The Soul of a New Machine

by Tracy Kidder

  Rasala stopped in at Jon Blau's station, in front of Coke. Blau is dark-haired with the smooth, clear skin of a child; but large black circles under his eyes stood out in sharp contrast to his white skin. He looked like a raccoon. Rasala sat down and visited with Blau for a while.

  Blau had become a Hardy Boy after Josh Rosen quit. Blau had been the logical choice to take over the debugging of the ALU, because he had written much of the microcode for the arithmetical instructions. It had been hard for him, though. "I was terribly excited by it, then very frightened," he said. "I knew what was going on with the microcode, but with the hardware, everyone is ten miles ahead of me, so I've got to run as hard as I can to catch up. Plus, I've got to keep pace. I'm keeping a note pad because people ask me questions and I think 'I should know the answer to that.'"

  Rasala was sitting cowboy-style, his chair reversed, his arms resting on its back. He was facing Blau and as Blau talked, Rasala seemed to study his face.

  Blau had presented to Rasala a problem of management at least as severe as Rosen had. No sooner did Blau start work on the ALU than he became so frightened that he had to take a week off. Blau also liked to work, said Rasala, "from one p.m. to whenever." He kept what Rasala called "good Alsing hours," and this habit created scheduling problems. They had four prototypes now, but many people needed to use them. Scheduling "machine time" was tricky, and an engineer keeping quirky hours could disrupt everything. Moreover, the bugs now tended to be synergistic; someone working on a problem in the IP might need to have the ALU's expert around.

  And Blau was at least as different from Rasala as Rosen; Blau was known as the introspective one, the philosopher of the group. He didn't fit Rasala's definition of a Hardy Boy exactly. "He's not as tough as the rest of us are," said Rasala. "He's a sensitive guy. He's a good guy. So if someone comes into his office, he'll tend to let them talk for hours, no matter how dumb their questions are." Nevertheless, Rasala did not speak brusquely to Blau or tease him roughly. They had what Rasala called "heart-to-hearts":

  "Well, here's my problem with you, Jon."

  "Well, Ed, here's my problem with you."

  If you saw them talking together, you might have thought that some of Blau had rubbed off on Rasala. Or vice versa. Or both.

  When Rosen left, Rasala went around for weeks saying, "I failed," and "If only I'd had more time ... a guy as smart as Josh." No manager wants to lose people; it doesn't look good. But Rasala's reaction had seemed to border on grief. He had read self-help manuals and shaved off his beard, and he looked less imposing without it. Nowadays, it seemed, he spoke wistfully on such topics as his bouts with diagnostics and his fear that there would be no stock options for "the kids."

  One day around this time Holberger came storming into Rasala's office, complaining about Jim Guyer. Guyer, it seemed, refused to stop working on a certain problem of low priority. Holberger and Guyer were like brothers, always arguing. "I'm gonna get Guyer off the machine," said Holberger.

  Rasala just listened calmly and nodded and nodded. Finally, Holberger said: "Ahhh, let Guyer do what he wants. It doesn't really matter."

  Rasala smiled and kept on nodding. It was catching perhaps. Now, in the lab that evening in August, Rasala asked Blau what he was working on. They talked about the particular problem for a while. Rasala turned to me and said, "It's the kind of problem where you push in the clutch and the horn works."

  "Are you trying to clean up the last crock?" I asked Blau.

  "The next one," he said. "I used to think I was getting close."

  Rasala was resting his chin on the back of his chair now. "It's hard to tell when you're finished," he said, without moving.

  One of the engineers working on the new prototype named Gallifrey called out to another, who was on his way out of the lab, "Don't burn that PAL yet. I have to ECO it." And then it was quiet again, except for the hum of the fans in the machines.

  "Well," I said to Blau. "I'll stop asking questions and let you go home. You look tired."

  "It's a long-term tiredness," said Rasala.

  "Going home won't solve it," said Blau.

