ON OCTOBER 3, 1969, two computers at remote locations “spoke” to each other over the Internet for the first time. Connected by 350 miles of leased telephone line, the two machines, one at the University of California in Los Angeles and the other at Stanford Research Institute in Palo Alto, attempted to transmit the simplest of messages: the word “login,” sent one letter at a time. Charlie Kline, an undergraduate at UCLA, announced to another student at Stanford by telephone, “I’m going to type an L.” He keyed in the letter and then asked, “Did you get the L?” At the other end, the researcher responded, “I got one-one-four”—which, to a computer, is the letter L. Next, Kline sent an “O” over the line. When Kline transmitted the “G” Stanford’s computer crashed. A programming error, repaired after several hours, had caused the problem. Despite the crash, the computers had actually managed to convey a meaningful message, even if not the one planned. In its own phonetic fashion, the UCLA computer said “ello” (L-O) to its compatriot in Stanford. The first, albeit tiny, computer network had been born.
The Internet is one of the defining inventions of the twentieth century, rubbing shoulders with such developments as aircraft, atomic energy, space exploration, and television. Unlike those breakthroughs, however, it did not have its oracles in the nineteenth century; in fact, as late as 1940 not even a modern Jules Verne could have imagined how a collaboration of physical scientists and psychologists would begin a communication revolution. The blue-ribbon laboratories of AT&T;, IBM, and Control Data, when presented with the outlines of the Internet, could not grasp its potential or conceive of computer communication except as a single telephone line using central-office switching methods, a nineteenth-century innovation. Instead, the new vision had to come from outside the businesses that had led the country’s first communication revolution—from new companies and institutions and, most importantly, the brilliant people working at them.
The Internet has a long and complicated history, peppered with landmark insights in both communications and artificial intelligence. This essay, part memoir and part history, traces its roots from their origin in World War II voice-communication laboratories to the creation of the first Internet prototype, known as ARPANET—the network through which UCLA spoke to Stanford in 1969. Its name derived from its sponsor, the Advanced Research Projects Agency (ARPA) in the U.S. Department of Defense. Bolt Beranek and Newman (BBN), the firm that I helped create in the late 1940s, built ARPANET and served for twenty years as its manager—and now provides me with the opportunity to relate the network’s story. Along the way, I hope to identify the conceptual leaps of a number of gifted individuals, as well as their hard work and production skills, without which your e-mail and web surfing would not be possible. Key among these innovations are man-machine symbiosis, computer time-sharing, and the packet-switched network, of which ARPANET was the world’s first incarnation. The significance of these inventions will come to life, I hope, along with some of their technical meaning, in the course of what follows.
Prelude to ARPANET
During World War II, I served as director at Harvard’s Electro-Acoustic Laboratory, which collaborated with the Psycho-Acoustic Laboratory. The daily, close cooperation between a group of physicists and a group of psychologists was, apparently, unique in history. One outstanding young scientist at PAL made a particular impression on me: J. C. R. Licklider, who demonstrated an unusual proficiency in both physics and psychology. I would make a point of keeping his talents close by in the ensuing decades, and they would ultimately prove vital to ARPANET’s creation.
At the close of the war I migrated to MIT and became associate professor of Communication Engineering and Technical Director of its Acoustics Laboratory. In 1949, I convinced MIT’s Department of Electrical Engineering to appoint Licklider as a tenured associate professor to work with me on voice communication problems. Shortly after his arrival, the chair of the department asked Licklider to serve on a committee that established Lincoln Laboratory, an MIT research powerhouse supported by the Department of Defense. The opportunity introduced Licklider to the nascent world of digital computing—an introduction that brought the world one step closer to the Internet.
In 1948, I ventured out—with MIT’s blessing—to form the acoustical consulting firm Bolt Beranek and Newman with my MIT colleagues Richard Bolt and Robert Newman. The firm incorporated in 1953, and as its first president I had the opportunity to guide its growth for the next sixteen years. By 1953, BBN had attracted top-flight post-doctorates and obtained research support from government agencies. With such resources right at hand, we began to expand into new areas of research, including psychoacoustics in general and, in particular, speech compression—that is, the means for shortening the length of a speech segment during transmission; criteria for prediction of speech intelligibility in noise; the effects of noise on sleep; and last but certainly not least, the still-nascent field of artificial intelligence, or machines that seem to think. Because of the prohibitive cost of digital computers, we made do with analog ones. This meant, however, that a problem that could be computed on today’s PC in a few minutes then might take a full day or even a week.
In the mid 1950s, when BBN decided to pursue research about how machines could efficiently amplify human labor, I decided we needed an outstanding experimental psychologist to head up the activity, preferably one acquainted with the then rudimentary field of digital computers. Licklider, naturally, became my top candidate. My appointment book shows that I courted him with numerous lunches in the spring of 1956 and one critical meeting in Los Angeles that summer. A position at BBN meant that Licklider would give up a tenured faculty position, so to convince him to join the firm we offered stock options—a common benefit in the Internet industry today. In the spring of 1957, Licklider came aboard BBN as a vice president.
