By the mid-1960s, the first time-sharing systems had already recapitulated the early history of the first telephone exchanges. Entrepreneurs built those exchanges as a means to allow subscribers to summon services such as a taxi, a doctor, or the fire brigade. But those subscribers soon found their local exchange just as useful for communicating and socializing with each other1. Likewise time-sharing systems, initially created to allow their users to “summon” computer power, had become communal switchboards with built-in messaging services2. In the decade to follow, computers would follow the next stage in the history of the telephone – the interconnection of exchanges to form regional and long-distance networks.
The first attempt to actually connect multiple computers into a larger whole was the ur-project of interactive computing itself, the SAGE air defense system. Because each of the twenty-three SAGE direction centers covered a particular geographical area, some mechanism was needed for handing off radar tracks from one center to another when incoming aircraft crossed a boundary between those areas. The SAGE designers dubbed this problem “cross-telling,” and they solved it by building data links on dedicated AT&T phone lines among all the neighboring direction centers. Ronald Enticknap, part of a small Royal Air Force delegation to SAGE, oversaw the design and implementation of this subsystem. Unfortunately, I have found no detailed description of the cross-telling function, but evidently each direction center computer determined when a track was crossing into another sector and sent its record over the phone line to that sector’s computer, where it could be picked up by an operator monitoring a terminal there3.
The SAGE system’s need to translate digital data into an analog signal over the phone line (and then back again at the receiving station) occasioned AT&T to develop the Bell 101 “dataset”, which could deliver a modest 110 bits per second. This kind of device was later called a “modem”, for its ability to modulate the analog telephone signal using an outgoing series of digital data, and demodulate the bits from the incoming wave form.
SAGE thus laid some important technical groundwork for later computer networks. The first computer network of lasting significance, however, is one whose name is well known even today: ARPANET. Unlike SAGE, it connected a diverse set of time-shared and batch-processing hardware each with its own custom software, and was intended to be open-ended in scope and function, fulfilling whatever purposes users might desire of it. ARPA’s section for computer research – the Information Processing Techniques Office (IPTO) – funded the project under the direction of Robert Taylor, but the idea for such a network sprang from the imagination of that office’s first director, J.C.R. Licklider.
As we learned earlier, Licklider, known to his colleagues as ‘Lick,’ was a psychologist by training. But he became entranced with interactive computing while working on radar systems at Lincoln Laboratory in the late 1950s. This passion led him to fund some of the first experiments in time-shared computing when he became the director of the newly-formed IPTO, a position he took in 1962.
By that time, he was already looking ahead to the possibility of linking isolated interactive computers together into a larger superstructure. In his 1960 paper on “man-computer symbiosis”, he wrote that
[i]t seems reasonable to envision …a ‘thinking center’ that will incorporate the functions of present-day libraries together with anticipated advances in information storage and retrieval and the symbiotic functions suggested earlier in this paper. The picture readily enlarges itself into a network of such centers, connected to one another by wide-band communication lines and to individual users by leased-wire services.
Just as the TX-2 had kindled Licklider’s excitement over interactive computing, it may have been the SAGE computer network that prompted Licklider to imagine that a variety of interactive computing centers could be connected together to provide a kind of telephone network for intellectual services. Whatever its exact origin, Licklider began disseminating this vision among the community of researchers that he had created at IPTO, most famously in his memo of April 23, 1963, directed to the “Members and Affiliates of the Intergalactic Computer Network,” that is to say the various researchers receiving IPTO funding for time-sharing and other computing projects.
The memo is rambling and shambolic, evidently dictated on the fly with little to no editorial revision. Determining exactly what Licklider intended it to say about computer networks therefore requires some speculative inference. But several significant clues stand out. First, Licklider revealed he sees the “various activities” funded by IPTO as in fact belonging to a single “overall enterprise.” He follows this pronouncement by discussing the need to allocate money and projects to maximize the advantage accruing to that enterprise, as network of researchers as a whole, given that, “to make progress, each of the active researchers needs a software base and a hardware facility more complex and more extensive than he, himself, can create in reasonable time.” To achieve this global efficiency might, Licklider conceded, requires some individual concessions and sacrifices by certain parties.
