THE SOUL OF A NEW MONGOOSE
The Story of Australia's First SPARC Unix Workstation
by Carlo Kopp
Designed by a Monash Computer Science postgraduate student, assisted by a Monash Computer Science honours graduate, this is a first hand account of the events of 1993.
In 1993, Carlo Kopp, then Chief Design Engineer of the GCS group of companies and also a part time Monash University Computer Science MSc student, commenced the design of Australia's first commercially manufactured SPARC architecture Unix workstation motherboard. This is the story of this project.
Kopp, an experienced digital logic and high speed analogue design engineer, and writer of embedded software, was hired by Graphics Computer Systems in 1990 as Chief Engineer. The Managing Director of GCS, Geoff Croker, intended to manufacture SPARC workstations in Australia. Croker had at that time entered a close commercial relationship with US SPARC compatible manufacturer, Solbourne Computer, with the specific intention to licence manufacture Solbourne's planned S4000 desktop workstation in Australia. Solbourne was mostly financed by the Japanese giant, Matsushita Electric, and staffed with American engineers, managers and sales personnel.
This was not to be, as design blunders by Solbourne's Silicon designers resulted in a single chip CPU with undersized caches. An interesting historical note was that Croker had warned Solbourne about the possible performance problems resulting from undersized caches 18 months before the CPU was cast into Silicon. Predictably, the S4000 with its single chip MN10501 CPU fell short of intended performance goals by a significant margin, with integer and floating point operations marginally better than the competitive Sun SPARCStation 1+. Shortly after the S4000 was released, Sun Microsystems released the 40 MHz SPARCstation 2 which destroyed any hope of success for Solbourne's stillborn machine. With the failure of the S4000 in the marketplace, Solbourne Computer suffered a major loss of investor confidence and thus supporting finance [Solbourne departed the hardware industry in 1994]. It is worth noting that Solbourne Computer grew to an annual turnover of USD 100 million per annum at a record rate, and was highly vulnerable to any product failures which might occur.
An ill advised attempt by Solbourne Computer to take over Solbourne Computer Australia destroyed the relationship between Solbourne Computer and its Australian partners. Geoff Croker's plans for an Australian built Unix workstation were in tatters.
Facing the peak of the early nineties economic recession, Croker stubbornly decided to follow his original plan. Together with his Chief Engineer, he developed an three year strategic plan to incrementally build up an indigenous design capability, though a series of products intended to culminate in a locally designed and built world class workstation design. The plan was structured around the model of controlled risk management and R&D funded from available company profits, as there was no hope at that time for any R&D investment. GCS had to fight their battle alone.
The starting point for this plan was to design and manufacture a series of computer chassis, initially to be used with imported motherboards and later to be fitted with the Australian board. The chassis products were to be followed by small I/O boards, used to replace imported US Sbus boards. Finally, a SPARC architecture motherboard design was to be produced and built.
The first product in this series was the DeskPAC-5 5.25" drive enclosure chassis, with a low noise high throughput fan and heavy duty power supply. The DeskPAC-5 was an outstanding success when produced in mid 1991 and was still being manufactured, without design revision, in 1997. It is almost certainly a record life cycle for any Australian computer product.
The DeskPAC-5 was soon followed by the larger DeskPAC-20 server chassis, and WorkPAC workstation chassis, the latter designed for power users and small servers. The DeskPAC-20 was notable because it was built not only in standard steel sheetmetal versions, but also in a stainless steel model, and a Gold plated display model, several of which were sold ! One customer complained about the "boring grey colour" of the basic DeskPAC-20, this was remedied by an early morning raid by GCS engineers who replaced the customer's standard grey model with a custom painted purple chassis. The customer's reported response is not printable !
As GCS shifted from being an importer of computers to being a design and manufacturing house, the company's Chief Financial Officer, Angela Croker, re-structured the company into a group of companies. This was necessary to protect the company's intellectual property base from predatory foreign competitors, to maximize possible federal government assistance and to introduce a new level of discipline to the management structure, based upon external audit. GCS is not a public company so this was a wholly self imposed regime of operation. It may be argued that this restructuring was an important contributor to the company's success in pursuing a goal which no other Australian company had ever achieved.