Lick, as he insisted that we call him, stood about six feet tall, appeared thin boned, almost fragile, with thinning brown hair offset by enthusiastic blue eyes. Outgoing and always on the verge of a smile, he ended almost every second sentence with a slight chuckle, as though he had just made a humorous statement. He walked with a brisk but gentle step, and he always found the time to listen to new ideas. Relaxed and self-deprecating, Lick merged easily with the talent already at BBN. He and I worked together especially well: I cannot remember a time when we disagreed.
Licklider had been on staff only a few months when he told me that he wanted BBN to buy a digital computer for his group. When I pointed out that we already had a punched-card computer in the financial department and analog computers in the experimental psychology group, he replied that they did not interest him. He wanted a then state-of-the-art machine produced by the Royal-McBee Company, a subsidiary of Royal Typewriter. “What will it cost?” I asked. “Around $30,000,” he replied, rather blandly, and noted that this price tag was a discount he had already negotiated. BBN had never, I exclaimed, spent anything approaching that amount of money on a single research apparatus. “What are you going to do with it?” I queried. “I don’t know,” Lick responded, “but if BBN is going to be an important company in the future, it must be in computers.” Although I hesitated at first—$30,000 for computer with no apparent use seemed just too reckless—I had a great deal of faith in Lick’s convictions and finally agreed that BBN should risk the funds. I presented his request to the other senior staff, and with their approval, Lick brought BBN into the digital era.
The Royal-McBee turned out to be our entrée into a much larger venue. Within a year of the computer’s arrival, Kenneth Olsen, the president of the fledgling Digital Equipment Corporation, stopped by BBN, ostensibly just to see our new computer. After chatting with us and satisfying himself that Lick really understood digital computation, he asked if we would consider a project. He explained that Digital had just completed construction of a prototype of their first computer, the PDP-1, and that they needed a test site for a month. We agreed to try it.
The prototype PDP-1 arrived shortly after our discussions. A behemoth compared to the Royal-McBee, it would fit no place in our offices except the visitors’ lobby, where we surrounded it with Japanese screens. Lick and Ed Fredkin, a youthful and eccentric genius, and several others put it through its paces for most of the month, after which Lick provided Olsen with a list of suggested improvements, especially how to make it more user-friendly. The computer had won us all over, so BBN arranged for Digital to provide us with their first production PDP-1 on a standard lease basis. Then Lick and I took off for Washington to seek research contracts that would make use of this machine, which carried a 1960 price tag of $150,000. Our visits to the Department of Education, National Institutes of Health, National Science Foundation, NASA, and the Department of Defense proved Lick’s convictions correct, and we secured several important contracts.
Between 1960 and 1962, with BBN’s new PDP-1 in-house and several more on order, Lick turned his attention to some of the fundamental conceptual problems that stood between an era of isolated computers that worked as giant calculators and the future of communications networks. The first two, deeply interrelated, were man-machine symbiosis and computer time-sharing. Lick’s thinking had a definitive impact on both.
He became a crusader for man-machine symbiosis as early as 1960, when he wrote a trailblazing paper that established his critical role in the making of the Internet. In that piece, he investigated the implications of the concept at length. He defined it essentially as “an interactive partnership of man and machine” in which
Men will set the goals, formulate the hypotheses, determine the criteria, and perform the evaluations. Computing machines will do the routinizable work that must be done to prepare the way for insights and decisions in technical and scientific thinking.
He also identified “prerequisites for … effective, cooperative association,” including the key concept of computer time-sharing, which imagined the simultaneous use of a machine by many persons, allowing, for example, employees in a large company, each with a screen and keyboard, to use the same mammoth central computer for word processing, number crunching, and information retrieval. As Licklider envisioned the synthesis of man-machine symbiosis and computer time-sharing, it could make it possible for computer users, via telephone lines, to tap into mammoth computing machines at various centers located nationwide.
Of course, Lick alone did not develop the means for making time-sharing work. At BBN, he tackled the problem with John McCarthy, Marvin Minsky, and Ed Fredkin. Lick brought McCarthy and Minsky, both artificial intelligence experts at MIT, to BBN to work as consultants in the summer of 1962. I had met neither of them before they started. Consequently, when I saw two strange men sitting at a table in the guest conference room one day, I approached them and asked, “Who are you?” McCarthy, nonplussed, answered, “Who are you?” The two worked well with Fredkin, whom McCarthy credited with insisting that “time sharing could be done on a small computer, namely a PDP-1.” McCarthy also admired his indominatable can-do attitude. “I kept arguing with him,” McCarthy recalled in 1989. “I said that an interrupt system was needed. And he said, ‘We can do that.’ Also needed was some kind of swapper. ‘We can do that.'” (An “interrupt” breaks a message into packets; a “swapper” interleaves message packets during transmission and reassembles them separately on arrival.)
