As U.S. directed forces gather in Saudi Arabia for the bigges military operation after Normandy, microchip designer Todd stumbles upon a previously unknow function hidden inside the Kali computer chip. That night, as he lies sleeping in his bed, he is shot
In the closing days of 1995, after a demanding week working in the fast lane of Silicon Valley, exhausted software engineer Nick, boards an overnight flight to Boston and ends up sitting next to a very unhinged man who declares he knows the secret of Gulf War Disease. Nick's travelling companion meets his dramatic death somewhere over Utah - and the police have a suspicion that Nick is the murderer. It's not long befeore everyone wants a piece of Nick - from the Airport cops to the CIA. obooko.
Todd Griffith was going to debug Kali or die trying.
Thirty-six strobe lines--an electroencephalogram of the Kali chip's brain waves--danced in parallel from left to right across the four monochrome monitors in his cluttered office. The answer to the riddle lay hidden within them, and he knew that if he looked hard enough he would eventually find it. Unless, of course, he went mad first.
The chip was six months off schedule, and before Kali no chip of Todd's had ever been late by so much as a millisecond. It was the curse of the new guy. It had to be. The original design team of Todd and Casey had met their milestones with monotonous regularity. Then management had stepped in with characteristic stupidity, reassigning Casey to some skunkworks kludge and replacing her with Pavel the Weirdo. For the last nine months, ever since Pavel had taken over as Todd's junior partner, glitches had popped up with distressing frequency. Things that had worked before suddenly stopped working. And although it was tempting to blame Pavel for the bugs, they always turned out to have been there in Todd's logic all along, lying dormant. Week by week Pavel added more capability to the Kali, and week by week this additional logic exposed the weaknesses in Todd's original architecture, as a cantilever added to the tenth floor of a building might expose a flaw in the foundation. Todd, in his arrogance, had built very little debug time into the schedule. And, being a hardware guy, he resolutely eschewed Brooks' famous dictum to "take no small slips." Therefore every new bug meant a new small schedule slip, and every time the schedule slipped management became exponentially more pissed--and Todd's life became exponentially more wretched.
If Casey were still on the project Todd wouldn't be in this mess. If Casey were still on the project, the design would be done by now and some silicon foundry in Texas or Sunnyvale would be cranking out Kalies like jellybeans. Casey would have found work-arounds for the subtle flaws in Todd's logic--she always did. If Todd's specification called for a two micro- second wait-state, Casey's logic would have tolerated any value between one and three. Casey accommodated Todd, subordinated her design goals to his. She reacted instinctively to the feel of his design--as a junior surgeon might address herself to a wound in the scalp so that the senior surgeon could focus his attention on the bullet in the brain.
Pavel wasn't like that. If two microseconds were specified, then Pavel's circuitry demanded two microseconds. Not one, not three, nor even 1.9 or 2.1. Pavel claimed to be a purist when it came to Very Large Scale Integration. Todd thought that 'anal retentive' was a better term.
But he had to give the devil his due. Pavel himself might be an unfriendly, humorless, compulsively orderly and geeky weirdo, but Pavel's designs were magnificent. They were economical of power, heat, real estate, and cycles. They were, in a word, cool, and anybody who knew anything about the aesthetics of VLSI design could see that. Pavel's logic demanded precision, but it paid good dividends. There was no denying that the chip that would ultimately result from the collaboration between Todd and Pavel was going to be vastly more cool than the one originally conceived by Todd and Casey. Therefore it was no good bitching to management about Pavel. Management could clearly see that coolness came from Pavel and schedule slippage came from Todd. Besides, there was no way he could blame an interface bug on the junior designer when his own circuitry didn't meet the spec. So whenever a new bug appeared, Todd owned it. He was not used to owning bugs, and he didn't like it.
As irritating and embarrassing as these bugs had been, they had been relatively easy to diagnose and correct. They were nothing but timing glitches, the kind any smart college kid working on his first chip would be expected to handle. You could find a timing glitch with the silicon equivalent of a flashlight, and you could fix it with the silicon equivalent of string and bubble gum. In the nine months that Pavel had been working on the Kali project, Todd had found and fixed forty-seven such bugs, with an average elapsed time of just four days from the time a report was opened in the bug- track database to the time it was closed. Today, March 28, 1990, only one open bug remained.
But this one was no simple timing glitch, and there was no simple fix for it. It was different from any bug Todd had met in nine years of design- ing computers for a living. It was a phantom. Irreproducible. Subtle. And fundamentally impossible. Bugtrack number K666. The Beast.
The Kali would run successfully for hundreds of millions--hundreds of billions--of cycles, then inexplicably shit the bed, as if deciding that just this one time two plus two equaled seventeen, or that just this one time the letter 'x' fell between 'q' and 'r' in the alphabet. Then Kali would resume giving the right answer, failing only once until a power-down and reset, and never, ever repeating the same mistake. It was like that Charles Addams cartoon where a barber holds a mirror to the back of a customer's head to produce an infinite regression of faces in the mirror in front of him, and the seventh face is a monster. The Beast was a monster. It was one true, hairy, son-of-a- bitchin' bug.
Retrospective diagnostics proved that the floating-point section of Kali's Arithmetic Logic Unit was unmistakably, and correctly, executing orders that it had not yet received and that the early arrival of results from the floating point messed up fifteen dependent steps elsewhere in the device. Clairvoyance in an ALU was an intriguing capability, but at this point in his life Todd was much more interested in the prosaic than in the paranormal. Since he didn't really believe that Kali had supernatural powers, he had to wonder if something almost as unexpected was going on.
He was certain that the Beast was caused by a race condition, when elec- trons on separate paths towards a common logic gate wound up in a dead heat, and the gate could not determine who got there first. When that hap- pened the output from the gate would be unpredictable. Debugging a race condition was usually the most satisfying part of chip design, but chasing the Beast had long ago ceased being fun. How could the decode be working correctly if the fetch had failed? Unless the laws of the universe had been suspended within the confines of his chip, it simply wasn't possible. And yet it happened. How?
Todd had been hunting the Beast relentlessly for three months now. He hardly ever left the Mill. If he wasn't in the lab huddled over prototype silicon with probes and oscilloscopes, he was in his cramped office running simulations and poring over schematics. Other than short exchanges with Pavel, he spoke to virtually nobody. He hadn't answered his phone or e-mail in a month. His food came from Mill vending machines. He slept on the floor of his office. Every third day or so he showered in the locker room, and once a week he went home for a change of clothes, timing his arrival to hours when his house mates would be asleep or at work. Todd didn't want to see anybody. Human companionship was a vague memory, and he wanted to keep it that way.