  For weeks afterward, Rasala repeated that thought again and again. "It's a tiredness going home won't solve." It fit his mood perfectly. The crew's wives — "and/or girlfriends," said Rasala for a joke — were growing weary themselves of the long hours. He felt under pressure from his crew to stick to one shift a day and to more or less normal working hours. Meantime, the pressure to finish by the end of September was building.

  Once, Alsing had tried to guess the plot ahead of time. Months ago, he had said: "Near the end, Tom's going to declare an imminent disaster and go into the lab himself. Sometime near the end we'll do that."

  A computer of Eagle's class must perform at high speed two special kinds of arithmetic called single-precision floating point and double-precision floating point. Scientific users, especially, care about this matter. Maybe more important, this is one of the few areas in which the quality of a computer can be measured and given a number, through a standard test called the Whetstone Benchmark. It is not the only or necessarily the most important standard by which to judge a computer — if you want to do trigonometry on your computer it's very important — but it is a popular test, partly thanks to Data General. For its time, the Eclipse had fine Whetstones, and Data General did not fail to brag effectively about them. West had hoped at the start that Eagle's Whetstones would be better than VAX's. When it became clear that the team could not match VAX's Whetstones at double precision and still keep the ALU on one board, West decided to sacrifice some performance at double precision. If they could surpass VAX at single-precision arithmetic, they could still say Eagle was faster than VAX.

  In August they ran a preliminary Whetstone on an unfinished prototype. The results were not encouraging. The numbers fell below their expectations. It appeared that in single-precision arithmetic, Eagle would be no faster than VAX.

  "It's just bullshit," said Jon Blau in a loud voice one evening around this time while sitting by Coke. "What's better, Crest or Colgate?"

  But Rasala did not respond. "It's discouraging," he said.

  "What's the problem?" I asked.

  "It's somewhere back in there," he said. Without turning, he jerked his thumb over his shoulder, pointing, pointing, pointing at the prototypes that stood humming against the wall.

  Allegedly, some computer companies maintain squads of debuggers called "finishers." They march into the lab and take over the job when a machine has been brought to the last stage of its incubation and the inevitable slowdown occurs. If West had owned such a team and had used them now, he would certainly have provoked a mutiny. The troops might have hung an effigy of him from the flagpole out front.

  But it was time for a change, Rasala believed. He decided to work a regular stint in the lab himself. He chose Guyer to be his lab partner. It was never clear just how well Rasala did at this eleventh-hour debugging, unacquainted as he was with the intricacies of the hardware. The others had come to know it as well as you do the rooms of your childhood home; they could find all the light switches in the dark. Rasala could not, though he was spending many hours studying up on his own. Questioned about this, Guyer shrugged his shoulders and gave his high laugh. "Oh, Ed's comin' along."

  Late in August, after work one day, the whole gang came trooping into the Cain Ridge Saloon, teasing Rasala by turns. It appeared that he had suspected a certain chip of overheating occasionally, and in order to find out if this was indeed the cause of the bug they were working on, he had subjected the chip to the heat gun, a glorified hair-dryer. But he had done so with too much enthusiasm. Rasala had melted the chip's socket.

  The socket cost just a few pennies, but the way they carried on, you might have thought it was made of gold. The way they all laughed and cried out, Rasala cowering in mock trepidation at their abuse — "Meltdown!" "We need more fire extinguishers in the lab, Ed!" — it was clear that under his influence, they had return
ed.

  When he first arrived at Westborough, Jon Blau heard from Rosemarie Seale that in five weeks the Eclipse Group's offices would be remodeled. It was ultimately on account of this that Blau, for one, never believed in the promise of stock options. It was like the debugging schedules. When five weeks elapsed, the promise was that the workmen had been delayed but would arrive five weeks hence — for sure this time. It became a standing joke. Of anything that clearly wasn't going to get done on time, they would say, "No sweat, we'll do it in five weeks."