The team quickly produced results, creating a modified PDP-1 computer screen divided into four parts, each assigned to a separate user. In the fall of 1962, BBN conducted the first public demonstration of time-sharing, with one operator in Washington, D.C., and two in Cambridge. Concrete applications followed soon after. That winter, for example, BBN installed a time-shared information system in the Massachusetts General Hospital that allowed nurses and doctors to create and access patient records at nurses’ stations, all connected to a central computer. BBN also formed a subsidiary company, TELCOMP, that allowed subscribers in Boston and New York to access our time-shared digital computers by using teletypewriters connected to our machines via dial-up telephone lines.
The time-sharing breakthrough also spurred BBN’s internal growth. We purchased ever-more advanced computers from Digital, IBM, and SDS, and we invested in separate large-disk memories so specialized we had to install them in a spacious, raised-floor, air-conditioned room. The firm also won more prime contracts from federal agencies than any other company in New England. By 1968, BBN had hired over 600 employees, more than half in the computer division. Those included many names now famous in the field: Jerome Elkind, David Green, Tom Marill, John Swets, Frank Heart, Will Crowther, Warren Teitelman, Ross Quinlan, Fisher Black, David Walden, Bernie Cosell, Hawley Rising, Severo Ornstein, John Hughes, Wally Feurzeig, Paul Castleman, Seymour Papert, Robert Kahn, Dan Bobrow, Ed Fredkin, Sheldon Boilen, and Alex McKenzie. BBN soon became known as Cambridge’s “Third University”—and to some academics the absence of teaching and committee assignments made BBN more appealing than the other two.
This infusion of eager and brilliant computer nicks—1960s lingo for geeks—changed the social character of BBN, adding to the spirit of freedom and experimentation the firm encouraged. BBN’s original acousticians exuded traditionalism, always wearing jackets and ties. Programmers, as remains the case today, came to work in chinos, T-shirts, and sandals. Dogs roamed the offices, work went on around the clock, and coke, pizza, and potato chips constituted dietary staples. The women, hired only as technical assistants and secretaries in those antediluvian days, wore slacks and often went without shoes. Blazing a trail still underpopulated today, BBN set up a day nursery to accommodate the staff’s needs. Our bankers—upon whom we depended for capital—unfortunately remained inflexible and conservative, so we had to keep them from seeing this strange (to them) menagerie.
In October 1962, the Advanced Research Projects Agency (ARPA), an office within the U.S. Department of Defense, lured Licklider away from BBN for a one-year stint, which stretched into two. Jack Ruina, ARPA’s first director, convinced Licklider that he could best spread his time-sharing theories around the country through the government’s Information Processing Techniques Office (IPTO), where Lick became Director of Behavioral Sciences. Because ARPA had purchased mammoth computers for a score of university and government laboratories during the 1950s, it already had resources spread across the country that Lick could exploit. Intent on demonstrating that these machines could do more than numerical calculation, he promoted their use for interactive computing. By the time Lick finished his two years, ARPA had spread the development of time-sharing nationwide through contract awards. Because Lick’s stockholdings posed a possible conflict of interest, BBN had to let this research gravy-train pass it by.
After Lick’s term the directorship eventually passed to Robert Taylor, who served from 1966 to 1968 and oversaw the agency’s initial plan to build a network that allowed computers at ARPA-affiliated research centers across the country to share information. According to the stated purpose of ARPA’s goals, the hypothesized network should allow small research laboratories to access large-scale computers at large research centers and thus relieve ARPA from supplying every laboratory with its own multimillion dollar machine. Prime reponsibility for managing the network project within ARPA went to Lawrence Roberts from Lincoln Laboratory, whom Taylor recruited in 1967 as IPTO Program Manager. Roberts had to devise the basic goals and building blocks of the system and then find an appropriate firm to build it under contract.
In order to lay the groundwork for the project, Roberts proposed a discussion among the leading thinkers on network development. Despite the tremendous potential such a meeting of the minds seemed to hold, Roberts met with little enthusiasm from the men he contacted. Most said their computers were busy full time and that they could think of nothing they would want to do cooperatively with other computer sites. Roberts proceeded undaunted, and he did eventually pull ideas from some researchers—primarily Wes Clark, Paul Baran, Donald Davies, Leonard Kleinrock, and Bob Kahn.
Wes Clark, at Washington University in St. Louis, contributed a critical idea to Roberts’s plans: Clark proposed a network of identical, interconnected mini-computers, which he called “nodes.” The large computers at various participating locations, rather than hooking directly into a network, would each hook into a node; the set of nodes would then manage the actual routing of data along the network lines. Through this structure, the difficult job of traffic management would not further burden the host computers, which had to otherwise receive and process information. In a memorandum outlining Clark’s suggestion, Roberts renamed the nodes “Interface Message Processors” (IMPs). Clark’s plan exactly prefigured the Host-IMP relationship that would make ARPANET work.