Once GCS had established their design office, documentation system, component purchasing system, production stock control system and design office technical library, the time had come to tackle a more ambitious target.
The next product was the BW2 ECL monochrome frame buffer (graphics board). This was an Sbus monochrome frame buffer designed to run in the latest SPARC workstations, but equipped to provide the ECL hardware interface and video timing required by a Sun 3/50 or 3/60 workstation monitor. This product allowed GCS to undercut US vendor "swapout" upgrades of Sun 3/50 or 3/60 workstations, by replacing only the "pizzabox" workstation base and retaining the 19" monochrome monitor, the most expensive single component of the workstation. The 32-bit SBus ECL frame buffer board prototype was debugged using a two channel Tektronix 475 analogue oscilloscope. A Tasmanian customer subsequently received the "Alpha 1" prototype board in his production system, when a US supplier failed to ship a BW2 board on time. To the knowledge of all involved, trusty BW2 "Alpha 1" is still running somewhere on the Apple Isle !
The strategy of replacing imported boards continued. GCS, like all suppliers of OEM computers, wanted to offer its customers a wide range of colour and true greyscale monitors. Unfortunately many of the colour monitors could not be used because they could not accept Sun's video synchronisation signals. Moreover, the pinout arrangement meant that greyscale monitors would only ever run in pseudo-greyscale mode, driven by the green video channel only. So Kopp devised an alternative, which was a variant of the Sun standard CG3 colour frame buffer board equipped to provide embedded synchronisation and jumper selectable true greyscale video, retaining full timing and driver compatibility with the original Sun board. Thus was born the home grown CG3 colour frame buffer board. The first CG3 boards were shipped in late 1992, and the product remained in production until 1995 when customer preference for the CG6 or GX accelerator saw the demise of the non-accelerated frame buffer in the marketplace.
By late 1992 a significant series of developments took place. Silicon Valley chipset manufacturer Nimbus produced the NIM-6000 SPARC motherboard chipset based upon Sun's new SPARC standard 64-bit Mbus, Taiwanese chipset manufacturer Twinhead produced a similar product, the Gemini chipset, and importantly, Ross Technologies and Cypress began to ship the CYM-6000 series of 40 MHz clock speed Mbus CPU modules. The technology had matured to the point, where a Mbus based high performance workstation could be designed and built in Australia, without the massive R&D overheads of designing a new chipset.
Croker and Kopp then reasoned though the strategic marketing problems to be dealt with. The basic technology offered the potential, at some expense in I/O expansion, to package compute performance characteristic of top tier workstations into a small and competitively priced product. The abundance of older Sun Microsystems SPARCstations in the customer base made it obvious, that the best choice for a board layout was an identical form factor to an existing machine. In this fashion, customers' existing workstations could be provided with a highly price competitive performance upgrade merely by swapping out the original Sun Microsystems motherboard with a GCS motherboard. The new workstation was code-named the Mongoose, this name was selected as the machine was to be a "Snake killer", a small and agile competitor to the then new HP Snake series of HP-PA workstations.
The first choice were the widely used low cost Sun SLC and ELC machines, which contained a motherboard embedded in the rear of the monitor case. Kopp subsequently proceeded to test an SLC power supply to verify whether it had the capacity to feed the required Amperes of current into the new chipsets and CPU modules. It did not, and moreover the convective cooling scheme in the SLC/ELC was simply inadequate to cope with the thirsty and hot new Mbus CPUs. So the Mongoose A workstation never got off the drawing board.
The next model up the range was the Sun IPC/IPX, a small footprint desktop workstation fitted with SPARCstation 1+ or SPARCstation 2 chipsets and monochrome or colour embedded frame buffers, respectively. The IPC/IPX had the volume, power supply capacity, layout and cooling capacity, if fan assisted, to fit either of the new chipsets, and either the new TI Viking/SuperSPARC or Ross Pinnacle/HyperSPARC superscalar SPARC Mbus CPU modules.
