Saturday, August 26, 2023

Scenes from the Solbourne Computer corporate video, March 1992

I've previously mentioned Solbourne Computer, which for a number of years was probably Sun's most significant early competitor in SPARC-based systems. Solbourne was the first to market in 1989 with multiprocessing SPARC servers based on their custom circuit-switched 108MB/s KBus interconnect, running a bespoke but highly compatible licensed SunOS 4.x clone named OS/MP; later, they even developed their own SPARC CPU, the benighted MN10501 "KAP" Kick-Ass Processor (yes, really). At least one Solbourne system even went on the Space Shuttle. Sun didn't have multiprocessor SPARCs until 1991 with the competing MBus-based Sun-4m systems.

My own affinity for them started when I had an account on a Series6 for a few months as an undergraduate college student and a few years later ended up with my own pizzabox S4000DX. I still have that machine, but my usual Sol is an S3000 portable workstation with that lovely garish gas plasma screen simply because it's so incredibly unique. In 1990, when this presentation slide was made, the sky seemed bright and the foundation seemed solid.

But by March 1992, cracks were showing, and now we have a view into that process then — a rare inside glimpse at the hardware development and executive management of a 1990s tech company. In a shipment of various Solbourne paraphernalia, a former employee from late in the company's dying days sent me a couple VHS tapes of corporate meetings. One of them will need some cleaning and rehabilitation, but the other was in good condition and absolutely playable, and I was able to digitize it on the Power Mac G5 Quad with my trusty Canopus ADVC-300. These are second-generation copies and the quality is worse than usual, but they tell the story adequately and serve as a fascinating time capsule of the company's later doom and nineties-era enterprise computing generally.

In addition to the anonymous contributor, I'd also like to thank (in alphabetical order) Stan Hanks, Al Kossow, Warner Losh and Dworkin Muller, ex-Solbourne staff and otherwise helpful folks who helped to identify the hardware, locations and people in the following screen grabs. Any errors in the history below are mine, not theirs, and if you can fill in any missing details or provide corrections, please post in the comments or E-mail me at ckaiser at floodgap dawt com. I've moved some of the grabs out of chronological order for ease of explanation where it did not obviously change meaning or significance.

The video is in two distinct halves, but both parts were taped at Solbourne's main facility in Longmont, Colorado, between Boulder and Fort Collins north of Denver. Their former building at 1900 Pike Rd appears to be currently leased by memory manufacturer Micron Technology, who is expanding to a nearby building that previously housed hard disk manufacturer Maxtor. While chips like the KAP were fabbed in Japan and the actual PCBs manufactured by a contractor, final assembly, testing and repair were all at Longmont. Similarly, the operating system group was also at Longmont handling release integration, installation software, bug fixes and mastering, with media mass reproduction done by a third-party.

