Where Wizards Stay Up Late: The Origins of the Internet

by Katie Hafner

  The chairman of the group, Abhay Bhushan, an MIT graduate student and systems architect, was an expert in Multics, an ambitious and complex operating system. Bhushan had studied problems of multitasking in a single computer. The next step was to eventually transfer blocks of data in a multicomputer system like the ARPA network.

  In the six months it spent working on the file-transfer protocol, the team usually met face to face in regular sessions of the Network Working Group. But it also frequently used real-time computer teleconferencing. Members of the team would all log on at once, from Palo Alto and Cambridge and L.A. and Salt Lake, and for an hour or two at a stretch trade comments back and forth. Conversing through keyboards and terminals was less spontaneous than speaking, but Bhushan believed it forced clarity into their thinking. It also had the advantage of creating a record of their work. In early July 1972, the final touches were put on FTP, and Jon Postel, now the editor and distributor of the Requests For Comments, released it as RFC 354.

  Showing It Off

  The only real problem with this network now was load. That is, there wasn’t much of it. At first, traffic was light because the protocols were late. Now, on a normal day the channels were virtually empty. In the fall of 1971, the network carried an average of just 675,000 packets per day, barely more than 2 percent of its capacity of 30 million packets a day. The UCLA Network Measurement Center continued generating test traffic to probe the weaknesses of all the network algorithms. But there wasn’t enough natural traffic on the network to really push the limits of the routing and anticongestion schemes.

  There were some interesting early uses. Programmers at SRI were using Utah’s PDP-10 compiler in preparation for installing their own PDP-10, and they generated the most actual traffic. Jon Postel at UCLA was using the network to run SRI’s oNLine System.

  Bob Metcalfe, a Harvard graduate student working at MIT, and a friend, Danny Cohen, who taught at Harvard, did one of the more exciting early experiments over the network. Metcalfe and Cohen used Harvard’s PDP-10 to simulate an aircraft carrier landing and then displayed the image on a graphics terminal at MIT. The graphics were processed at MIT, and the results (the view of the carrier’s flight deck) were shipped back over the ARPA network to the PDP-1 at Harvard, which also displayed them. The experiment demonstrated that a program could be moved around the network at such high speed as to approximate real time. Metcalfe and others wrote up an RFC to announce the triumph and titled it “Historic Moments in Networking.”

  Others weren’t so successful. There was an early attempt to devise something called a data-reconfiguration service; it was a failed attempt to write a programming language that would interpret incompatible files on different computers.

  The ARPA network, however, was virtually unknown everywhere but in the inner sancta of the computer research community. And for only a small portion of the computer community, whose research interest was networking, was the ARPA network developing into a usable tool. To them it was a fantastic experiment—but you had to be involved in things like queueing theory or communications theory to appreciate it. If, on the other hand, your work was in artificial intelligence or robotics or computer graphics or almost anything else the community was investigating, the utility of this grand transcontinental packet-switching system had yet to be realized. No one had come up with a useful demonstration of resource-sharing; the protocols to make it work weren’t even in place. The ARPA network was a growing web of links and nodes, and that was it—like a highway system without cars.

  Yet ostensibly throughout the community there were rich resources to be shared. Carnegie-Mellon had a premier artificial-intelligence department and unique applications attendant to it. Stanford had an extraordinary robotics group. Theoretically, at just about every site some unique capability could be contributed to the network.

  If the network was ever going to become anything more than a testbed for the artificial traffic generated by the Network Measurements and Network Control Centers, word of its potential had to spread. Larry Roberts knew it was time for a public demonstration. Roberts was on the program committee of the first International Conference on Computer Communication, to be held in Washington in October 1972. He circled the date and called Bob Kahn, who was still at BBN, and asked him to organize a demonstration of the ARPA network as the sole exhibit at the meeting. The conference was about a year away. Roberts asked Kahn to start planning immediately. Kahn had in fact already been planning to leave BBN and go work for Roberts at ARPA. But both men decided it would be a good idea for Kahn to stay at BBN for a while to plan the demonstration.