Paul Baran, of the RAND Corporation, unwittingly supplied Roberts with key ideas about how the transmission could work and what the IMPs would do. In 1960, when Baran had tackled the problem of how to protect vulnerable telephone communication systems in case of a nuclear attack, he had imagined a way to break one message down into several “message blocks,” route the separate pieces over different routes (telephone lines), and then reassemble the whole at its destination. In 1967, Roberts discovered this treasure in the U.S. Air Force files, where Baran’s eleven volumes of explanation, compiled between 1960 and 1965, languished untested and unused.
Donald Davies, at the National Physical Laboratory in Great Britain, was working out a similar network design in the early 1960s. His version, formally proposed in 1965, coined the “packet switching” terminology that ARPANET would ultimately adopt. Davies suggested splitting typewritten messages into data “packets” of a standard size and time-sharing them on a single line—thus, the process of packet switching. Although he proved the elementary feasibility of his proposal with an experiment in his laboratory, nothing further came of his work until Roberts drew on it.
Leonard Kleinrock, now at the University of Los Angeles, finished his thesis in 1959, and in 1961 he wrote an MIT report that analyzed data flow in networks. (He later expanded this study in his 1976 book Queuing Systems, which showed in theory that packets could be queued without loss.) Roberts used Kleinrock’s analysis to bolster his confidence on the feasibility of a packet-switched network, and Kleinrock convinced Roberts to incorporate measurement software that would monitor the network’s performance. After the ARPANET was installed, he and his students handled the monitoring.
Pulling together all of these insights, Roberts decided that ARPA should pursue “a packet switching network.” Bob Kahn, at BBN, and Leonard Kleinrock, at UCLA, convinced him of the need for a test using a full-scale network on long-distance telephone lines rather than just a laboratory experiment. As daunting as that test would be, Roberts had obstacles to overcome even to reach that point. The theory presented a high likelihood of failure, largely because so much about the overall design remained uncertain. Older Bell Telephone engineers declared the idea entirely unworkable. “Communications professionals,” Roberts wrote, “reacted with considerable anger and hostility, usually saying I did not know what I was talking about.” Some of the big companies maintained that the packets would circulate forever, making the whole effort a waste of time and money. Besides, they argued, why would anyone want such a network when Americans already enjoyed the world’s best telephone system? The communications industry would not welcome his plan with open arms.
Nonetheless, Roberts released ARPA’s “request for proposal” in the summer of 1968. It called for a trial network made up of four IMPs connected to four host computers; if the four-node network proved itself, the network would expand to include fifteen more hosts. When the request arrived at BBN, Frank Heart took on the job of administering BBN’s bid. Heart, athletically built, stood just under six feet tall and sported a high crew cut that looked like a black brush. When excited, he spoke in a loud, high-pitched voice. In 1951, his senior year at MIT, he had signed up for the school’s very first course in computer engineering, from which he caught the computer bug. He worked at Lincoln Laboratory for fifteen years before coming to BBN. His team at Lincoln, all later at BBN, included Will Crowther, Severo Ornstein, Dave Walden, and Hawley Rising. They had become experts at connecting electrical measuring devices to telephone lines to gather information, thus becoming pioneers in computing systems that worked in “real time” as opposed to recording data and analyzing it later.
Heart approached each new project with great caution and would not accept an assignment unless confident that he could meet specifications and deadlines. Naturally, he approached the ARPANET bid with apprehension, given the proposed system’s riskiness and a schedule that didn’t allow sufficient time for planning. Nonetheless, he did take it on, persuaded by BBN colleagues, myself included, who believed that the company should push ahead into the unknown.
Heart started by pulling together a small team of those BBN staff members with the most knowledge about computers and programming. They included Hawley Rising, a quiet electrical engineer; Severo Ornstein, a hardware geek who had worked at Lincoln Laboratory with Wes Clark; Bernie Cosell, a programmer with an uncanny ability to find bugs in complex programming; Robert Kahn, an applied mathematician with a strong interest in the theory of networking; Dave Walden, who had worked on real-time systems with Heart at Lincoln Laboratory; and Will Crowther, also a Lincoln Lab colleague and admired for his ability to write compact code. With only four weeks to complete the proposal, no-one in this crew could plan on a decent night’s sleep. The ARPANET group worked until nearly dawn, day after day, researching every detail of how to make this system work.
The final proposal filled two hundred pages and cost more than $100,000 to prepare, the most the company had ever spent on such a risky project. It covered every conceivable aspect of the system, beginning with the computer that would serve as the IMP at each host location. Heart had influenced this choice with his adamance that the machine must be reliable above all else. He favored Honeywell’s new DDP-516—it had the correct digital capacity and could handle input and output signals with speed and efficiency. (Honeywell’s manufacturing plant only stood a short drive from BBN’s offices.) The proposal also spelled out how the network would address and queue the packets; determine the best available transmission routes to avoid congestion; recover from line, power, and IMP failures; and monitor and debug the machines from a remote-control center. During the research BBN also determined that the network could process the packets much more quickly than ARPA had expected—in only about one-tenth the time originally specified. Even so, the document cautioned ARPA that “it will be difficult to make the system work.”