The first, longer half was a corporate and financial briefing that took place in the lunchroom. In 1990, Solbourne had over 250 employees; this meeting looks like about a hundred were present.
The first speaker, complete with photocopied transparencies and an overhead projector, is Carl Herrmann, then Solbourne's vice-president of international operations. However, this is near the end of his service in that capacity, as the Board of Directors accepted founder and CEO Doug MacGregor's resignation in April 1992 after a disastrous year-end earnings report, and Herrmann was appointed in his stead as CEO.
As you would expect, his portion of the presentation largely talked about bottom line (reporting $3.2 million in revenue in January and mentioned several "million dollar" large orders) and fiscal responsibility. However, when you're telling the rank and file to "SPEND OUR LIMITED RESOURCES CAREFULLY" ("including your time and other limited resources"), this is usually an unambiguous signal to update your résumé. Calling the cash flow "neutral" rings a little hollow when the third line on the screen is actually a $31.1 million loan from Panasonic Finance (Matsushita, the company's largest stakeholder).
Herrman was followed by Travis White, the vice-president of marketing. White's talk was largely about product, "focus" and market positioning, using Macintosh's successful inroads into desktop publishing to justify Solbourne's newer focus on database and consultant-centric offerings ("it gives us a chance to position away from Sun [hardware] and their fairly intense price pressure ... we think there's a lot of good reasons to go after commercial, databases, their servers"). Oracle's offerings were mentioned prominently as a potential "tip of the arrowhead," and White estimated database work could account for as much as "20 or 30 percent of our revenues." Unfortunately, what could have been a deliberate, intentional expansion instead started having the feel of a hasty pivot, and employees even reported tales of sales staff saying existing hardware customers they didn't need their business anymore. The net effect was to substantially impair revenue from their existing lines of business while trying to enter a new, arguably more cutthroat market with multiple established players; it had exactly the result one might expect.
This point was further driven home by the final speaker, Walt Pounds, who was then the CFO. Pounds talked at length about some of the customers he visited, including Equifax, Hayes (using Oracle) and NCR, among others, including Equifax's initiative to make credit report disputes run "machine to machine," something that couldn't be done on "these nasty mainframes or ... minicomputers like Sequent [Symmetry servers]." Solbourne's bottom line was on their clients' minds too, as "the first question" out of customers' mouths was, "'are you going to be solvent?'" Later, he mused that "[i]t was the first time I had the opportunity to respond directly to the question of, 'so you're in the database business; what does that mean for me?' And it was very simple to say, 'Look, this is where we're focusing. This is where we're putting most of our energy.' ... We're not myopic in that the efforts we're doing are going to pay off in other areas as well." As we'll show in the hardware segment, by this time Solbourne was deploying the same or similar processors as Sun and other SPARC vendors, meaning they had to have a hook to distinguish their offerings and that hook appeared to be "really great at databases." Part of this attempt was through software as well as hardware, such as improving OS/MP's ability to more rapidly context-switch. However, the fact that customers were asking such fundamental questions — solvency and market focus — was further proof that Solbourne's corporate retrenching was not seen positively by new buyers.
With that theme of transition suffusing most of the tape, the hardware portion of the video — which I naturally found more interesting — was therefore the shorter of the two. The map of the hardware section is not legible in the grab (I told you it was a copy of a copy), but seems to have been limited to one wing of the building.
Wandering about in the lab. The panel wasn't sure who many of these people were. Some may have been invited guests.
A "prototype board for KBus [with] a parallel general purpose interface which we can connect to other boards for testing purposes" was being demonstrated by one of the senior hardware engineers, possibly Ed Reihms (possibly misspelled).
This gentleman also wasn't known to the group, but he had a number of interesting toys. The first was a Hewlett-Packard impedance analyzer of some sort (I can't identify the exact model) for checking traces, which he described as mostly to evaluate how clock pulses distribute through the logic board.
The next device was another HP testing unit, a Hewlett-Packard 4145B semiconductor analyzer to "evaluate components, how they'll react when they're on the board. Right now I'm measuring how much of a load a given low logic signal out of the device can take." (Not sure if I transcribed him correctly as there was a lot of background noise.)
The last was a microscope set up as a (as he describes it) "touchy feely type of thing" showing "all of the integer units that were used for all the individual boards that were designed by Solbourne."
"We start out with the Series4 [lower left, a 16.67MHz Fujitsu MB86900, the very first commercial SPARC implementation], the Series5 [lower right, initially a 33MHz Cypress CY7C601 with a Weitek Abacus 3171 FPU; the chip here is a later Series5E CPU that ran at 40MHz], the IDT [upper left, KAP MN10501], and the new Series6 'Viking' chip." Solbourne was developing the Series6 at the time, a large KBus system supporting up to 11 slots and eight CPUs. The SuperSPARC Viking CPU it used came out that same year in 1992 and was a 3.1 million transistor SPARC V8 developed by Sun and fabricated by Texas Instruments on an 800nm process. It came in two speeds, this first version at 33MHz; the later version used in the Series6E, Solbourne's last official hardware product, ran at 50MHz. Sun used these chips too, which is why Solbourne needed some way to distinguish themselves.
Several examples of Solbourne's hardware offerings were tacked up on the walls in the hardware lab. The boards in the first grab are all system backplanes. Clockwise from upper left are a VME backplane, an option on many KBus systems but not on the desktop /500 chassis or the SBus-based pizzabox IDTs; a KBus backplane out of a /900 rack system chassis; and a KBus backplane from a /600 deskside chassis.
The second grab shows actual KBus boards. The shot is too blurry to positively identify the left two, but they are "least unlike" I/O boards, based on what looks like coax Ethernet and their particular board stiffener layout. The red board at the top right is a prototype Series6 CPU board and below it a single-slot VME adaptor with Ethernet and SCSI connectors hanging off it.
An actual Series6 prototype was demonstrated by Ty Sell, technical head of hardware engineering, using an MBus simulator called MSIM. MBus was a high-speed circuit-switched multiprocessor interconnect designed by Sun and Ross Technology in 1991, intended to replace the earlier SBus between the CPU, memory and logic board with SBus remaining for add-on cards. (SBus was later replaced by PCI.) Solbourne's uniprocessor IDT pizzaboxes, including the lovely S3000, were completely SBus from cards to CPU as were their SPARCstation contemporaries. However, the Series6 was intended for enterprise customers who likely already had an investment in Solbourne KBus hardware, so the new system returned to using KBus for add-on cards instead. In this application the KBus backplane effectively links together not only the KBus cards but also multiple MBuses which are independent of each other.
The Series6 processor module itself uses a different private Sun interconnect called VBus (SuperSPARC-I could use either VBus or MBus directly). The module Sell is holding has eight SRAMs for cache, a Viking CPU, and an MXCC L2 cache controller. VBus runs between the Viking and the MXCC, which handles the L2 and provides the asynchronous MBus interface.
With the processor module so mounted on the CPU board, the CPU is thus connected to the rest of the system via VBus to the MXCC, MBus from the MXCC to the CPU board, and KBus from the CPU board to memory and other CPUs and devices. In production Series6 units, both the Viking and MXCC are on the CPU board, not on a daughtercard.
The prototype was mounted in a /600 deskside chassis, though it would never be offered with one; Solbourne was exiting the workstation market at this point and the only chasses the Series6 was sold with were the larger 7- and 11-slot /700 and /900 configurations respectively. The simulation part manifests in an umbilical "tail" that provides programmable system facilities.
With the simulator running, the Viking on the CPU module has managed to start up from a simulated boot ROM, the console of which is also provided by the simulator, and is displaying an operator prompt. However, the system was also capable of simulating the CPU as well using an ISP (Instruction Set Processor) interpreter, and earlier versions prior to fabricating the Viking used software cores for testing.
The next segment was presented by Karl Whiting, whose exact role at Solbourne wasn't recalled, demonstrating programming Xilinx FPGAs using Synopsys tools and some sort of circuit routing program which might have been Cadence Virtuoso. The workstation he appeared to be using is an S4100 (S4000DX), the top end IDT system in Solbourne's former workstation line.

These screenshots aren't particularly legible either, but you can see the state of electronic design automation tools at the time.

Comparing this view with this printout from Sell's demo,
it appears the board Whiting was working on (or at least playing around with) was also the Series6.
This designer was also not known to the group except for actually being a Matsushita employee, as were most of the Japanese technical staff, who were described as being more or less a technology transfer scheme. He got two S4100s to himself, one of them dual-headed.
I'm not sure what program he was using here, though the name "Routesmith" (possibly a subcomponent) was mentioned for following the traces. He demonstrated several examples of placing and removing components from the board.
Finally, some more views of the hardware lab before we finish.
And of course you weren't here for the hardware demos or the financial blather. You were here for the beer.
"This is how you pour a beer. This is live, not Memorex. Remember the Walnut Brewery when you're in Boulder." Note the dot-matrix printer inside an enclosure to muffle the sound, next to the laser printer. No one recognized these folks either. Do you?

History tells us, of course, that the switch to databases didn't save Solbourne. As losses mounted, Doug MacGregor was forced out of the company shortly after this video and was replaced by Herrmann. The Series6 hit the market in December 1992 and certainly had sales — including the one I used as an undergraduate — but failed to expand the business or revenue beyond the existing base even with the revamped focus and enhancements. Additionally, Sun refused to license Solaris to competitors, including Solbourne, leaving OS/MP stuck in time with the older SunOS 4.x kernel.

In 1993, the beginning of the end was visible when Solbourne forged an agreement with Grumman Systems Support Corporation for on-site maintenance and support of existing customers, and in May 1994 Solbourne fired 65% of its 180 employees and exited the hardware market completely, killing the planned twenty-processor Series7. Some employees became Grumman consultants while transitioning to other jobs. Herrmann was replaced by Pounds as CEO in November 1994, the final step in the transition between "Solbourne I" the hardware company and "Solbourne II" the Oracle database consultants, the latter after another round of layoffs with around 10 employees left. Deloitte bought out the remnant in 2008.

The individual who provided me the videos was one of the later casualties, who stayed on to wind down the satellite office they worked for and ship everything back to either Longmont or Grumman. But this person didn't send everything, and now computing history can remember a little more. If you can fill in the gaps, post in the comments or send me a message at ckaiser at floodgap dawt com.

No comments:

Post a Comment

Comments are subject to moderation. Be nice.