The Soul of a New Machine

by Tracy Kidder

  One of the other diagnostic programs has produced a failure.

  "Oh, no."

  "We didn't do it," says Holberger. "We didn't do it right."

  He and Veres go to lunch. Holberger feels ill and picks at his food. When they return, they hook up a trace on the new failure, but without much enthusiasm. They take some pictures. It looks as if something complex is going wrong, but they can't immediately tell what, and they don't feel like opening up another long search right away. In part, their feeling stems from weariness. But there must also be something instinctive about their reluctance to dig into the new problem. For it seems they've forgotten to do something basic, and in a moment, Veres remembers.

  Veres takes out the NAND gate. He runs the failing program. The machine still commits the new failure. So it's not the NAND gate that did it. Greatly relieved, smiling now, Holberger points out that they have the IP board out on the "extender." The board is hooked up to Gollum but it's sitting in a small frame of its own, outside of the main frame of the machine. This is standard debugging practice, but boards aren't made with extenders in mind, and in some cases a perfectly good board won't work while out on an extender. With nimble fingers, they put the IP back in its proper place among the other boards, and the failure no longer occurs. Now Gollum successfully negotiates all of the basic Eclipse diagnostics, including Eclipse 21.

  They've reached a milestone, but one that they thought they had reached before. There's no celebration, no sitting around Gollum with their feet up on analyzers, savoring the victory, rehearsing the battle. Plenty of diagnostics stand before them. Much trickier ones, in fact.

  "A feeling of accomplishment" is what Veres says he has. "But then again, there's lots more feeling of accomplishment to go."

  SHORTER THAN A SEASON

  In the lab, at the bench along the far wall, the other members of the night shift are bending low over wire-wrapped boards.

  "Oh no. Not me. I did that last night," Josh Rosen has said to Ed Rasala. Now stillness has fallen over the room. The prototypes' cooling fans, the crickets of this place, drone on and on. Rosen stands all alone next to Coke. He looks rather small among the machinery. Rosen has very black hair, cut short enough to withstand a first-sergeant's inspection. His complexion, though naturally swarthy, still manages to reveal that he has been spending his days indoors. He is wearing corduroy pants, a plain cotton shirt with a button-down collar and no necktie. He often wears a sports jacket to work, even in the lab, and his black, laced shoes and the upper quadrant of a white T-shirt showing at his throat make his dress seem more conventional than a typical Hardy Boy's uniform. He could by his looks be any young age, from late adolescence to his mid-twenties; in fact he is twenty-four. Periodically, he raises one hand to his mouth and nibbles at his nails, while he works with the other.

  Rosen, who designed the board called the ALU, is trying to get h to perform addition. It is not a large exaggeration to say that everything else in a computer exists in order to bring information swiftly to the ALU for manipulation; and for the ALU, adding is the mechanical equivalent of breathing. But this evening, whenever the diagnostic program has asked the ALU to add two packets of bits, the ALU has sent out a wrong answer and then performed a series of incomprehensible actions. "It goes," as Rosen likes to say, "to never-never land." At this moment, he is trying to take a picture with the logic analyzer of what's going wrong inside.

  A straight white line runs horizontally across the little blue screen of the analyzer. From a cabinet in a corner, Rosen gets an object almost exactly the size and shape of a 45-rpm record and he inserts this "floppy disk" into the tall disk-drive machine that stands nearby Coke. Lights flash immediately on the disk drive. Rosen turns to the console and types a short message. At once the console starts typing by itself. A scratchy sound, which lasts just a moment, then stops. Rosen bites absentmindedly at the nails of his left hand as he leans over and reads what the console has typed. Nails still at his lips, he turns to the analyzer.

  Something has happened. The straight white line that was running across the little blue screen has rearranged itself into a jagged shape, like a diagram of two teeth on one side of a zipper. Rosen is staring at the picture, his nails raised to his mouth. Slowly, still staring, he rotates his hand and takes most of his knuckles in his teeth. For a long moment, he holds this position, frozen like the image on the screen.

