TBT: A Race to Engineering Discipline
October 2, 2014
A two-page ad, many months after the reckoning of the prior fall, asked the business community to imagine More Raw Horsepower from a system called the HP 3000 Series 930.
October was a month to remember from an engineering era that Hewlett-Packard would rather forget. The era was the cycle of what independent contractors called Destructive Testing, the repeated, broad-spectrum hammering on the new MPE/XL operating system that was going to power the first PA-RISC Series 900 HP 3000. HP paid these experts to break what it had built.
The computer rolled out in the early fall of 1987, a full year after its Unix counterpart. It was just 12 months earlier that HP's tech czar Joel Birnbaum swore that PA-RISC would emerge from a swamp of too-sweet project management.
More than 1,000 engineers would eventually work on pulling MPE/XL and its Reduced Instruction Set Computing steed 900 Series out of a ditch. During 1985 and through much of 1986, status reports about the development of this faster 3000 were encouraging. No show-stoppers there, not with so much pressure on horsepower improvements. The Series 70 was released in 1985, a stop-gap server that ran faster than the Series 64. But not fast enough at plenty of major HP customers, the group called the Red Accounts.
Those lab updates were being sweetened because a replacement for the Series 64 and 70 was overdue. HP had already scrapped the 3000's update plans for HP Vision, broadening the replacement project to call it HP Spectrum. This was design to be used in all HP servers, working through an HP-invented RISC chip architecture. The twinkle in Birnbaum's eye while he was in IBM, RISC was going to be a business success. HP hired him away to deliver on the RISC promise.
But by October 1986 at a conference whose theme was Focus on the Future, the 900 Series was undeliverable as addressed. Birnbaum had to deliver the news to pressmen, reporters assembled in a conference room. We circled him, standing and taking notes, quizzing Birnbaum as he said the horsepower would arrive. More important was stability. Birnbaum explained patiently that interfaces between MPE software modules were not working as forecast. Not yet.
This didn't appear to be a man accustomed to explaining delays to the public, especially critics from the press. But he uttered a phrase that afternoon in Detroit at Interex '86 that seemed to close down the probing questions. We wanted to know, after all, could anyone believe that the vanguard Series 930 server would appear after more than two years of reboots and delays?
This problem, this issue, will yield to engineering discipline.
And he spoke that sentence as a vow, in a tone with some anger. You had to believe in engineering discipline, if you were to write stories about computing. It was easy enough to believe in the 32-bit computing that MPE/XL was aiming at, since it represented HP's entry in the cutting-edge design of the day. At that same Interex '86, DEC rented a set of rooms in the hotel across the street from the downtown Renaissance of the HP show. Its Vax servers were already running there in 32-bit mode. DEC even tried to offer an IMAGE-workalike on the servers in the display rooms. One of the accounts was even named REGO, a nod to Adager's founder Alfredo.
"Digital Has It Now," was the theme of the competition's campaign that October. Big two-page ads in Byte and Computerworld, printed on silver ink backgrounds and massive white letters on top, assured the markets they didn't have to wait for anyone's discipline if they bought DEC. There was lots of hubris back then, as now. Cullinet ran one ad that used a headline it only made sense to switch to their application suite, since it was the only one built for use in the '80s and '90s.
By the fall of 1987, the Series 930 was squeezed out to a handful of sites including Northern Telecom, as evidence that Birnbaum's discipline had yielded 32 bits of success. But few of those 930s ever booted up in production. Just like the original HP 3000 had buckled under the demands of MPE, the 1.0 release of MPE/XL drove the RISC hardware to regular halts. The problem lay not in the hardware, but in the software which had not been destructively tested.
It took another year for the first, genuinely effective 900 Series HP 3000 to ship. The Series 950 might not have been the first horse out of HP's RISC 3000 gate. But it could promise at least 10 percent more horsepower than the biggest Series 70 running MPE. It was the work of those independent engineers -- many of them former HP employees, developers and SEs -- that let MPE/XL get free of its starting gates.