Then Licklider began to explicitly discuss computer (rather than social) networks. He wrote of the need for some sort of network control language (what would later be called a protocol) and his desire to eventually see an IPTO computer network consisting of “..at least four large computers, perhaps six or eight small computers, and a great assortment of disc files and magnetic tape units–not to mention the remote consoles and teletype stations…” Finally, he spent several pages laying out a concrete example of how a future interaction with such a computer network might play out. Licklider imagines a situation where he is running an analysis on some experimental data. “The trouble is,” he writes, “I do not have a good grid-plotting program. …Is there a suitable grid-plotting program anywhere in the system? Using prevailing network doctrine, I interrogate first the local facility, and then other centers. Let us suppose that I am working at SDC, and that I find a program that looks suitable on a disc file in Berkeley.” He asks the network to execute this program for him, assuming that, “[w]ith a sophisticated network-control system, I would not decide whether to send the data and have them worked on by programs somewhere else, or bring in programs and have them work on my data.”
Taken together, these fragments of thought appear to reveal a larger scheme in Licklider’s mind: first, to parcel out particular specialties and areas of expertise among IPTO-funded researchers, and then to build beneath that social community a physical network of IPTO computers. This physical instantiation of IPTO’s “overall enterprise” would allow researchers to share in and benefit from the specialized hardware and software resources at each site. Thus IPTO would avoid wasteful duplication while amplifying the power of each funding dollar by allowing every researcher to access the full spectrum of computing capabilities across all of IPTO’s projects.
This idea, of resource-sharing among the research community via a communications network, sowed the seeds within IPTO that led, several years later, to the creation of ARPANET.
Despite its military provenance, originating as it did in the halls of the Pentagon, ARPANET thus had no real military justification. It is sometimes said that the network was designed as a war-hardened communications network, capable of surviving a first-strike nuclear attack. There is a loose connection, as we’ll see later, between ARPANET and an earlier project with that aim, and ARPA’s leaders occasionally trotted out the “hardened systems” idea to justify their network’s existence before Congress or the Secretary of Defense. But in truth, IPTO built ARPANET purely for its own internal purposes, to support its community of researchers – most of whom themselves lacked any direct defense justification for their activities.
Meanwhile, by the time of his famous memo Licklider had already begun planning the germ of his intergalactic network, to be led by Len Kleinrock at UCLA.
Kleinrock, the son of working class immigrants from Eastern Europe, grew up in Manhattan in the shadow of the George Washington Bridge. He worked his way through school, taking evening sessions at City College to study electrical engineering. When he heard about a fellowship opportunity for graduate study at MIT, capped by a semester of full time work at Lincoln Lab, he jumped at the opportunity.
Though built to serve the needs of SAGE, Lincoln had since diversified into many other research projects, often tangentially related to air defense, at best. Among them was the Barnstable Study, a concept floated by the Air Force to create an orbital belt of metallic strips (similar to chaff) to use as reflectors for a global communication system4. Kleinrock had fallen under the spell of Claude Shannon at MIT, and so decided to focus his graduate work on the theory of communication networks. The Barnstable Study provided Kleinrock with his first opportunity to apply the tools of information and queuing theory to a data network, and he extended that analysis into a full dissertation on “communications nets,” combining his mathematical analysis with empirical data gathered by running simulations on Lincoln’s TX-2 computers. Among Kleinrock’s close colleagues at Lincoln, sharing time with him in front of the TX-2, were Larry Roberts and Ivan Sutherland, whom we will meet again shortly.
By 1963, Kleinrock had accepted a position at UCLA, and Licklider saw an opportunity – here he had an expert in data networking at a site with three local computer centers: the main computation center, the health sciences computer center, and the Western Data Processing Center (a cooperative of thirty institutions with shared access to an IBM computer). Moreover, six of the Western Data Processing Center institutions had remote connections to the computer by modem, and the IPTO-sponsored System Development Corporation (SDC) computer resided just a few miles away in Santa Monica. IPTO issued a contract to UCLA to interconnect these four centers, as a first experiment in computer networking. Later, according to the plan, a connection with Berkeley would tackle the problems inherent in a longer-range data connection.
Despite the promising situation, the project foundered and the network was never built. The directors of the different UCLA centers didn’t trust one other, nor fully believe in the project, and they refused to cede control over their computing resources to one another’s users. IPTO had little leverage to influence the situation, since none of the UCLA computing centers were funded directly by ARPA5.