  It might be a painting of a nightmare by Goya. Your eye is drawn from the young man's face and the hand resting in his teeth, to the jagged line on the screen, which is in fact a picture of an electronic event that took place, in infinitesimal time, just a moment ago. Though it is a common sort of picture, often seen in the lab, all of a sudden it has become dreadful. But who can say why?

  From the moment they started designing the computer, engineers were dropping out The reasons varied, from the feeling that SHORTER THAN A SEASON 213 the machine would be a kludge, to disappointment over positions in the group's pecking order. Some may have tired of the competition within the group, of what Ken Holberger called the peer pressure: "If I screw up this, then I'll be the only one, and I'm not gonna be the only one." Some may have had trouble keeping up with the others. A few did not participate in the group's social life, and some seemed to drift away from the project. Building Eagle wasn't the best of times for everyone.

  Rosen came to the group in the middle of the summer of 1978 and went to work on the all-important ALU. He felt constrained to start designing right away, before he could really study the architectural spec — before, indeed, a complete spec existed. A few months later, in August, he decided that he had chosen to use the wrong sort of chips. He told Rasala he wanted to redesign the whole board. Rasala replied, 'There isn't time." In effect, Rosen felt, Rasala was saying: 'This'll probably work. Put a Band-Aid fix on it."

  In December Rosen brought in a design that called for far more chips than it was supposed to contain. West assigned another engineer to examine Rosen's work — a necessary act from West's point of view, but a form of censure for Rosen. A while after that, Rosen underwent his review, a periodic ritual at Data General in which your boss evaluates your performance and sometimes gives you a raise. He was handed a report card less flattering than he was used to. It was the only unflattering one he had ever seen in his brief but distinguished career as a builder of computing equipment.

  Rosen felt that West and Rasala were treating him unfairly. "They backed me into a corner." Certainly, they gave him a challenging task: he was to make an ALU that would perform certain kinds of arithmetic faster than VAX and yet occupy only one board. Bob Beauchamp would say later on that this just couldn't be done. In fact, West and Rasala eventually came to the same conclusion. They decided to sacrifice certain features and keep the ALU on one board. They had Rosen proceed. In any project such compromises inevitably occur, and in the end, according to Beauchamp, the ALU turned out nicely. "I think Josh did a damn good job on that board personally. There are some very slick things in that design," said Beauchamp — and he was in a position to know, because he had written some of the microcode that primarily ordered the ALU around and because, at the time he offered his opinion, he was himself designing an ALU for another computer.

  When the Eclipse Group finished designing Eagle and began to debug, West told me: "Josh is doing all right. He's hangin' in there." West never said that to Rosen. But a few kind words just then could not have cured what ailed the young man.

  Rosen had come over from Data General's Special Systems Division, which produces equipment for customers' special needs. "I was the star at Special Systems," he said. "I got all the sexy jobs. I went to the Eclipse Group and I wasn't treated like a star." Ken Holberger, about the same age as Rosen and endowed with roughly the same amount of experience, occupied that position among the Hardy Boys, if anyone did. Clearly, Rasala considered Holberger to be the team's stellar designer. Rosen, by his own account, found himself competing with Holberger. He wanted to be "the driving
force" behind Eagle's hardware. He was used to being in control of entire designs. Over at Special Systems, he had often felt free to pursue pure technical excellence. Three weeks after joining the Eclipse Group, he said to himself, "This is all wrong."

  Over at Special Systems, Rosen had felt that his boss was also his friend. I remember following West into the lab one evening early in the debugging. Bound on some urgent mission, he strode briskly. Rosen turned around. He saw West coming. Grinning, he intercepted West, saying: "It does an increment now. So we know that it can add." West did not seem to hear him. He flicked out his hand as if brushing something away and strode on. 'The man never said hello to me after he hired me. Literally never said hello!" Rosen cried. "I wanted to feel I was part of what was going on, but with West the only channel of communication was through Rasala, and Rasala, I felt, was a pretty narrow filter. Everything was on a need-to-know basis." Mushroom management came as a shock to him. "I felt I had no more rights than an oscilloscope."

