Dealers of Lightning (51 page)

The geographical rift was only one of numerous headaches. The proj­ect's schedule had slipped almost from Day One. Building the ambitious
Star "was taking longer than everyone thought and it was harder than
everyone drought," recalled Bob Belleville, a former Engelbart engineer
who helped Metcalfe supervise Thacker's work. And there lay the main
predicament. Thacker's processor design, which was dubbed "Dolphin,"
satisfied no one. It busted its specifications in almost every measure
—
too
big, too slow, too hot, too expensive
—
as if Thacker, the minimalist paragon,
had unaccountably succumbed to an alarming attack of biggerism.

Most of the blame, however, belonged to the specs themselves. SDD's
ambition was to bundle together two separate technologies: office
automation, including programmable word processing and networking;
and high-quality digital copying and printing with high-speed lasers.
Designing a processor to handle either task might have been manage­able, but putting them together was like squaring the circle—simple on
its face and impossible in practice.

"Chuck ended up designing a machine that wasn't very good for either
purpose," remarked his friend Butler Lampson. "At a very early stage
they should have said, 'This is impossible, we can't meet both of these
requirements.' But instead he soldiered on and designed this thing which
was kind of big and clunky for the office automation application and
didn't have the power that was needed for the imaging application. It was
a bust."

Moreover, Thacker was fighting his battle using the weapons of the last
war. Because the new system's components were exponentially bigger
and faster than the Alto's—that the Ethernet had tripled in speed only
hinted at the size of the problem—his design plan was to take the origi­nal Alto processor and simply scale it up. In fact, what was needed was an
entirely new architecture.

"The Alto was a machine that was a happy confluence of technologies,"
Belleville observed. "It was built at the right time. It took a mess of ideas
and made them into a machine that was for the time extraordinarily pow­erful and cheap. But now the display was bigger and faster, and Ethernet
was bigger and faster, and the disks were bigger and faster. The complex­ity rose very quickly and Thacker ran into a brick wall."

The central processors inability to keep up with its supercharged
peripherals increased pressure on the rest of the system. One could
never be sure the processor would be finished with one task in time to
handle the next—data bits might stream in so fast from the disk, for
example, that the processor, like a tennis player trying to return the vol­leys of a souped-up ball machine, might not be in position to receive
them. In that event, the system would crash.

Thacker felt deadlocked. "You can trade off cost and performance and
time to market in various ways," he explained. "But if you try to bind all
three of them you may wind up in an infeasible part of the design space.
That's what happened with the Dolphin. In order to cut the costs we cut
the performance down quite a bit—and still couldn't meet the cost goals.
It was still faster than the Alto and it had some things the Alto didn't have,
like virtual memory and caches, but it was just too expensive."

Metcalfe pelted him with acerbic demands for progress while he
tried to maintain a gruelling schedule. The pressure brought him to the
edge of burnout. "My group had about fifteen people in it and half
were in Palo Alto and half in El Segundo," Thacker recalled. "And I
would commute two days a week to El Segundo. That just broke me."

To a great extent, Metcalfe was only passing on the same tension he
received from above. "Xerox had these staff guys who would come in
from Connecticut to check up on what we were doing," he recalled. "It
was my job to stand up in front of these bastards and give them these pre­sentations. I had
the fun
job. I'd say, 'Remember last time I told you how
well it was going? Well, it hasn't worked out that way. Here's the unfore­seen problem
. .
.' And this went on forever."

SDD had few options. If Thacker could not get the job done, the
entire project was in jeopardy, for he was the ace processor designer
and the font of knowledge piped direct from PARC. Yet there was only
so far they could push him, for he was not technically an employee of
SDD—just a contractor, formally on loan.

"Dave could not simply bang on Chuck's desk and say, 'Dammit, I'll fire
you!'" Belleville recalled. During one morose staff meeting, "They were all
worrying about whether Chuck would be hit by a truck, because he was the
one developing the hardware for SDD. Right then Metcalfe joked, "What
you guys don't realize is that I'm waiting in an alley
driving
the truck.'"

Against these great odds, Thacker's team finally produced the Dolphin.
It was as problematic as everyone feared, for the fundamental issues had
not been solved. All the design compromises made had still left it too big,
too slow, too expensive. Weary and demoralized, Thacker returned to
CSL, where he could work once again sheltered from the merciless pres­sure of commercial deadlines. As though in exchange, CSL provided one
more favor for Liddle's division. This was the intercession, like the
deus
ex machina
in a Greek tragedy, of Butler Lampson.

Lampson had become aware while Thacker was still struggling with the
Dolphin design that SDD's hardware effort was hitting the wall. Thacker,
Metcalfe, and Liddle were his friends and he could almost smell their
panic. "I kibitzed with them quite a bit and I noticed at some point that
the Dolphin wasn't going to be satisfactory. But they were in denial. I
decided myself to go back to the roots of the Alto and give it another
spin." This was classic PARC (and classic Lampson)—an unsolicited
answer to an unasked question dropped casually over the transom.

For him the task of designing a substitute processor resembled a
jazzman's noodling on a handy sax. The raw materials were his ability to
take a fresh look at the problem, and the appearance of a new integrated
circuit that had been announced by National Semiconductor Corpora­tion. Because the chip was actually not yet in production, he based his
work entirely on its written specifications. But the flash of insight that
enabled him to overcome the obstacles that had stymied Thacker came
when he realized that the Dolphin processor—like the Alto's—did not
keep time with its own internal clock.

This so-called "asynchronous" architecture had never before been a
problem, which is why Thacker had re-implemented it for the new
machine. In simple terms, the Dolphin and Alto varied the number of
instructions they performed each clock cycle according to the proce­dure they were executing. More complicated procedures required
more instructions (and more time) than easy ones. Therefore the sys­tem could only guess when the processor would be available to service
the next needy peripheral, whether
that was
the Ethernet transceiver
or the
disk controller. Since the processor speed of the
Alto
was more
than fast enough to handle all these
functions
together, nothing ever
went awry. Given the stepped-up
demands
of the Dolphin machine,
however, such an informal
arrangement would
not do.

Lampson's "synchronous" design
differed by
feeding a fixed number of
instructions through the processor
and memory
per clock cycle, which
enabled him to synchronize all the
peripheral
devices to the processor
and eliminate the guesswork about
when an
incoming bit could be
processed.
With
everything ticking
along according
to the
same
clock, a
much smaller volume of data
had to be held in
buffers—which
relieved
the design of the Dolphins hardware
bloat.

Belleville and the others saw
instantly that the
design
Lampson
had
sketched out on seven sheets of
lined yellow
paper would run faster,
more efficiently, and cooler than
the Dolphin. After
making
a few
minor
changes, they renamed it
"Dandelion" (most of
the machines being
turned out by Xerox's Palo Alto engineers
in this
period got names start­ing with
"D"
for "Digital" and became
known
collectively as the
"D
machines") and installed it in the Dolphin's
place
as the heart of the
Star.
A
major hurdle had finally been cleared.

But the
Star was still more than two years
from
its launch
—
and that
would make all the difference in its future.
For
while the Star was
designed to be the office professional's personal nirvana, meticulously
assembled from the finest technological components money could buy,
it
was destined to get blindsided in the marketplace by a new species of
machine that was quite explicitly less than nirvana, assembled quick and
dirty from the cheapest components available, and so low-tech it bore
almost
the
same relationship to the Star that a roller-skate does to a
Mer
cedes-Benz.

A
legend in its own time, boasting capabilities still unmatched nearly
twenty years after its launch, the Star would end up as one of the most
resplendently obsolescent machines ever sold.