IPTO’s second try at networking proved more successful, perhaps because it was significantly more limited in scope – a mere experimental trial rather than a pilot plant. In 1965, a psychologist and disciple of Licklider’s named Tom Marill left Lincoln Lab to try to profit from the excitement around interactive computing by starting his own time-sharing business. Lacking much in the way of actual paying customers, however, he began casting about for other sources of income, and thus proposed that IPTO fund him to carry out a study of computer networking. IPTO’s new director, Ivan Sutherland, decided to bring a larger and more reputable partner on board as ballast, and so sub-contracted the work to Marill’s company via Lincoln Lab. Heading things from the Lincoln side would be another of Kleinrock’s old office-mates, Lawrence (Larry) Roberts.
Roberts had cut his teeth on the Lincoln-built TX-0 as an undergrad at MIT. He spent hours each day entranced before the glowing console screen, eventually constructing a program to (badly) recognize written characters using neural nets. Like Kleinrock he ended up working at Lincoln for his graduate studies, solving computer graphics and computer vision problems, such as edge-detection and three-dimensional rendering, on the larger and more powerful TX-2.
Up until late 1964, Roberts had remained entirely focused on his imaging research. Then he came across Lick. In November of that year, he attended an Air Force-sponsored conference on the future of computing at the Homestead hot springs resort in western Virginia. There he talked late into the night with his fellow conference participants, and for the first time heard Lick expound on his idea for an Intergalactic Network. Roberts began to feel a tickle at the back of his brain – he had done great work on computer graphics, but it was in effect trapped on the one-of-a-kind TX-2. No one else could use his software, even if he had way to provide it to them, because no one else had equivalent hardware to run it on. The only way to extend the influence of his work was to report on it in academic papers in the hopes that others would and could replicate it elsewhere. Licklider was right, he decided, a network was exactly the next step needed to accelerate computing research.
And so Roberts found himself working with Marill, trying to connect the Lincoln TX-2 with a cross-country link to the SDC computer in Santa Monica, California. In an experimental design that could have been ripped straight from Licklider’s “Intergalactic Network” memo, they planned to have the TX-2 pause in the middle of a computation, use an automatic dialer to remotely call the SDC Q-32, invoke a matrix multiply program on that computer, and then continue the original computation with the answer.
Setting aside the basic sensibility of using dearly-bought cutting-edge technology to span a continent in order to use a basic math routine, the whole process was painfully slow due to the use of the dial telephone network. To make a telephone call required setting up a dedicated circuit between the caller and recipient, usually routed through several different switching centers. As of 1965, virtually all of these were electro-mechanical6. Magnets shifted metal bars from one place to another in order to complete each step of the circuit. This whole process took several seconds, during which time the TX-2 could only sit idle and wait. Moreover the lines, though perfectly suited for voice conversation, were noisy with respect to individual bits and supported very low bandwidth (a couple hundred bits per second). A truly effective intergalactic, interactive, network, would require a different approach.[^others]
The Marill-Roberts experiment had not shown long-distance networking to be practical or useful, merely theoretically possible. But that was enough.
In the middle of 1966, Robert Taylor took over the directorship of IPTO, succeeding Ivan Sutherland as the third to hold that title. A disciple of Licklider and a fellow-psychologist, he came to IPTO by way of a position administering computer research for NASA. Nearly as soon as he arrived, Taylor seems to have decided that the time had come to make the intergalactic network a reality, and it was Taylor who launched the project that produced ARPANET.
ARPA money was still flowing freely, so Taylor had no trouble securing the extra funding from his boss, Charles Herzfeld. Nonetheless, the decision carried significant risk of failure. Other than the very limited 1965 cross-country connection, no one had ever attempted anything like ARPANET. One could point to other early experiments in computer networking. For example, Princeton and Carnegie-Mellon set up a network of time-shared computers in the late 1960s in conjunction with IBM.7 The main distinction between these and the ARPA efforts was their uniformity – they used exactly the same computer system hardware and software at each site.
ARPANET, on the other hand, would be bound to deal with diversity. By the mid-1960s, IPTO was funding well over a dozen sites, each with its own computer, and each of those computers had a different hardware design and operating software. The ability to share software was rare even among different models from a single manufacturer – only the brand-new IBM System/360 product line had attempted this feat.