Kopp immediately commenced on the design of the IPC/IPX form factor Mongoose B motherboard, based upon the Nimbus chipset and the Nimbus SPARCstation 2 (SS2) layout board design. The Nimbus chipset used a unique and proprietary Mbus frame buffer, rather than the Sun standard Sbus BW2/CG3 used in the IPC/IPX, and differed significantly from the IPC/IPX and SS2. The electrical design was completed, when Nimbus ran into difficulties with investors and the future of their chipset was placed in jeopardy. With the prospect of having a motherboard but no available Silicon, the Mongoose B was laid to rest. The Mongoose B, like its earlier sibling, never got off the drawing board.
After much contemplation, it was decided that the Twinhead Gemini/TW1 chipset was the only remaining option. Twinhead had designed their chipset in Silicon Valley with a generous Taiwanese government subsidy, and were keenly interested in finding a customer for their Silicon to satisfy their government requirement that the taxpayer's investment result in some useful exports. At that time, Twinhead began to ship an SS2 "pizzabox" format workstation using the Gemini chipset and Cypress CYM-6000 series Mbus scalar CPU modules.
The Twinhead option was carefully considered by Croker and Kopp, understandably cautious after the failure of the Nimbus based Mongoose B project. Clearly Twinhead had the ability to produce viable Silicon, indeed their chief engineer was a former member of the design team which created the chipset for the Sun Microsystems Galaxy server machine, the first multi-processor server released by Sun Microsystems. The Galaxy was an outstanding commercial success for Sun Microsystems.
However, some significant risks were evident. Twinhead's board level design was clearly inferior to the chipset design, and their SunOS 4.1.2 4.3BSD operating system port had serious problems, the resolution of which was evidently not within the grasp of Twinhead's operating systems group. A Mongoose based on the Twinhead chipset would almost certainly require a completely new board level electrical design to meet the robustness and quality standards needed in the fickle workstation market, and a new operating system port would be needed, as Twinhead had failed to exercise any discipline in revision control and the existing SunOS port was beyond repair.
After much careful consideration, Croker and Kopp opted for the Twinhead chipset, negotiated access to Twinhead's Valid Logic (Cadence) electrical schematic database for the motherboard design, and Kopp made his third attempt at producing a SPARC motherboard.
Once the Twinhead schematics arrived, the expectation that a completely new motherboard design would be required was indeed confirmed. The Taiwanese board level design did not conform with a number of Sun's electrical design rules and conventions, and employed non-standard electrical interfaces on most of its I/O ports. To produce a high-quality design which was to be as electrically robust as Sun Microsystems' IPC/IPX motherboards would require a new motherboard electrical design, from scratch.
Evidently, more manpower was needed, and Croker tasked his Chief Engineer with finding himself an assistant. After screening thirty six graduates from a number of Melbourne institutions, and subjecting the best of these to a rigourous two hour fundamentals exam, and further screening applicants through practical tests, Ross Bennett, BSc BE(hons), honours first class, a Monash Electrical Engineering and Computer Science dual major graduate was hired for the task. To this very day GCS apply the same rigourous selection procedures for job applicants.
The electrical design and printed circuit board layout of the new Mongoose C board took three months, with Barry Olney, formerly of QPSX, doing the Cadence (Valid Logic) component libraries, schematic entry and board layout to Kopp's tight specifications. Olney was no stranger to this engineering regime, ten years earlier at the University of Western Australia he endured Carlo Kopp the engineering honours student.
While this transpired, the Mongoose D was conceived. This new Mongoose was to use a low cost PC plastic case but retain the electrical design of the Mongoose C. Both the Mongoose C and D would contain an embedded version of the existing GCS enhanced CG3 frame buffer. Both the Mongoose C and D were built as 8 layer boards with over 1000 electrical nets, making them the most complex high density boards designed in Australia, to that date.
The first two Mongoose C prototypes were built in July, 1993, and took several weeks to debug. The Twinhead chipset displayed a great sensitivity to clock skew, and the unfortunate Bennett spent many late hours attempting to empirically determine the undocumented timing idiosyncrasies of the Twinhead chipset. It was later determined that the skew problem related wholly to a floating pin on the Mbus/Sbus controller chip, this in turn resulting from an error in Twinhead's schematic. Without access to Twinhead's jealously guarded chipset electrical internal design, Kopp and Bennett were left guessing. Without the capital to afford complex chip level test equipment, GCS had to tackle the problem with a two channel analogue oscilloscope. The hired HP 1651 Logic State Analyser was too slow for the task.
