The Soul of a New Machine

by Tracy Kidder

  One of Wallach's posters was a blown-up reproduction of Saul Steinberg's droll cartoon depicting the United States: New York City occupies the entire foreground of the picture; California is awarded a much smaller, but significant, space in the background; and the states in between take up about as much room as the lettuce in a club sandwich. Wallach said he liked this poster for its accuracy. It also made him think of North Carolina, one of those states that got short shrift in the cartoon. Wallach had been offered the chance to go south with FHP. Indeed, he had been encouraged to go. Why had he refused?

  "I thought North Carolina sucked," he explained. "I told them, 'That's not a place where I want to bring up my wife and family.' I was given a tour of Chapel Hill and somebody told me that more people per capita read the New York Times the Chapel Hill than in all the rest of the state. I said, 'It sounds like the Warsaw Ghetto you just described.' Hey, when I wake up in the morning I want to hear the stock market report on the radio, not the price of tobacco and hogs. And what they considered good food — I wouldn't wish it on my worst enemy." Wallach sniffs, wrinkles his nose, and then delivers such lines with his chin forward, as if daring someone to hit it. "I said, 'That's not a place where I want to bring up my wife and children.'" He made me laugh. He watched, smiling.

  Carl Alsing gleefully related that once, when an engineer in the basement was describing how rough it had been to grow up in India, Wallach burst in declaiming that India was easy — that man should have tried growing up in Brooklyn, where you had to fight your way from one ethnic neighborhood to another just to get to school.

  Wallach remembered that when he was very young, another boy on his street punched him and he didn't hit back. His father punished him for this failure to retaliate. Asked to describe his old neighborhood, Wallach said: "Let's put it this way. You learn very quickly that if someone hits you, you hit them back twice as hard, so they won't hit you again."

  When I met Wallach, he was living in a residential neighborhood in Framingham, Massachusetts, in a new and prosperous looking housing development. His house was gray and clap- boarded in the Colonial style. It was pretty. Inside, it was immaculate. He sat in his living room on a white easy chair telling what amounted to old war stories. One concerned an executive whom Wallach had believed to be an enemy of the Eclipse Group. Wallach had decided to "play with the turkey's mind." For a week, everytime he ran into this man, Wallach would smile at him. The following week, however, whenever Wallach encountered the man, he would make himself look sad. Finally, the executive could stand it no longer and asked for an explanation — which was what Wallach had been waiting for.

  "Steve, how come you're so happy to see me sometimes, and then sometimes you look like you wish I wasn't there?"

  "I don't know," said Wallach. "I'm very moody."

  "Then," recalled Wallach, "I went into my office, closed the door, and got hysterical."

  Wallach's wife overheard this tale. "Steve, that's terrible," she said. "I live with this," she added, turning to me. But I noticed that she was smiling.

  While Wallach was talking in this vein, his four-year-old daughter was circling the room on roller skates, across the wall-to-wall carpeting. Considering the whiteness of his chair and his proclivity to battle, I expected Wallach to yell at her at any moment, but I soon got the feeling that he did not often raise his voice in anger while at home. When the little girl skated up behind him and planted a wet, loud kiss on his cheek, Wallach beamed. "That's why you have daughters."

  You had to understand the context in which Wallach practiced his martial arts; His swords were usually technical issues, and he left them at work. Moreover, in the last few years, he had been sorely provoked. He felt, he said, "like an animal thrown in a cage."

  In this situation, Wallach allowed, he was a rough customer. On one of his many visits to North Carolina during the so-called EGO wars, he was presented with a poster depicting the hideous face of the .arch villain of Star Wars, Darth Vader himself. "We got this just for you," Wallach was told. He liked to tell that story.

  When FHP's removal south had been announced, Wallach had felt robbed and insulted. Then he had felt astonished and dismayed when he had read the article in the Globe that seemed to have de Castro denigrating the engineers at Westborough. Wallach called up the reporter. Pretending to be an important Data I General executive, he badgered the fellow into reading his notes over the phone. On the basis of this conversation, Wallach decided that de Castro had been quoted out of context. But the story left a bad taste. Wallach heard the people who were going south with FHP say that North Carolina was where the action was and would be from now on. He became convinced that neither the defectors to North Carolina nor de Castro himself believed that the stay-behinds at Westborough would produce significant machines in the future. He took the whole affair as a personal insult. "A slap in the face." He had to retaliate. Working on EGO was his first attempt to do so. "Like I said, I'm a fighter, and the challenge was to prove the man in the corner office and North Carolina wrong."

  In the EGO wars, Wallach served as West's Hessian. West didn't feel angry toward North Carolina, nor could he afford to appear hostile. But arguments are arguments; certain nasty things had to be said. Wallach was good at saying them and happy to do it. He served, he said afterward, as "West's gun" in the "shootouts."

  But in the end Wallach got beaten. EGO was canceled — not because it wasn't a good design, but because it was too good, too much of a threat to steal the show from North Carolina. West never fully believed this, but Wallach always did.