Although 140 companies received Roberts’s request and 13 submitted proposals, BBN was one of only two that made the government’s final list. All the hard work paid off. On December 23, 1968, a telegram arrived from Senator Ted Kennedy’s office congratulating BBN “on winning the contract for the interfaith [sic] message processor.” Related contracts for the initial host sites went to UCLA, the Stanford Research Institute, the University of California at Santa Barbara, and the University of Utah. The government relied on this group of four, partly because East Coast universities lacked enthusiasm for ARPA’s invitation to join in the early trials and partly because the government wanted to avoid the high costs of cross-country leased lines in the first experiments. Ironically, these factors meant that BBN was fifth on the first network.
As much work as BBN had invested in the bid, it proved infinitesimal compared to the work that came next: designing and building a revolutionary communications network. Although BBN had to create only a four-host demonstration network to start with, the eight-month deadline imposed by the government contract forced the staff into weeks of marathon late-night sessions. Since BBN was not responsible for providing or configuring the host computers at each host site, the bulk of its work would revolve around the IMPs—the idea developed from Wes Clark’s “nodes”—that had to connect the computer at each host site to the system. Between New Year’s Day and September 1, 1969, BBN had to design the overall system and determine the network’s hardware and software needs; acquire and modify the hardware; develop and document procedures for the host sites; ship the first IMP to UCLA, and one a month thereafter to the Stanford Research Institute, UC Santa Barbara, and the University of Utah; and, finally, oversee the arrival, installation, and operation of each machine. To build the system, the BBN staff broke into two teams, one for the hardware—generally referred to as the IMP team—and the other for software.
The hardware team had to begin by designing the basic IMP, which they created by modifying Honeywell’s DDP-516, the machine Heart had selected. This machine was truly elementary and posed a real challenge to the IMP team. It had neither a hard drive nor a floppy drive and possessed only 12,000 bytes of memory, a far cry from the 100,000,000,000 bytes available in modern desktop computers. The machine’s operating system—the rudimentary version of the Windows OS on most of our PCs—existed on punched paper tapes about a half inch wide. As the tape moved across a light bulb in the machine, light passed through the punched holes and actuated a row of photocells that the computer used to “read” the data on the tape. A portion of software information might take yards of tape. To allow this computer to “communicate,” Severo Ornstein designed electronic attachments that would transfer electrical signals in it and would receive signals from it, not unlike the signals the brain sends out as speech and takes in as hearing.
Willy Crowther headed the software team. He possessed the ability to keep the whole software skein in mind, as one colleague said, “like designing a whole city while keeping track of the wiring to each lamp and the plumbing to every toilet.” Dave Walden concentrated on the programming issues that dealt with communication between an IMP and its host computer and Bernie Cosell worked on process and debugging tools. The three spent many weeks developing the routing system that would relay each packet from one IMP to another until it reached its destination. The need for developing alternate paths for the packets—that is, packet switching—in case of path congestion or breakdown proved especially challenging. Crowther responded to the problem with a dynamic routing procedure, a masterpiece of programming, that earned the highest respect and praise from his colleagues.
In a process so complex that invited it occasional error, Heart demanded that we make the network reliable. He insisted on frequent oral reviews of the staff’s work. Bernie Cosell recalled, “It was like your worst nightmare for an oral exam by someone with psychic abilities. He could intuit the parts of the design you were least sure of, the places you understood least well, the areas where you were just song-and-dancing, trying to get by, and cast an uncomfortable spotlight on parts you least wanted to work on.”
In order to insure that all of this would work once staff and machines were operating at locations hundreds if not thousands of miles apart, BBN needed to develop procedures for connecting host computers to the IMPs—especially since the computers at the host sites all had different characteristics. Heart gave the responsibility for preparing the document to Bob Kahn, one of BBN’s best writers and an expert on the flow of information through the overall network. In two months, Kahn completed the procedures, which became known as BBN Report 1822. Kleinrock later remarked that anybody “who was involved in the ARPANET will never forget that report number because it was the defining spec for how the things would mate.”
Despite the detailed specifications that the IMP team had sent Honeywell about how to modify the DDP-516, the prototype that arrived at BBN didn’t work. Ben Barker took on the job of debugging the machine, which meant rewiring the hundreds of “pins” nestled in four vertical drawers at the back of the cabinet (see photo). To move the wires that were tightly wrapped around these delicate pins, each roughly one-tenth of an inch from its neighbors, Barker had to use a heavy “wire-wrap gun” that constantly threatened to snap the pins, in which case we would have to replace an entire pin board. During the months that this work took, BBN meticulously tracked all the changes and passed the information on to the Honeywell engineers, who could then ensure that the next machine they sent would function properly. We hoped to check it over quickly—our Labor Day deadline was looming large—before shipping it to UCLA, the first host in line for IMP installation. But we were not so lucky: the machine arrived with many of the same problems, and again Barker had to go in with his wire-wrap gun.