  Now they were back on double shifts. They had gone back underground. The Hardy Boys were attacking those bugs. The Microkids were cranking out the last parts of their code. Suddenly, the carpenters arrived. All the walls to all their offices and cubicles came down. Then the carpenters went away for a week. Rosemarie did what she could. They managed somehow. At least nothing important got lost.

  "The place is all torn up," said West, "and when it's back together, I suspect that they'll find it's been emotionally torn up, too."

  As it turned out, the group did not get much new space. Alsing and Rasala got walls and a real door apiece, and most of the rest got cubicles again — but ones of slightly different size, so that an engineer with some seniority, such as Holberger, got several square feet more than the recruits. Thus, for the first time, the group's pecking order was made completely visible. It was carved out in steel partitions.

  The atmosphere was different. It was hard to put your finger on the change, but the veneer of equality was gone, and with it perhaps some of what had made the Eclipse Group seem unusual. Perhaps it was only my imagination, but afterward there seemed to be more talk among the younger members of the team about being "peons," more sentences in which the phrase "someone at my level" cropped up.

  By early September Holberger felt he was back in "burnout city." Around this time he found, in a wastebasket in the lab, the stub of a paycheck. It belonged to one of the technicians who had come down from Portsmouth. These were first-rate technicians, and being technicians, they got paid for overtime. Engineers were professionals; they did not.

  Holberger glanced at the stub. He couldn't help it. He was astonished. The technicians were taking home more than twice as much as he was, on account of all the overtime.

  Holberger took the stub to Rasala. Together, they burned it, so that the troops wouldn't see it, and then they went back to work. Holberger laughed. He said once again, "I don't work for money."

  * * *

  By Saturday, the fifteenth of September, Rasala felt that they were close. He inaugurated what he called "the last big push."

  "We're gonna work round the clock, Saturday, Sunday," he said. "We're gonna do it all, and bathe in glory, guys." Late on Saturday afternoon, however, while running one of the last, advanced Eagle diagnostics, Gollum came down with a bad-looking bug.

  The failure occurred when the diagnostic program called for an "interrupt," an order that the computer pause, store all unfinished work, and start in on another job. A computer must deal with interrupts smoothly and regularly if it is going to handle many terminals at once.

  Guyer, Holberger and Rasala took on the bug. They studied the printed listing of the steps in the diagnostic program, and they saw that in addition to everything else — as if interrupts weren't tricky enough — the program was trying to test Eagle's ability to move from one block of memory to another. Then they saw that this program also contained "a page boundary crossing," another fairly arduous maneuver through memory.

  Rasala felt stymied. There were too many possibilities, boundary crossings on top of block crossings on top of interrupts. "It's some complex interaction of the IP, the ATU, Micro- sequencer and the IOC. And probably the ALU, for that matter," thought Rasala. In effect, he was thinking, "It's the whole damn machine." So they hooked up an analyzer and they followed a signal, out of the System Cache, to Main Memory, territory into which the analyzers cannot look. And when the signal came back from Main Memory, it had changed, from a I to a 0 — from a high to a low voltage. Essentially, it had disappeared.

  Four hours passed. Guyer left, worn out at last. Holberger and Rasala lingered awhile. They suspected it was one of those time bombs, hidden far back in the diagnostic program, but they couldn't find it and they went home. The last big push ground to a halt. Next morning, Sunday, all three gathered in front of Gollum again. Perhaps it was a timing problem. Gollum was running with the normal, 220-nanosecond clock. Coke, next door, was still using a slow clock. They took the diagnostic program to Coke and on Coke it ran without a hitch. So it was a timing problem. Had to be. Holberger and Rasala began trading theories.

  In the lab, three-way conversations often led to confusion. Guyer bowed out of this one. He pushed his chair to the table in the center of the room and rocked back. He was just thinking, gazing absently at Coke, at the boards lined up in a row inside the frame. It was like looking at a shelfful of books, and he noticed before he even realized it that Coke had one more board on its shelf than Gollum did. It had an extra memory module.

  "Oh shit," said Guyer.