This diversity of systems was a risk that added a great deal of technical complexity to the network design, but also an opportunity for Licklider-style resource sharing. The University of Illinois, for example, was in the midst of construction on the massive, ARPA-funded ILLIAC IV supercomputer. It seemed improbable to Taylor that the local users at Urbana-Champaign could fully utilize this huge machine. Even sites with systems of more modest scale – the TX-2 at Lincoln and the Sigma-7 at UCLA, for example, could not normally share software due to their basic incompatibilities. The ability to overcome this limitation by directly accessing the software at one site from another was attractive.
In the paper describing their networking experiment, Marill and Roberts had suggested that this kind of resource sharing would produce something akin to Ricardian comparative advantage among computing sites:
The establishment of a network may lead to a certain amount of specialization among the cooperating installations. If a given installation, X, by reason of special software or hardware, is particularly adept at matrix inversion, for example, one may expect that users at other installations in the network will exploit this capability by inverting their matrices at X in preference to doing so on their home computers.[^ricardo]
Taylor had one further motivation for proceeding with a resource-sharing network. Purchasing a new computer for each new IPTO site, with all the capabilities that might be required by the researchers at that site, had proven expensive, and as one site after another was added to IPTO’s portfolio, the budget for each was becoming thinly stretched. By putting all the IPTO-funded systems onto a single network, it might be possible to supply new grantees with more limited computers, or perhaps even none at all. They could draw whatever computer power they needed from a remote site with excess capacity, the network as whole acting as a communal reservoir of hardware and software.
Having launched the project and secured its funding, Taylor’s last notable contribution to ARPANET was to select someone to actually design the system and see it through to completion. Roberts was the obvious choice. His engineering bona fides were impeccable, he was already a respected member of the IPTO research community, and he was one of of a handful of people with hands-on experience designing and building a long-distance computer network. So in the fall of 1966, Taylor called Roberts to ask him to come down from Massachusetts to work for ARPA in Washington.
But Roberts proved difficult to entice. Many of the IPTO principal investigators cast a skeptical eye on the reign of Robert Taylor, whom they viewed as something of a lightweight. Yes, Licklider had been a psychologist too, with no real engineering chops, but at least he had a doctorate, and a certain credibility earned as one of the founding fathers of interactive computing. Taylor was an unknown with a mere master’s degree. How could he oversee the complex technical work going on within the IPTO community? Roberts counted himself among these skeptics.
But a combination of stick and carrot did their work. On the one hand Taylor exerted a certain pressure on Roberts’ boss at Lincoln, reminding him that a substantial portion of his lab’s funding now came from ARPA, and that it would behoove him to encourage Roberts to see the value in the opportunity on offer. On the other hand, Taylor offered Roberts the newly-minted title of “Chief Scientist”, a position that would report over Taylor’s head directly to a Deputy Director of ARPA, and mark Roberts as Taylor’s successor to the directorship. On these terms Roberts agreed to take on the ARPANET project.8 The time had come to turn the vision of resource-sharing into reality.
Janet Abbate, Inventing the Internet (1999)
Katie Hafner and Matthew Lyon, Where Wizards Stay Up Late (1996)
Arthur Norberg and Julie O’Neill, Transforming Computer Technology: Information Processing for the Pentagon, 1962-1986 (1996)
M. Mitchell Waldrop, The Dream Machine: J.C.R. Licklider and the Revolution That Made Computing Personal (2001)
- This history is described in Only Connect. ↩
- As described in Extending Interactivity. ↩
- There are brief discussions of cross-telling in John F. Jacobs, “The Sage Air Defense System: A Personal History” (1986); Redmond and Smith, From Whirlwind to MITRE (2000); Computer History Museum, “Oral History of Severo Ornstein,” November 20, 2015. ↩
- Delbert R. Terril, Jr., “The Air Force Role in Developing International Outer Space Law”, Air University Press, 1999 ↩
- This political problem points to one of the central questions in the history of the Internet. Convincing peer systems that cooperate and interconnection is in their best interest is difficult. So why, or rather, how, does the Internet exist? We’ll return to this question more than once in the essays to come. ↩
- AT&T installed its first electronic switching center that very year, in Succasunna, New Jersey. ↩
- Ronald M. Rutledge, et. al., “An Interactive Network of Time-Sharing Computers”, ACM ’69. ↩
- Many sources report on the stick, but few on the carrot, or on the reasons for Roberts’ reluctance. Waldrop, Dream Machine, 268-69, gives the full details. ↩