  Kahn’s first move was to recruit Al Vezza, of MIT’s Project MAC, to assist him.Vezza always made a good impression. He was sociable and impeccably articulate; he had a keen scientific mind and first-rate administrative instincts. Between the two men there probably wasn’t a key computer project in the U.S. research community they didn’t know about or a key player they couldn’t persuade to join them.

  In mid-1971, Kahn and Vezza called a small group of about eight principal investigators from around the country to come to a meeting at MIT’s Tech Square in Cambridge. They presented the idea of a highly accessible, engaging demonstration of the community’s most interesting resources—accessible over the network. Vezza knew it would have to be a live, interactive demonstration if it was going to have any impact. Someone at the meeting argued strenuously in favor of a videotaped presentation, to ensure against computer crashes during the show. Vezza was incredulous and argued just as strenuously that anything but a hands-on, live demonstration using actual equipment and software would signal uncertainty and potential failure for the whole ARPA network experiment. It had to be done in real time, it had to be something that could be touched and controlled by anyone just sitting down at a terminal. It was a gamble. There still wasn’t much operational experience with the Telnet and File Transfer protocols, which the participants would have to employ. And this increased the risk that the demonstration would fail. But if the demonstration succeeded, it would prove the network was not only real but useful.

  For the next nine months, Kahn and Vezza traveled the country on Roberts’s budget. “There were lots of blind alleys,” Vezza recalled. They met with dozens of vendors in the computer industry, asking each to participate by bringing its own terminals to the Washington Hilton, where the meeting would be held and where a TIP would connect them to the ARPA network. Roberts was arranging for AT&T to bring two 50-kilobit lines into the hotel. The plan was to run demonstrations on as many machines as possible, connected to as many sites as possible. Organizers would invite conference-goers to come in, sit down, log on, and use the network’s resources.

  Dozens of meetings took place at the various network sites to design interesting scenarios. Teams of graduate students and principal investigators signed on. And almost as soon as they did, they began to feel a certain panic. To pull this off, they would have to step up their efforts to finalize the unfinished network tools and protocols. Roberts, ever so correctly, had foreseen the likelihood that scheduling a highly visible public demonstration for October 1972 would build pressure on the community to mobilize and make sure the network would be functioning flawlessly by that date. Kahn too acknowledged that the demonstration was “set up to force the utility of the network to occur to the end users.”

  In the fall of 1971, BBN had a prototype TIP running at 50 Moulton Street. Two other TIPs were operational elsewhere in the network, at this time consisting of just nineteen nodes. The TIPs, based on the Honeywell 316, were completely compatible with all of the older 516s. By early 1972 several additional 316 IMPs and TIPs had been installed and the central part of the network between the East and West Coasts was beginning to fill in. By August 1972, a third cross-country line had been added. In addition to IMPs scattered around the center of the country, there were now clusters of IMPs in four geographic areas: Boston; Washington, D.C.; San Francisco; and Los Angeles. As the ICCC demonstration app
roached, there were twenty-nine nodes in what was now being widely referred to as theARPANETor, more often, just the Net.

  And people at the sites were madly scrambling. Dozens of major contributors from throughout the U.S. academic and research communities had become involved. Intensive efforts were under way to debug applications and get host computers up and running in time for the public demonstration. Every terminal manufacturer had been invited to prove its equipment could work with theARPANET: They were lining up to show more than forty different computer terminals in the demonstration. Vezza negotiated with a local vendor in the Washington, D.C., area who agreed to lend a large section of antistatic, raised computer-room flooring for installation in the meeting room at the Hilton where the TIP and terminals would be located. AT&T promised it would come through with the data link. Getting such a circuit installed anywhere in less than six months was no small issue, and it certainly wasn’t a small matter for AT&T to have that line into the Hilton as the ICCC approached.