  Rosen described his own habits of thought as eclectic. He liked to think up fresh approaches, and he liked to talk about them. Something of a perfectionist, he felt offended when he was told to perform quick-and-dirty repairs. Rasala, who had the job of enforcing haste, felt that Rosen might design and redesign the ALU forever if he let him. He took seriously Rosen's casual, digressive talk about other ways the thing might be done, and he really dreaded it. "We can't think about that," he would often snap when Rosen came to him with some new idea. Rosen would walk away to contemplate Rasala's "sheer rudeness." They had almost nothing in common. One was big, gruff, athletic, and intent on getting the machine out the door on time. The other was of a more delicate sensibility, and eager to make comely designs.

  Rosen had grown up near the University of Chicago. He did not care for sports. "I have a sort of very antiathletic point of view. That's the sort of thing nice Jewish boys don't do. I was captain of a softball team in college and we used to get beaten by fifty runs and be proud of it." He said, "I'm not much of a team player, I guess." This was a defect in his character, he felt, but only in the strange context of the Eclipse Group. He did not hide his attitude. That made Rasala angry. He spoke gruffly to Rosen. Rosen's interest in working on others' parts of the machine, never high, declined still more.

  Rosen belonged to the generation for whom computers made up part of the scenery. He himself might have had block diagrams of ALU's encoded somewhere in his genes. Like practically everyone else on the team, he started becoming an engineer at about the age of four, picking on ordinary household items such as lamps and clocks and radios. He took them apart whenever his parents weren't looking. At ten he turned to rockets. First he made them out of match heads. Then he experimented with more powerful fuels. At twelve, he got some gunpowder in the mail from a firm of dubious repute and concocted his most powerful missile. He set it off electrically, from his "blockhouse," a basement stairwell. The rocket ascended some distance and exploded, loudly. A few minutes later, he saw a police car turn into the alley behind his home. "I gotta go visit grandma in Sheboygan," he told his parents that very evening.

  He went to an extremely competitive high school, the University of Chicago's Lab School. "If you had seven hundreds on your College Boards, you were at the low end of the class." In college he had no confidence until, thinking he'd like to have a stereo — and being "cheap," as he put it — he decided to learn how to make one himself, and to that end enrolled in a course in basic electronics. "I totally crushed the course," he remembered. He became a physics major. For his senior-year thesis he began to build a device called a floating-point processor, and suddenly he was getting A's in all his courses. "No one knew what I was doing. No one understood it. They thought, 'Hey, he must be something.' "

  The little processor was his first concoction in computers, and he remembered it fondly, as some remember old girlfriends or certain football games. "It wasn't a bad little box at all." He felt gratitude toward that first processor, because it enabled him to graduate magna cum laude.

  Rosen went on to Northwestern, to earn his master's degree in electrical engineering. He spent his summers building electronic gear. While still in school, he built a pattern recognition processor for Fermi Labs, worked on an earth station for Fairchild Space and Electronics, and designed a signal processor, also for Fairchild.

  Data General recruited Rosen by promising him interesting work, and he got it. The first truly commercial product he designed was something known as a cluster controller, a kind of computer terminal, which he called Hydra. After it had been built and shipped, a small microcode error cropped up in it, and Rosen's boss sent him to California to make the repair. Out west, Rosen was ushered into a room and saw a dozen people using the machine that he had designed. The sight made him tremble. It took his breath away. He felt scared. "My God!" he thought. "Don't use one of those. Why don't you use a real terminal?" He felt thrilled. "That's something I designed. That's my machine. That's not Data General's. That's me.

  "You don't get to see that very often," Rosen said. "But that's the biggest satisfaction of all."

  He was only twenty-two, and he had done it all — except that he had never helped to build a commercially important, big and brand-new computer. He expressed an interest in doing so, and the word got back to the Eclipse Group. They recruited him; he had fine credentials. Then, of course, everything went sour for him. But perhaps his personal catastrophe had started earlier. Maybe he volunteered for Eagle looking for a way out of a malaise already upon him. He thought that was probably the case.

