that the relatively new Defense Communications Agency was responsible for the long-distance communications of all of the US military services. Baran and some higher-level Defense Department officials did not feel DCA had the technical competence to do the job. The project was killed, at least until a more skilled implementing agency could be found.

Missed Opportunity

The Air Force thus perhaps missed a chance to make computing history, but Baran still feels it was the right decision.

“If the project turned into a botch, it would be extremely difficult to get it going again,” he said. “Detractors would have proof that it couldn’t be done.”

Enter ARPA. In 1966, Robert Taylor, then the head of ARPA’s Information Processing Techniques Office, began actively looking for someone to run a computer network project. In part, Mr. Taylor was looking for ways to allow ARPA-funded researchers to work more efficiently. Linking computers at ARPA centers, he felt, would help accomplish this goal.

Taylor found his man in Larry Roberts, an intense scientist at Massachusetts Institute of Technology’s Lincoln Laboratory, who had the requisite computer background. Mr. Roberts had twice turned down the ARPA project director job before he finally accepted. Many ARPANET pioneers say that, in retrospect, it was Roberts’s energetic management and direction that made the project the success that it was.

Working with others in the field, Roberts by 1968 had pulled together the specifications for the new ARPANET. It would mirror Baran’s idea—a distributed net, carrying message bits for reassembly at their destination. (Somewhere along the way, the message-block aspect of the system came to be called “packet switching,” perhaps to elicit the image of millions of little packages racing to their electronic addresses. That’s the term—and the underlying technology—that’s still in use.)

In January 1969, BBN won the contract to build the first pieces of what would become the Internet. BBN had begun life as an acoustical design firm, consulting on such things as the shape of movie theaters, and was still relatively new in the computer field. Therefore, its ARPANET victory was something of a surprise. But by September, BBN was ready to install the first ARPANET computer. Based on the Honeywell 516, the machine was known as an Interface Message Processor.

The first IMP went to UCLA, a center of ARPA-funded research. By the end of the year, IMPs had been shipped to three other ARPA hotbeds: Stanford Research Institute, the University of California, Santa Barbara, and the University of Utah. ARPANET was born, though initially it was little more than a local area network.

Hidden Agendas

The system was not an overnight success. To much of the US computer science world, the idea of packet switching seemed both revolutionary and dubious. Furthermore, the idea of sharing computer resources was not an attractive one for many academics. ARPA wanted them to use a computer on somebody else’s campus, via phone line; what many of them really wanted was for ARPA to buy them spanking new computers of their own.

“For most of the people at any given site, it was at best neutral and at worse maybe a little antithetical to their own interests and aspirations,” BBN scientist Alexander McKenzie told his Babbage Institute interviewer.

Each of the original ARPANET nodes was intended as a specialized software site. The University of Utah, for instance, was then experimenting with cutting-edge computer graphics. Among ARPANET’s heaviest users in its early years were scientists who had changed jobs and moved from one site to another. They would log on to the system largely to take advantage of the specialized software back at their old universities.

By 1971, ARPANET consisted of 15 nodes. Around then, something began to happen, something not anticipated by the system’s planners. Scientists began to use the system as a means of accessing data banks and swapping information, as opposed to a method of accessing somebody else’s computer programs.

E-mail had not even been one of the Internet’s original services. It was an afterthought, an add-on that did not start up until around 1970. Once the researchers realized the power of computer communications, cooperative software use was quickly forgotten. To a certain extent, that original vision for ARPANET has yet to be realized, note its pioneers.

“Other things—people interactively working together, people being able to communicate with each other more easily, people being able to have joint projects, and just being able to use software they both had developed over the net—became much more important,” said Frank Heart, BBN manager of the ARPANET project for more than 10 years.

In 1972, an ARPANET demonstration for the International Conference on Computer Communications became something of a coming-out party for the system. ARPA moved an IMP into a Washington, D. C., hotel and demonstrated everything from a remote air traffic control system to a tiny scooting robot directed from across the country. A new computer began hooking up to ARPA every 20 days, on average.

“Resistance” Within

Throughout ARPANET’s early days, the influence of the military was somewhat muted. BBN and other contractors working on the system dealt largely with civilian ARPA officials whom they considered peers—not uniformed officers interested in the net’s defense applications. Some of the ARPANET pioneers were even antiwar protestors. Ornstein, for one, used to joke that he was going to take a little “Resistance” button into the Pentagon and pin it on a colonel’s jacket when he wasn’t looking. (Ornstein, one of the Internet’s key hardware designers, later became a founder of the antinuclear group Computer Professionals for Social Responsibility.)

Government bureaucracy wasn’t much of a problem. ARPA was small and flexible, with project director Larry Roberts allowed considerable power to make key technical decisions, select projects for funds, and otherwise push his programs forward.

ARPA was so nimble that contractors at times found it easy to forget they were working for Uncle Sam. Ornstein was surprised one day when Roberts picked him up at an airport driving a cheap, battered rental car. Asked why he was driving such a turkey, Roberts, according to Ornstein, “muttered something I didn’t understand at the time about government rules and government expenses and so forth. I had always thought of him as passing out these millions of dollars, but it hadn’t occurred to me that he was, in fact, living personally on quite a limited budget.”