  Several days before the meeting, the networking equipment and people began arriving at the hotel. Kahn and Vezza had drawn up a floor plan. The TIP was located up on a section of raised flooring in the center of the large meeting room. Around the perimeter of the room were to be the dozens of terminals, virtually no two alike. It would take a couple of days for all the gear to be moved into place, connected, powered up, and checked out. In a matter of hours, the room was a teeming tangle of wire and people speaking technical jabber. Technicians were stretching cables everywhere. Members of Heart’s team were all around, tools in hand, deeply engaged in helping the various terminal manufacturers modify the connector cables on each of the multitude of terminal devices, so that each could be connected to the TIP. Hours were spent stripping wires, rewiring the connectors, reconnecting, testing, and debugging.

  Many of the participants were working at a fever pitch. Many had packed up while still finishing their projects and came to Washington to add the final touches. It was the first time the whole community showed up in one place at one time. “If somebody had dropped a bomb on the Washington Hilton, it would have destroyed almost all of the networking community in the U.S. at that point,” observed Kahn. Not to mention the international community, for even Donald Davies, father of the term “packet-switching,” had come over from England to see how this would all work out. “It was just an amazing experience,” said Vint Cerf. “Hacking away and hollering and screaming and saying, ‘No, no . . . you got this one wrong. ’Getting all the details right.”

  At the end of Saturday (the conference opened on Monday), the BBN TIP was like a king on a throne of wire running to all corners of the room. AT&T had done its job and turned up at the right moment with the right line. Sunday was another frantic day of preparation, but now the TIP was in action, so people were starting to run programs and do their final checkouts. Many didn’t finish until late Sunday afternoon, just before a preview demonstration was scheduled for a group of VIPs—a Washington coterie of congressmen, Pentagon officials, and others. At about six o’clock in the evening, minutes before the doors were to open,Vezza was standing near the TIP when Metcalfe said, with no faint urgency in his voice, ”We’re losing packets!”

  Vezza shot a look to McKenzie, who was standing right there: “Alex, what changed?” McKenzie reached for the hot line to Cambridge and shouted into the phone, “Get it out! Get it out!” The Network Control Center had been watching and monitoring a slightly glitchy line in the network for the past few days. They thought they’d solved the problem that afternoon and had added the circuit back into the network. Within thirty seconds of McKenzie’s call, the link was removed by the operators at the NCC and packets were flowing smoothly at the TIP again.

  BBN’s remote-management technology had never had a finer moment. Later that evening, Jon Postel was in the exhibition room sitting at a keyboard, logged on to the host at UCLA. His team had designed a demonstration in which someone in Washington could pull up a file in Boston via the host computer in Los Angeles. The idea was to then print out the file in the exhibition room at the Hilton. When Postel shipped the file to the printer sitting beside him, nothing happened. He looked around the room. There were a lot of other demonstrations, one of which was a small robotic turtle built at MIT. The turtle was built to demonstrate how a computer program could be written to direct the motion of a machine. Kids could write their own programs in the LOGO language that said, “go left, go right, go forward, back up, move sideways,” and when the program was run, the turtle would do that. At the moment, however, the turtle was jumping up and down, twitching and jerking crazily. Instead of sending Postel’s file to the printer, the system had accidentally sent it to the turtle port, and the robot dutifully offered its interpretation of what it took to be motion commands.

  As an enthusiastic graduate student, Bob Metcalfe had undertaken the task of writing a booklet to accompany the demonstrations. It described nineteen scenarios for using theARPANET, listed resources at various sites, and showed how to log on to a remote host, how to gain access to one of the applications, and how to control a program or engage in some kind of interactive communication over the network. There were several chess games, an interactive quiz about the geography of South America, a way of reading the Associated Press news wire over the network, and many other games, tools, and demonstrations. One of the more practical applications simulated an air traffic control scenario in which responsibility for monitoring an airplane flight is automatically handed off from one computer to another, representing different air traffic control centers, as the plane crosses geographic boundaries. Metcalfe’s scenarios book was designed to walk participants, most of whom knew little about theARPANET, through each demonstration step by step.