  When West started talking, after all of that, about building a 32-bit Eclipse, Wallach sneered—-partly because it seemed to him that this was just the sort of project that North Carolina would have wished on him: an insignificant, kludgey machine that probably wouldn't get out the door. He'd rather stand in honorable defiance than work on such a thing. He was quitting the fight, he said, and soon he would leave Data General.

  Meanwhile, thinking in the fastness of his office, West decided that he had made a mistake. He had not presented Eagle in the proper light to Wallach. Over the next few weeks, he had many long conversations with the reluctant architect. He listened to Wallach's complaints. He agreed that Wallach had been treated badly.

  "But haven't you realized yet that the way to prove someone wrong is to build the right thing?" West asked, time and time again. Finally, somewhat exasperated, but also suspecting that Wallach was ready to yield, West forced the issue. "Either you do this or your job description is inoperative," he said to Wallach one day.

  The phrasing appealed greatly to Wallach, and he could see that West was probably right. Conceiving architectures was his job and Eagle was the only project around that needed an architect. But Wallach had seen too many projects canceled in spite of their merits to believe that Eagle would go out the door just because it promised to be a good commercial machine. He wasn't going that route again. Wallach wanted to talk to de Castro. This was not a completely extraordinary request; de Castro's office, Wallach knew, would usually open for an engineer who had something to say.

  As Wallach remembered it later, for those who questioned him about it, this was the gist of his conversation with the president

  "Can I be straight with you?" asks Wallach.

  De Castro nods.

  "Okay," says Wallach. "What the fuck do you want?"

  "I want a thirty-two-bit Eclipse," says de Castro.

  "Are you sure? If we do this, you won't cancel it on us? You'll leave us alone?"

  "That's what I want, a thirty-two-bit Eclipse and no mode bit." Wallach returned to West's office, and now, at long last and sniffing, he said, "Okay, Tom, one more time." . "The document's yours," West replied. "You gotta do it fast."

  Wallach went to his office and closed the door. Some months later, a careful examination of Wallach's quarters revealed many scuff marks etched in shoe polish low down along his walls, and there was a dent higher up on one wall. These were br
uises, left over from Wallach's labor on the machine.

  When he sat down, alone, in his office, Wallach reasoned that since the whole purpose of this ridiculous undertaking was 32-bit- hood — the enlargement of the Eclipse's logical-address space from 65,000 to 4.3 billion storage compartments — he might as well begin by figuring out how the compartments would be organized ("managed") and the information'' in them protected. He further decided — he called this "the methodical engineering approach" — to worry about memory management first Clearing a space on his desk, he placed a yellow legal pad in front of him and drew a picture of the standard 32-bit address — a box containing 32-bits, which looked like this:

  12 345 .32

  He began to divide up the space inside this box.

  If you imagine the computer's storage — its memory — to be a large collection of telephones, then what Wallach was doing might be described as designing a logical system by which phones and groups of phones could be easily identified — a system of area codes, for instance.

  Next to Wallach's desk stood several tall metal bookcases. Their shelves from floor to ceiling were stuffed, with looseleaf binders and fat volumes in plain covers that bore titles such as

  Parallelism in Hardware and Software: Real and Apparent Concurrency. The binders contained specifications for nearly every sort of computer ever made and for some that had never been constructed. Wallach called his bookcase "Data General's de facto library." He claimed that he had most of this library in his head. From time to time, he wheeled around in his chair and took a binder down from the bookcase. By the end of the day, Wallach had roughed out the divisions of the standard 32-bit address, and he said to himself: "Okay, great. So far I've done nothing." He gave the wall near his door a kick as he left.

  Wallach wasn't ready to admit that he was having fun. But he was back in his office early the next morning. He had a workable general plan for managing memory. Now for the question of how to protect the stored information.

  This had become one of the important questions in the computer industry. It had arisen largely because of a practice called time-sharing. In the basement of Westborough, for example, every engineer had a computer terminal — typewriterlike keyboard plus video screen — and most of these terminals were hooked up to a single large Eclipse, located some distance away, behind one of the locked doors that warned against unauthorized entry. So long as the central computer didn't become overloaded, each of the engineers, at his separate desk, could enjoy the illusion that he alone had access to the computer, though in fact they were all sharing it. It was a common sort of system. Much more exotic ones existed. Time-sharing already took place not only within buildings, but between them and across continents and oceans. Increasingly, computers communicated with other computers across vast distances.

  Many organizations stored very valuable information in their computers. In a sense, banks store their money in computer systems, and oil companies store their crown jewels, their seismic information, in computerized data banks. Publicized cases of thievery and espionage by computer were legion at this time; experts generally agreed that most electronic criminals were never caught and that many who were never got prosecuted, because an organization that had been successfully attacked either didn't want the embarrassing publicity or feared that the news would encourage other raids.