Finally, with wires all properly wrapped and only a week or so to go before we had to ship our official IMP No. 1 to California, we ran into one last problem. The machine now worked correctly, but it still crashed, sometimes as often as once a day. Barker suspected a “timing” problem. A computer’s timer, an internal clock of sorts, synchronizes all its operations; the Honeywell’s timer “ticked” one million times per second. Barker, figuring that the IMP crashed whenever a packet arrived between two of these ticks, worked with Ornstein to correct the problem. At last, we test drove the machine with no accidents for one full day—the last day we had before we had to ship it to UCLA. Ornstein, for one, felt confident that it had passed the real test: “We had two machines operating in the same room together at BBN, and the difference between a few feet of wire and a few hundred miles of wire made no difference…. [W]e knew it was going to work.”
Off it went, air freight, across the country. Barker, who had traveled on a separate passenger flight, met the host team at UCLA, where Leonard Kleinrock managed about eight students, including Vinton Cerf as designated captain. When the IMP arrived, its size (about that of a refrigerator) and weight (about half a ton) amazed everyone. Nonetheless, they placed its drop-tested, battleship-gray, steel case tenderly beside their host computer. Barker watched nervously as UCLA staff turned the machine on: it worked perfectly. They ran a simulated transmission with their computer, and soon the IMP and its host were “talking” to each other flawlessly. When Barker’s good news arrived back in Cambridge, Heart and the IMP gang erupted in cheers.
On October 1, 1969, the second IMP arrived at the Stanford Research Institute exactly on schedule. This delivery made the first real ARPANET test possible. With their respective IMPs connected across 350 miles through leased, fifty-kilobit telephone line, the two host computers stood ready to “talk.” On October 3, they said “ello” and brought the world into the age of the Internet.
The work that followed this inauguration certainly wasn’t easy or trouble-free, but the solid foundation was undeniably in place. BBN and the host sites completed the demonstration network, which added UC Santa Barbara and the University of Utah to the system, before the end of 1969. By spring 1971, ARPANET encompassed the nineteen institutions that Larry Roberts had originally proposed. Furthermore, in little more than a year after initiation of the four-host network, a collaborative working group had created a common set of operating instructions that would make certain the disparate computers could communicate with each other—that is, host-to-host protocols. The work this group performed set certain precedents that went beyond simple guidelines for remote logins (allowing the user at host “A” to connect to the computer at host “B”) and file transfer. Steve Crocker at UCLA, who volunteered to keep notes of all the meetings, many of which were telephone conferences, wrote them so skillfully that no contributor felt humbled: each felt that the rules of the network had developed by cooperation, not by ego. Those first Network Control Protocols set the standard for the operation and improvement of the Internet and even the World Wide Web today: no one person, group, or institution would dictate standards or rules of operation; instead, decisions are made by international consensus.
ARPANET’s Rise and Demise
With the Network Control Protocol available, the ARPANET architects could pronounce the entire enterprise a success. Packet switching, unequivocally, provided the means for efficient use of communication lines. An economical and reliable alternate to circuit switching, the basis for the Bell Telephone system, the ARPANET had revolutionized communication.
Despite the tremendous success achieved by BBN and the original host sites, ARPANET was still underutilized by the end of 1971. Even the hosts now plugged into the network often lacked the basic software that would allow their computers to interface with their IMP. “The obstacle was the enormous effort it took to connect a host to an IMP,” one analyst explains. “Operators of a host had to build a special-purpose hardware interface between their computer and its IMP, which could take from 6 to 12 months. They also needed to implement the host and network protocols, a job that required up to 12 man-months of programming, and they had to make these protocols work with the rest of the computer’s operating system. Finally, they had to adjust the applications developed for local use so they could be accessed over the network.” ARPANET worked, but its builders still needed to make it accessible—and appealing.
Larry Roberts decided the time had come to put on a show for the public. He arranged for a demonstration at the International Conference on Computer Communication held in Washington, D.C., on October 24–26, 1972. Two fifty-kilobit lines installed in the hotel’s ballroom connected to the ARPANET and thence to forty remote computer terminals at various hosts. On the exhibition’s opening day, AT&T; executives toured the event and, as if planned just for them, the system crashed, bolstering their view that packet switching would never replace the Bell system. Aside from that one mishap, however, as Bob Kahn said after the conference, the “public reaction varied from delight that we had so many people in one place doing all this stuff and it all worked, to astonishment that it was even possible.” Daily use of the network jumped immediately.
Had ARPANET been restricted to its original purpose of sharing computers and exchanging files, it would have been judged a minor failure, because traffic seldom exceeded 25 percent of capacity. Electronic mail, also a milestone of 1972, had a great deal to do with drawing users in. Its creation and eventual ease of use owed much to the inventiveness of Ray Tomlinson at BBN (responsible, among other things, for choosing the @ icon for e-mail addresses), Larry Roberts, and John Vittal, also at BBN. By 1973, three quarters of all traffic on the ARPANET was e-mail. “You know,” Bob Kahn remarked, “everyone really uses this thing for electronic mail.” With e-mail, the ARPANET soon became loaded to capacity.