Military involvement with the ARPANET system gradually increased throughout the early 1970s. By 1975, military-related net traffic had reached the point that the DCA finally decided to take control of the system. Meanwhile, ARPA was experimenting with new military applications for ARPANET’s basic packet-switching technology. The experiments were to have a direct bearing on the concept of the Internet—a network formed of other networks, not a single, self-contained system.

The Packet Radio Project was one such experiment. It was the brainchild of ARPA official Robert Kahn, a cousin of the famed nuclear theorist Herman Kahn. As outlined in the early 1970s, packet radio was to be a kind of wireless ARPANET. A central minicomputer near a powerful radio station would be used to communicate with smaller, mobile computing sites. If successful, the idea would have applications throughout the military but particularly within the Army. Mobile computers carried in tanks or trucks would be able to link together whole divisions with ease.

In the mid-1970s, packet radio prototypes were tested successfully in exercises with the XVIII Airborne Corps at Fort Bragg, N. C., and Strategic Air Command at Offutt AFB, Neb. Among other applications, ARPA showed how the equipment could be used for airborne forces on the tarmac to log on to a central computer and automatically reconfigure division load plans if airlift availability suddenly changed. Today, such a capability seems basic. Back then, it was revolutionary.

The system never made it into production or even into development. However, experiments with packet radio and a packet satellite project got Mr. Kahn thinking. What would happen if he could link all these different packet networks together—and with the ARPANET? The result would be a powerful information tool.

Enter the Protocol

If this vision were to become reality, Kahn needed to find a way to connect separate computer nets that, in effect, spoke different languages. He needed a flexible translator—software called a “computer protocol.”

The protocol would have to manage message traffic between many different computers. It would have to detect relay errors, sort out addresses, and perform all sorts of electronic “postal” tasks. With colleague Vinton Cerf, Kahn hammered out the Transmission Control Protocol over a period of several years in the 1970s. Published in a technical journal in May 1974, TCP introduced the concept of networking networks to the wider world.

Kahn described the development in an interview with the Babbage Institute oral history team: “I think the conception of the internetting effort is one that can be credited to Vint Cerf and myself. He and I . . . laid out the grand design, and then Vint worked with the community to develop and evolve it over time.”

By the early 1980s, all the computers hooked to the now-exploding ARPANET had to be capable of using this internetting protocol. Thus was laid the foundation of today’s Internet. Recognizing the civilian importance of computer networking, the Defense Department split ARPANET in two in 1984. The new MILNET would ensure the military had its own reliable network for communications, while a rump ARPANET continued to serve other users.

Soon, the civilian remnant of ARPANET faced a form of competition. The National Science Foundation set up NSFNET in 1984 as a high-speed computer net “backbone” between supercomputer research centers. NSFNET quickly diversified and upgraded, linking universities all across America on lines that were up to 25 times faster than ARPANET lines. This speed attracted many users who might earlier have wanted an ARPANET connection. By the late 1980s the number of computers hooked up to NSF’s net was far greater than the number of ARPANET users.

Within a few years, it became clear that ARPANET was totally obsolete. On June 1, 1990, it was “deinstalled,” an act that ended the system’s 21-year life.

Shortly thereafter, two events cemented NSFNET’s future. In 1991, NSF officials decided to allow commerce on the net, exponentially expanding the universe of potential users. And in 1992, physicist Tim Berners-Lee, working at the Geneva, Switzerland–based Center for Nuclear Research (CERN), developed a method of organizing and linking Internet information that would underpin what soon would be called the World Wide Web.

By the end of 1992, there were one million host computers linked to the Internet. Since then, a graph line of the system’s growth goes almost straight up, like a rocket ascending.

ARPANET left a legacy of both technical and social advances. Whole communities of like-minded souls—one of the first “chat groups” was for science fiction fans, for instance—were pulled together over its wires. Meanwhile, creating the net had produced a cadre of the finest computer scientists in the world for the US.

“I think that there isn’t any doubt that the investment that DARPA made in the ARPANET put packet-switching technology on the map,” said Cerf, now an official at MCI. “It convinced people it was real and has spawned a phenomenal explosion in new kinds of computer communications techniques.”

Many ARPANET pioneers remain convinced that the ARPA bureaucratic model of the early 1970s, with a few bright people in a government office allowed the freedom to pursue particular goals, led to a very effective return on taxpayer investment. Some feel that today’s defense bureaucracy is more ossified and that endless paperwork and a relentless search for the low bidder would make it difficult to duplicate ARPANET’s success.

“It was a joy to be associated with the ARPANET project,” said BBN scientist McKenzie. “It was fun. It was challenging, and I think it was good for the country. It’s not so easy to find that mix now, and I think regulation is a big part of it.”

Peter Grier, the Washington bureau chief of the Christian Science Monitor, is a longtime defense correspondent and regular contributor to Air Force Magazine. His most recent articles, “The Legacy of Airpower” and “Aerospace Technology Exposition,” appeared in the November 1996 issue.