  On Monday morning, theARPANETcomputer scientists eagerly awaited their public. When curious conference-goers approached, the network guys, like Jehovah’s Witnesses handing out copies ofThe Watchtower, thrust Metcalfe’s scenarios book into their hands and ushered them into the room. Although it was possible to follow the instructions, to all but the initiated the scenarios book was fairly incomprehensible, and it was easy to foul up the system. One man sat down in front of a terminal and typed in an instruction from the book. For some reason or other, the host he was trying to reach wasn’t functioning, or he miscued the thing. The message came back: “HOST DEAD.”

  “Oh, my God. I’ve killed it!” he exclaimed. He wouldn’t touch a terminal after that. Other funny things happened. Two people had logged in to the University of Utah. One saw that somebody else he knew but had never met was logged in. They were in talk mode, and so he typed, “Where are you?” The other replied, “Well, I’m in Washington,” “Where in Washington?” “At the Hilton.” “Well, I’m at the Hilton, too.” The two turned out to be only a few feet from each other.

  Some things weren’t so funny. As the author of the scenarios book, Metcalfe was chosen to take ten AT&T executives on a virtual tour of theARPANET.It was an odd sight: Young Metcalfe, with his big red beard, showing ten AT&T pinstriped businessmen around the network. In the middle of the demonstration the computers crashed. It was the first and only time the computers went down. The phone company executives’first reaction was to laugh.

  “I looked up in pain,” said Metcalfe, “and I caught them smiling, delighted that packetswitching was flaky. This I will never forget. It confirmed for them that circuit-switching technology was here to stay, and this packet-switching stuff was an unreliable toy that would never have much impact in the commercial world, and now they could go home to New Jersey. It was clear to me they were tangled up in the past.”

  Had they looked beyond the luckless Metcalfe and the failed demo, the AT&T executives would have seen the exuberance in other corners of the room. Not only did packetswitching work but it made wondrous things possible.

  Some of the most ingenious demonstrations involved English-language conversational programs. These were elaborate programs constructed to engage a u
ser in a verbal dialogue with a machine. There were four programs on display, two of which offered an especially fascinating glimpse into interactive computing.

  PARRY, the first of these virtual conversationalists, mimicked the belief system of a paranoid psychotic. He held up his end of a conversation by offering canned responses to statements he thought he understood. Otherwise, his response was noncommittal. PARRY was the brainchild of Dr. Kenneth Colby at Stanford University.

  The Doctor was another conversational program, based on ELIZA, a natural-language program written by MIT’s Joseph Weizenbaum. The Doctor, which mimicked the behavior of a psychiatrist, was rather more diffident than PARRY. The Doctor accepted statements, then tried to produce new sentences based on the words and syntax of the other conversant. While the resulting conversation was altogether banal, users were often struck by how lifelike the quality of their conversations with the Doctor actually seemed. “People would sit down and interact with this thing,” observed Cerf. “And sometimes they’d ask other people to leave the room, because they thought they were having a private conversation.”

  Cerf and others had toyed with the idea of setting up Colby’s paranoid to have a “session” with the psychiatrist. Just a few weeks before the ICCC demonstration, PARRY indeed met the Doctor for an unusual conversation over theARPANET, in an experiment orchestrated at UCLA. It perhaps marked the origin, in the truest sense, of all computer chat. There was no human intervention in the dialogue. PARRY was running at Stanford’s artificial-intelligence lab, the Doctor was running on a machine at BBN, and at UCLA their input and output were cross-connected through theARPANET, while the operators sat back and watched.

  “Good evening,” said the Doctor. “Tell me your problems. Please terminate input with a period or a question mark.”