Reloading, Redux: How To
Why Good News Can Stay Under Wraps

TBT: When Spectrum's Debut Missed Its Cue

Detroit Interex LogoFew Interex conferences ever convened in October. When the Detroit meeting of 1986 marked its high point, September was already in the year's rear-view mirror. It was a late meeting in the 3000's calendar. Late turned out to be a theme for the meeting's tech papers, because the Spectrum project for MPE was more than a year behind schedule.

By October 1986, the HP Series 930 was supposed to be churning bits using the new PA-RISC architecture and running MPE/XL. The latter was the successor to MPE V, while the former was the first 3000 to leap past Complex Instruction Set Computing. RISC had already succeeded CISC on the HP 9000 side of the server line. But moving MPE while maintaining application compatibility, as well as acceptable performance, was providing unscheduled downtime for the newest HP 3000s. "Where's our Spectrum?" was the cry of dismay from performance-bound big shops.

Detroit's conference proceedings were dotted with accepted HP papers that couldn't be written or submitted. Guidelines on Migration Solutions for MPE/XL, Organization and Direction for MPE/XL, Migrating to the Series 900s as Variables Affect System Performance, an overall Commercial Spectrum Progress Report — all were listed in the two-volume proceedings' index, but each bore a cover page that said "We regret that this paper was not received for inclusion in these proceedings."

The customers had been looking at tomorrows for much of 1985, since Spectrum was the engineering project which replaced HP's Vision effort in 1984. CISC and its memory constraints were holding back large-scale DP, as computing was called then. But like the first version of MPE and the initial 3000s, the stability and performance of the new generation 3000s was not ready for release. Some of the challenges came from testing's optimistic reports.

3000 pioneers already knew about the scant uptime for the earliest 3000s. The computer which made HP a computer company, instead of just one of the world's greatest instrument and calculator vendors, couldn't remain up for more than 24 days. That flaw was in the design, which overflowed a clock register. At the core of the problem was a gaggle of engineers telling HP that everything would be just fine with the 3000, instead of reporting what was well off schedule of being fixed.

Joel BirnbaumSomething very similar happened at the 3000's next debacle when the server tried to embrace RISC and head toward 32-bit computing. HP responded in both instances with an enforcer. In the 1970s it was Paul Ely, feared by many, according to Michael S. Malone's Bill & Dave: How Hewlett and Packard Built the World's Greatest Company. Bad press prompted a fierce HP response once again with RISC, as the industry's journalists called out the woes of a long-awaited HP product. In 1986 when PA-RISC was a 3000 bust, the muscle was HP's labs chief Joel Birnbaum. The man who left IBM to lead HP toward RISC said to reporters in Detroit, "we expect that these problems will yield to engineering discipline."

Pass-PassHP's lab management must have made little room for the prospect of a late Spectrum HP 3000. Some clue of the problems with testing might have been inside one of the few HP-authored RISC papers published in Detroit. Meeting the Challenge: An Inside Look at the Spectrum Testing Program included a flow chart for tests. The graphic flowed down to a decision diamond labelled Pass. Not Pass/Fail, or even a yes-no marker on the corners of the diamond. The software was caught up in a loop that headed back to testing. The testing paper acknowledged the challenge of making something faster and different that ran old software was a stout task.

The MPE/XL operating system is a new product. It has the unenviable task of behaving precisely like its predecessor, MPE, while running code which has not been recompiled, and supporting a whole new set of capabilities as well. If the code is recompiled, it must deliver vastly increased performance, using techniques radically different than those of its predecessor.

HP-UX didn't labor under those tasks because the Unix marketplace was not a commercial market. Software at Unix sites was built for technical tasks and used in lab settings. These users had not established business revenue requirements based on software stability and performance. HP 3000s had to stay online, extend their power to meet company growth, and run mature applications to protect business investments.

Test TreeThe irony of having a testing report appear at a conference while the migration, commerce, and performance papers — as well as the computer itself — were missing was rich, as well as disappointing. Outside MPE experts, contracted into the MPE labs, eventually gained the ear of top HP lab management. A Destructive Testing unit probed MPE/XL alpha releases for failures and found many. 

It would be another year until MPE/XL 1.0 hit the Manufacturing Release milestone, and even then, the 1.0 version was called a career crossroads for the managers who had to put it into production. 1986 had its computing demands that sound a lot like those of today: lower power, higher speed, higher yields to manufacture chips. A R.I.S.C Tutorial in the proceedings summed up the dilemma HP worked more than a year to overcome. People were bringing old software onto new hardware, not rewriting for a new system.

In Hewlett-Packard's case, simply implementing a new RISC-based computer system would have alienated a large potential customer base. Not only must the new generation of HP RISC-based systems have more power than the current HP 1000, 3000 and 9000 systems, but they must be compatible with these systems as well. The simplest compatibility approach is to require each application to be re-compiled on the new system. This method is common practice in the Unix community.

HP chose to approach compatibility by emulating the HP 3000's instructions in software. The Tutorial explained that "A program will examine each instruction and branch to the appropriate subroutine (dubbed milli-code) for execution." HP and the 3000 users came to call this magic Compatibility Mode. CM gave the new 3000 an instant field of software and apps, even though it made the 3000 late for its cue to take the stage.