By 1983, the ARPANET contained 562 nodes and had become so large that the government, unable to guarantee its security, divided the system into MILNET for government laboratories and ARPANET for all others. It also now existed in the company of many privately supported networks, including some instituted by corporations such as IBM, Digital, and Bell Laboratories. NASA established the Space Physics Analysis Network, and regional networks began forming across the country. Combinations of networks—that is, the Internet—became possible through a protocol developed by Vint Cerf and Bob Kahn. With its capacity far outstripped by these developments, the original ARPANET diminished in significance, until the government concluded that it could save $14 million per year by closing it down. Decommissioning finally occurred by late 1989, just twenty years after the system’s first “ello”—but not before other innovators, including Tim Berners-Lee, had devised ways to expand the technology into the global system we now call the World Wide Web.
Early in the new century the number of homes connected to the Internet will equal the number that now have televisions. The Internet has succeeded wildly beyond early expectations because it has immense practical value and because it is, quite simply, fun. In the next stage of progress, operating programs, word processing, and the like will be centralized on large servers. Homes and offices will have little hardware beyond a printer and a flat screen where desired programs will flash up at voice command and will operate by voice and body movements, rendering the familiar keyboard and mouse extinct. And what else, beyond our imagination today?
LEO BERANEK holds a doctorate in science from Harvard University. Besides a teaching career at both Harvard and MIT, he has founded several businesses in the USA and Germany and has been a leader in Boston community affairs.
1. Katie Hafner and Matthew Lyon, Where Wizards Stay Up Late (New York, 1996), 153.
2. The standard histories of the Internet are Funding a Revolution: Government Support for Computing Research (Washington, D. C., 1999); Hafner and Lyon, Where Wizards Stay Up Late; Stephen Segaller, Nerds 2.0.1: A Brief History of the Internet (New York, 1998); Janet Abbate, Inventing the Internet (Cambridge, Mass., 1999); and David Hudson and Bruce Rinehart, Rewired (Indianapolis, 1997).
3. J. C. R. Licklider, interview by William Aspray and Arthur Norberg, Oct. 28, 1988, transcript, pp. 4–11, Charles Babbage Institute, University of Minnesota (cited hereafter as CBI).
4. My papers, including the apppointment book referred to, are housed in the Leo Beranek Papers, Institute Archives, Massachusetts Institute of Technology, Cambridge, Mass. BBN’s personnel records also shored up my memory here. Much of what follows, however, unless otherwise cited, comes from my own recollections.
5. My recollections here were augmented by a personal discussion with Licklider.
6. Licklider, interview, pp. 12–17, CBI.
7. J. C. R. Licklider, “Man-Machine Symbosis,” IRE Transactions on Human Factors in Electronics 1 (1960):4–11.
8. John McCarthy, interview by William Aspray, Mar. 2, 1989, transcript, pp. 3, 4, CBI.
9. Licklider, interview, p. 19, CBI.
10. One of the primary motivations behind the ARPANET initiative was, according to Taylor, “sociological” rather than “technical.” He saw the opportunity to create a countrywide discussion, as he explained later: “The events that got me interested in networking had little to do with technical issues but rather with sociological issues. I had witnessed [at those laboratories] that bright, creative people, by virtue of the fact that they were beginning to use [time-shared systems] together, were forced to talk to one another about, ‘What’s wrong with this? How do I do that? Do you know anyone who has some data about this? … I thought, ‘Why couldn’t we do this across the country?’ … This motivation … came to be known as the ARPANET. [To succeed] I had to … (1) convince ARPA, (2) convince IPTO contractors that they really wanted to be nodes on this network, (3) find a program manager to run it, and (4) select the right group for the implementation of it all…. A number of people [that I talked with] thought that … the idea of an interactive, nation-wide network was not very interesting. Wes Clark and J. C. R. Licklider were two who encouraged me.” From remarks at The Path to Today, the University of California—Los Angeles, Aug. 17, 1989, transcript, pp. 9–11, CBI.
11. Hafner and Lyon, Where Wizards Stay Up Late, 71, 72.
12. Hafner and Lyon, Where Wizards Stay Up Late, 73, 74, 75.
13. Hafner and Lyon, Where Wizards Stay Up Late, 54, 61; Paul Baran, “On Distributed Communications Networks,” IEEE Transactions on Communications (1964):1–9, 12; Path to Today, pp. 17–21, CBI.
14. Hafner and Lyon, Where Wizards Stay Up Late, 64–66; Segaller, Nerds, 62, 67, 82; Abbate, Inventing the Internet, 26–41.