  Many people had taken a crack at solving such problems, notably a group of engineers and computer scientists at MIT who worked with money from the Department of Defense on a project called MULTICS. In the late sixties, they produced a complex plan for making time-sharing systems secure. It was a clever plan. But many experts believed that no system of protection yet devised could withstand the efforts of smart.pranksters or thieves bent on foiling it. One organization had purchased a very fine system for protecting its computerized data banks, but a determined group had cracked it. Pretending to be the manufacturer of the security system, they had sent to the organization a bogus set of revisions to the system's software. Without affecting the system's performance, these revisions left open a "trapdoor," through which the thieves were able to withdraw important information from the organization's data bank. The anecdote was one of Wallach's favorites. "I'm not gonna try and solve th« world's problems here," he conceded. He'd forget about trying to stymie the malicious user and concentrate instead on preventing accidental damage.

  Inadvertently, users of a time-sharing system could alter the contents of the host computer's memory and in this way destroy valuable data and foul up system software. Plans such as MULTICS were workable solutions to this problem.

  Wallach was proceeding almost on instinct now. In his two years of work on FHP, he believed, he had read every published description of every system for protection that had ever been devised. In the ordinary case, an architect might linger over the various possibilities for months. Wallach didn't have time for that, and he really didn't need it. Very quickly, he chose what he believed to be the simplest and best general solution, the general idea that came out of MULTICS — which DEC, as it happened, had used in the VAX. This was a system of "rings."

  Picture an Army encampment in which all the tents are arranged in several concentric rings. The general's tent lies at the center, and he can move freely from one ring of tents to another. In the next ring out from the center live the colonels, say, and they can move from their ring into any outer ring as they please, but they can't intrude on the general's ring without his dispensation. The same rules apply all the way to the outermost ring, where the privates reside. They have no special privileges; they can't move into any ring inside their own without permission.

  Analogously, each ring defines some part of the computer's memory. How does the computer mediate a user's access to these various areas? In the general case, it compares two numbers — the number of the ring to which the user has free access and the number of the ring that the user wants to get into. If the user's ring number is smaller than or equal to the number of the ring that the user wants to enter, the user is allowed to go in. But how should ring numbers be assigned to the compartments of memory? That was the question before Wallach.

  With the VAX, DEC's engineers had solved the problems of memory management and the protection of memory separately. Each compartment in memory had an address, and each compartment also had a separate ring number, which was produced and checked by a special set of hardware. Wallach had studied the specs on VAX. He didn't like that approach. From the back of his mind came the recollection of a conversation he'd had at a conference of engineers from different companies some years before. At such convocations the usual way of getting acquainted is to ask other engineers what projects they've worked on. Wallach remembered an engineer telling him about a ring system he'd thought up and never built, in which ring numbers and addresses were mingled. Wallach had gotten the fellow to send him a spec; it had looked a little clumsy to him. But there was the germ of some thing there. Wallach drew another diagram of the standard 32 bit address:

  I 2 3 4 ,32

  segment address

  The first three bits of the address would contain the segment number of a memory compartment — in the telephone analogy, a given compartment's area code. The other bits would define the rest of the address; but Wallach wasn't interested in them just now. He was pondering the first three bits. Suddenly, without thinking about it, he was drawing another box below the first box. The diagram now looked like this:

  I 2 3 4 .32

  segment address

  ring number address

  The segment number — the area code — would be the same as the ring number, which defined the level of security to which the compartment would be assigned. Three bits can be combined in eight different ways. So there would be eight rings (eight levels of security) and eight segments (eight area codes) in the memory system. The area codes would themselves indicate which ring was forbidden to whom.

  Although they are generally shy about claiming to have had one, engineers
often speak of "the golden moment" in order to describe the feeling — it comes rarely enough — when the scales fall from a designer's eyes and a problem's right solution is suddenly there. The chief virtue of Wallach's scheme was its simplicity. It would be relatively cheap and easy to implement in hard ware and software, and it should work efficiently and reliably. When Alsing saw Wallach's brief description of the plan, he said to Wallach, "That's nice." Later, out of Wallach's earshot, he said more. "Rings have been around. They're old hat. What makes Wallach a good Data General engineer is that he came up with a really elegant subset of those ideas — simple, sweet, cheap, efficient, clean. And I can't believe I just said that about Wallach."

  As for Wallach, after he had drawn the diagram, he stared at it, wondering for a moment, "Where did that come from?"

  He kept eyeing it. "That looks pretty cool."

  For the next couple of days Wallach played with the idea, until he was sure it would work. Then he pulled his computer terminal in front of him and wrote up a memo, which told how others had handled memory management and protection in the past and which described his own general scheme. Near the end of the memo, his sarcasm rose. He described the plan as a kludge, but a better one than DEC had put into VAX. He kicked his walls a few times. The idea of placing that neat, clean structure on top of the outdated structure of the Eclipse repelled him. It was as if he had invented a particularly nice kind of arch for the doorway of a supermarket.

  Wallach was a scholar of computer architectures. He knew by heart the works of the Michelangelos, the Frank Lloyd Wrights and the Gaudis of the computer profession. He imagined himself standing in front of a roomful of experts when Eagle was all done. They would question him about the architecture, inserting sharp little knives in his flank and then twisting them. "Why didn't you do it this way, Steve, when it's obviously better?" And, as he imagined it, his only possible defense would be to plead weakly and traitorously that the company had prevented him from really showing his stuff.