  Rosen said that when he first came to Data General a few years before Eagle, the staffer at Personnel told him, "We know how you fellows work, and we will remind you if you forget to take your vacations." But, he said, they never did remind him. Probably, it wouldn't have mattered if they had. He went to work at Special Systems, and in his first year there, he was assigned so many important, challenging projects that he not only forgot to take his vacation, he also failed to take a weekend off. What would obtain in the Eclipse Group also held at Special Systems. "There was no question of deadlines. You'd already missed it, whatever it was." He worked many eighty-hour weeks — without extra pay, of course, but that wasn't the issue. "I had a lot of control over the things I did, and the price was a lot of pressure. If I spent only a sixty-hour week, I felt intensely guilty."

  He told himself that he was having the time of his life. During his second year at Special Systems, he began to remind himself of this with some regularity. "Josh," he would say to himself, "you're designing the sexy machines."

  The dialogue with himself continued when he joined the Eclipse Group and began working on Eagle. "You've always revered the people who built the NOVA and the PDP-11. Now you're one of them. You're the guy you always wanted to be," he said in his mind.

  "So why," he asked himself, "am I not happy?"

  By the time I saw Rosen standing in front of the logic analyzer in the lab, he did not have to ask himself that question anymore. He did have some friends in the company, but few of any sort outside of work. How could he have made any? He had spent nearly half of all the hours of all the days and nights of the last three years at work.

  His experience had made him feel that Data General, more than any of the other companies he'd heard about, exploited "child labor—kids right out of school." "It's something of a sweatshop. It's expected that you'll ruin your health for the company." But he knew it wasn't that simple. He went on: "It's sort of self-imposed. Everybody's trying to prove themselves. Eventually you burn out."

  Now he thought to himself: "I have no social life. Nothing." He looked back on his career and saw that ever since adolescence he had never strayed very far from work. He needn't have taken a job every summer; his parents would have given him spending money. But that was what he had done. "I'd been doing this all my life, it seemed. In college, you know, physics majors are masochists and proud of it. Constantly pulling all-nighters in the lab or the c
omputer centers. But you find yourself becoming really sort of narrow." When he went to some of the parties that members of the Eclipse Group threw, he found himself and most everyone around him talking about computers. That was nothing new, but now he also found himself thinking: "This is a party. We're not supposed to be talking about work."

  He suspected that Rasala and West were suffering from the same disorder as he, their irritability and rudeness toward him the symptoms of their burning out. As for himself, he had no doubt. As the debugging continued, he felt the pressure in his stomach. It hurt every day. This sort of work, even the occasional bad stomach, used to be fun. "Part of the fascination," he said, "is just little boys who never grew up, playing with Erector sets. Engineers just don't lose that, and if you do lose it, you just can't be an engineer anymore." He went on: "When you burn out, you lose enthusiasm. I always loved computers. All of a sudden I just didn't care. It was, all of a sudden, a job."

  Rosen went on working, through the winter and spring, peering into Coke with logic analyzers and doing a creditable job of repairing errors in his own board. But debugging Eagle had long ago stopped being fun, and from time to time, when Rasala teased or reprimanded him, or when he simply couldn't face that tangle of wire and silicon anymore, he would take advantage of Carman's escape valve and go away from the basement for an afternoon or an evening. Sometimes, before he left, he would leave a note behind, on top of the terminal in his cubicle — usually a faintly humorous note that would nonetheless alert his friends to what had happened in case he didn't come back this time.

  With an old friend, Rosen had once visited what he called "a very liberal arts college" in Vermont. He was strolling through something known as "an alternative-energy farm," when a young woman, bare to the waist, walked by. "She was," Rosen said, "a miracle of biological engineering." He continued: "I was so stunned that I walked into the door of a geodesic dome. Although blood was pouring down the bridge of my nose, I was completely oblivious to it." Now, one day back at Data General, his weariness focused on the logic analyzer and the small catastrophes that come from trying to build a machine that operates in billionths of a second. On this occasion, he went away from the basement and left this note on his terminal: