Tom Johnson - who reports that he still "dabbles" with Apex - wrote to remind me of Ramachandran Krishnaswamy and Dan Ehrenfried.
David wrote to fill in some details of Rational's earliest history. In 1979, David had been funded by DARPA to explore techniques for the compilation of Ada tasking. Dave Stevenson and Howard Larsen were then grad students working with David to build the Ada-M compiler, which was written in Lisp. Mike and Paul had been developing their idea of an Ada machine and so looked to the Ada-M effort to bootstrap Rational's first validated compiler. David reminisced about a meeting in his dining room in December 1980 which brought together the gang with some venture capitalists. David is a professor emeritus at Stanford and is currently working on R&D in event processing.
Joe Marasco sent me a lengthy missive detailing the R1000 family tree, with Dave Lofgren and Jim Archer adding some additional bits. Rational's machine evolved through four series of different models, each representing some significant packaging or architectural change. We indeed shipped our first machine (which was named simply R1000, with the series 100 moniker added later) to Rolm on December 31st, 1984. Rational ended hardware production in 1993, shipping its last machine to the Danish Navy. There are still several R1000s under support and maintenance contracts; Jim Archer continues to support the software side of Rational's earliest products and Dave Lofgren handles the hardware end. Joe estimates that a total of approximately 300 machines were shipped over the R1000's lifetime. As Joe further reported:
The "Series 200" was led by marketing manager Yosi Amram in 1986, and was launched in the fall of that year at Ada Expo. It came in three models: a model 10, model 20, and model 40. The low-end model 10 was never purchased. The model 40 was two model 20's literally put into one larger cabinet. Each half of a 40 had its own control terminal and its own peripherals; as I recall, it was challenging to get them to share a printer. The first Series 200s were shipped to Philips AB at the very end of 1986. The Series 200, especially the model 40, was the "workhorse" of the product line for several years (1987 - 1989), with many of them at sites such as Philips, CSC, Rockwell, and others. Customers who bought them in substantial quantities began to understand the difficulties of doing large-scale development on multiple R1000's. The original R1000 employed a PDP-11/24 as an I/O processor, which enable the use of off-the-shelf disk, tape, and communications controllers In the series 200, this was replaced by a Motorola 68K microprocessor that generated the DEC Unibus so we could continue to use those controllers. Wayne Meretsky did both the hardware and software for this revamped I/O system.
In the fall of 1989 yours truly fielded the Series 300, which was basically a "skin job." It used the same boards as the 200, but Mr. Druke and his hardy band of hardware engineers weaseled the form factor down somewhat further. The idea was to get the price down as much as possible without doing any major engineering, although disk and tape drive progress allowed us to continue to reduce the size of the package. The 300 had the ability to increase the main memory up to 64 Meg from 32; doubling the memory had a big effect on performance. The 300 product line also included a coprocessor variant, which meant you could use the processor and memory in conjunction with a disk farm from Sun. In an unexpected turn of events, the pesky salespeople continued to sell the more expensive "standard" R1000 configuration to the detriment of the coprocessors. I believe the leader in this regard was Tom "Too Tall" Smith, who sold a raft of full-up Series 300 "model 40 equivalents" to Lockheed for Space Station Freedom in 1990.
The immediate reaction to fielding the 300 was to get to work on the 400. Quickly abandoning the co-processor concept, the 400 took the next logical step of reducing the package size still further by incessant shrinking of the drives and a re-do, in under a year, of the processor and memory boards, as well as redesigning the I/O system yet again. The basic architecture of the processor and memory boards was unchanged, but the critical "zero insertion force" edge connectors got revamped so the boards could be made smaller. This feat had previously been postulated as impossible, but when faced with imminent extinction, Druke's hardware boys somehow managed to make it happen. The Series 400 came out in the fall of 1990, and was as small as the R1000 ever got. We sold 400's from late 1990 through 1993, when we finally ceased producing new hardware.
Cross-correlation of these dates would indicate that we commenced the Apex development project AFTER we fielded the series 400. I believe that during the latter part of 1990 and into early 1991, the hardware people were still working on a "next generation" R1000 that would completely re-architect the basic boards. But that effort got killed at the end of 1990 in the famous decision to abandon the hardware business. Curiously, it was about nine months after that that the Apex team was formally constituted and launched. During the two-year period it took to build Apex, most of the hardware people either left, worked mostly in support, or transitioned to a number of "Delta" improvement projects. Delta was the R1000 OS of the day, and the "inside joke" was that everything wrong with it would be "fixed" in Epsilon. Of course, Epsilon never happened; when the R1000 was replaced by Apex, the operating system became one of the standard vendor's version of Unix.