  "What," said Rasala, without turning away from Gollum.

  Guyer went leafing through the listing of the diagnostic program. "I don't wanna tell you yet," he called over. "But I think I've got it."

  "You tell me now!" said Rasala.

  So Guyer brought over the listing and reminded them that Coke had two boards of Main Memory, whereas Gollum only had one, and the conversation turned cryptic.

  "Yup. More than two-fifty-six K," said Holberger, looking at the listing of the diagnostic.

  In essence, Holberger was saying that the diagnostic programmer, in trying to test Eagle's ability to cross from one block of memory to another, had inadvertently directed the machine to cross from the block located at the very end of one board of Main Memory to the block at the beginning of the next board. But in Gollum there was no next board. So when the program told Gollum to go to the block on the board that it didn't contain, the prototype "fell off the end of the world."

  The ratiocination proceeded. They had failed to recognize these facts because they were used to thinking in the numbering system that Eclipses employed. Eagle used a different system. "And we didn't change our way of thinking," said Rasala. They stuck a second board of Main Memory into Gollum and the interrupt worked. But they had lost the best part of a day. "It's kinda embarrassing," said Guyer.

  Sitting in his office, chewing up a toothpick, West scowled. Rasala should have known better. Would West have known better? "Maybe."

  By the evening of September 20 the machines had accomplished the Eclipse Multiprogramming Reliability Test, but not quite the one designed for Eagles. Probably, Rasala thought, they would reach that milestone tomorrow. At any rate, he decided to try out Adventure the following evening.

  After the M/600, his last machine, had been sent to dozens of customers, a defect had cropped up in it — an important one, which, among other things, prevented the machine from playing Adventure. If he'd tried to play the game in the lab, they'd have found the problem and fixed it cheaply. Although he had never ventured into Colossal Cave himself, Rasala resolved that he would never declare a machine debugged again until it had mastered Adventure. Now the time for Eagle to do it had come. So he thought.

  When Rasala walked into the lab on the evening of the twenty- first, Holberger called out to him, "It looks like we've achieved hard interrupts." There were diagnostic programmers in the room, and some writers of system software, as well as Microkids and most of the Hardy Boys, and they were all doing something. The little room was a forest of equipment now. A couple of months had passed since Eagle had been cloned. Rasala had named the two new prototypes Tartis and Gallifrey, after the home planet and time machine of Dr. Who, the protagonist of a science fiction show on public TV. The two new machines were the first to run with the normal, full-speed 220-nanosecond clock. Like Dr. Who, lasala explained, the purpose of these new prototypes was "to conquer time."
Mag tapes were spinning. There were disk drives everywhere.

  Steve Wallach was there, looking haggard, pale and excited. In addition to being emissary to Software, Wallach had become an important cheerleader. He was forever dragging important people in from the hall to show them the prototype Eagles these days — "signing up the remainder of the corporation," he said. At the moment, he was trying to get Holberger's attention, without success. Several people were trying to get Holberger's ear, and, busy as a blue jay over at Gollum, he didn't seem to be paying attention to any of them.

  "Pretty exciting," I said to Wallach.

  "Yeah," he said. "But there are other problems now. I'm trying to get the documentation done. Some people's writing styles are terrible. They make mistakes they should have learned to correct in My Weekly Reader."

  Over at Gollum, Holberger was pulling a chip out of a board with a pair of tweezers.

  "Aren't you gonna take power down first?" Rasala called to him.

  "No," said Holberger without looking up.

  "Holberger's risking his job right here," said Rasala to the room in general, in his best deadpan manner.

  There was so much equipment in the little lab now that I couldn't sort it all out. Many contraptions had their innards bared, bringing bright colors into the lab — bundles of orange and yellow wires, and brightly colored connector straps. Eleven fat logbooks for Gollum and Coke lay on the central table. Images danced on the screens of analyzers. On a table in a corner lay several printed-circuit boards. They had messages taped over their wires: Doesn't work. Can't be written into.