15. Hafner and Lyon, Where Wizards Stay Up Late, 69, 70. Leonard Kleinrock stated in 1990 that “The mathematical tool that had been developed in queuing theory, namely queuing networks, matched [when adjusted] the model of [later] computer networks…. Then I developed some design procedures as well for optimal capacity assignment, routing procedures and topology design.” Leonard Kleinrock, interview by Judy O’Neill, Apr. 3, 1990, transcript, p. 8, CBI.
Roberts didn’t mention Kleinrock as a major contributor to the planning of the ARPANET in his presentation at the UCLA conference in 1989, even with Kleinrock present. He stated: “I got this huge collection of reports [Paul Baran’s work] … and suddenly I learned how to route packets. So we talked to Paul and used all of his [packet switching] concepts and put together the proposal to go out on the ARPANET, the RFP, which, as you know, BBN won.” Path to Today, p. 27, CBI.
Frank Heart has since stated that “we were unable to use any of the work of Kleinrock or Baran in the design of the ARPANET. We had to develop the operating features of the ARPANET ourselves.” Telephone conversation between Heart and the author, Aug. 21, 2000.
16. Kleinrock, interview, p. 8, CBI.
17. Hafner and Lyon, Where Wizards Stay Up Late, 78, 79, 75, 106; Lawrence G. Roberts, “The ARPANET and Computer Networks,” in A History of Personal Workstations, ed. A. Goldberg (New York, 1988), 150. In a joint paper authored in 1968, Licklider and Robert Taylor also envisioned how such access could make use of standard telephone lines without overwhelming the system. The answer: the packet-switched network. J. C. R. Licklider and Robert W. Taylor, “The Computer as a Communication Device,” Science and Technology 76 (1969):21–31.
18. Defense Supply Service, “Request for Quotations,” July 29, 1968, DAHC15-69-Q-0002, National Records Building, Washington, D.C. (copy of original document courtesy of Frank Heart); Hafner and Lyon, Where Wizards Stay Up Late, 87–93. Roberts states: “The final product [the RFP] demonstrated that there were many problems to surmount before ‘invention’ had occurred. The BBN team developed significant aspects of the network’s internal operations, such as routing, flow control, software design, and network control. Other players [named in the text above] and my contributions were a vital part of the ‘invention.'” Stated earlier and verified in an e-mail exchange with the author, Aug. 21, 2000.
Thus, BBN, in the language of a patent office, “reduced to practice” the concept of a packet-switched wide-area network. Stephen Segaller writes that “What BBN did invent was doing packet switching, rather than proposing and hypothesizing packet switching” (emphasis in original). Nerds, 82.
19. Hafner and Lyon, Where Wizards Stay Up Late, 97.
20. Hafner and Lyon, Where Wizards Stay Up Late, 100. BBN’s work reduced the speed from ARPA’s original estimation of 1/2 second to 1/20.
21. Hafner and Lyon, Where Wizards Stay Up Late, 77. 102–106.
22. Hafner and Lyon, Where Wizards Stay Up Late, 109–111.
23. Hafner and Lyon, Where Wizards Stay Up Late, 111.
24. Hafner and Lyon, Where Wizards Stay Up Late, 112.
25. Segaller, Nerds, 87.
26. Segaller, Nerds, 85.
27. Hafner and Lyon, Where Wizards Stay Up Late, 150, 151.
28. Hafner and Lyon, Where Wizards Stay Up Late, 156, 157.
29. Abbate, Inventing the Internet, 78.
30. Abbate, Inventing the Internet, 78–80; Hafner and Lyon, Where Wizards Stay Up Late, 176–186; Segaller, Nerds, 106–109.
31. Hafner and Lyon, Where Wizards Stay Up Late, 187–205. After what was really a “hack” between two computers, Ray Tomlinson at BBN wrote a mail program that had two parts: one to send, called SNDMSG, and the other to receive, called READMAIL. Larry Roberts further streamlined e-mail by writing a program for listing the messages and a simple means for accessing and deleting them. Another valuable contribution was “Reply,” added by John Vittal, which allowed recipients to answer a message without retyping the whole address.
32. Vinton G. Cerf and Robert E. Kahn, “A Protocol for Packet Network Intercommunication,” IEEE Transactions on Communications COM-22 (May 1974):637-648; Tim Berners-Lee, Weaving the Web (New York, 1999); Hafner and Lyon, Where Wizards Stay Up Late, 253–256.
33. Janet Abbate wrote that “The ARPANET … developed a vision of what a network should be and worked out the techniques that would make this vision a reality. Creating the ARPANET was a formidable task that presented a wide range of technical obstacles…. ARPA did not invent the idea of layering [layers of addresses on each packet]; however, the ARPANET’s success popularized layering as a networking technique and made it a model for builders of other networks…. The ARPANET also influenced the design of computers … [and of] terminals that could be used with a variety of systems rather than just a single local computer. Detailed accounts of the ARPANET in the professional computer journals disseminated its techniques and legitimized packet switching as a reliable and economic alternative for data communication…. The ARPANET would train a whole generation of American computer scientists to understand, use, and advocate its new networking techniques.” Inventing the Internet, 80, 81.
By LEO BERANEK