In August 1993, the first prototype Mongoose C motherboard came to life, with a half speed 20 MHz heartbeat animating its soul. History was made, Australia's first home grown SPARC machine was alive !
Once the hardware was working, Mark Goroff, a Los Angeles consultant who had previously ported the SunOS operating system kernel and boot PROM to the Nimbus chipset arrived for a week in Melbourne to port the SunOS kernel and Open Boot PROM, the latter written in Forth, to the new motherboard. Assisted by Kopp, Goroff got the kernel and boot PROM running in 6 days. Kopp then completed the kernel port and conducted regression testing for several weeks to weed out any bugs, which Goroff rectified in Los Angeles. Using the Internet, physical distance became irrelevent. The port included fixes to accommodate the TI SuperSPARC and Ross HyperSPARC superscalar CPU modules, as well as the older Cypress scalar CPU modules.
The Mongoose C prototypes made their public debut at AUUG 1993, running with 40 MHz Cypress scalar CYM-6001K-40 CPU modules. Kopp presented a Work In Progress paper which sadly received very little interest from the audience. It was a disappointing experience for the GCS design team, who had worked long and hard to produce the first Australian SPARC workstation. Croker stills recalls with some amusement his comment to Kopp at that time: " Nobody in that audience was qualified to appreciate what we had achieved, so you should not be surprised that they were not very interested". Certainly the only praise GCS received at AUUG was from a number of visiting US Unix industry luminaries, some of whom had gone through similar development project experiences in the past.
Once the hardware and operating system port were proven with about a thousand hours of diagnostic regression testing on the two boards, the production revision of the printed circuit board layouts was commenced. The lengthy prototype "buglists" were fixed, production details such as pin access for automated production testing were incorporated, and production board stencilling was added. A comprehensive technical manual and documentation package were produced.
The first production batch of Mongoose C boards, given the marketing name MX, were manufactured by LSE in Sydney in early 1994. The production MX Mongoose C boards came to life on Australia Day, 1994, their little hearts beating at 40 MHz for the first time.
The total development cost of the Mongoose C and D machines, and the aborted development of the Mongoose A and B, reached about AUD 250,000. Compared to similar US projects which typically cost in excess of USD 2,000,000 , the Mongoose was without any doubt a bargain.
Carlo Kopp departed GCS in early 1994 to complete his MSc at Monash Computer Science and develop a career as a consultant, subsequently returning to Monash to do a full time PhD under the supervision of Professor Chris Wallace. Ross Bennett assumed responsibility for design maintenance of the MX machines. In excess of 200 units were produced in Australia, providing sufficient return on investment to cover its development cost, and to finance the follow-on product, the MX Peacemaker workstation.
Sadly, the reluctance of the Australian investment community to provide genuine venture capital meant that the full commercial potential of the Mongoose was never realised.
The MX Peacemaker was based upon Sun Microsystems' SPARCstation 20 chipset, repackaged into a 5.25" disk drive form factor. Bennett carried out the design of this machine in 1995, with Olney providing the production engineering input and board layout. Where the Mongoose did not achieve its lofty commercial objectives, the Peacemaker most certainly did. It received an IEAust Engineering Excellence Award, and was subsequently licenced to a US company, jointly owned by Fujitsu of Japan and Sun Microsystems of Silicon Valley.
The licence manufactured machine, a US or Canadian built MX Peacemaker, won the Best Performance category in the 1997 US AIM "Hot Iron" Awards for a machine priced under USD 10,000. This was the ultimate realisation of Geoff Croker's vision of an Australian built world class Unix workstation machine. Sadly it is being manufactured and marketed overseas, with 30,000 units delivered at the time of writing.
The story of the MX machines illustrates what can be achieved by applying the rigourous academic training of Monash University Computer Science postgraduate research and honours degree courses to the demanding environment of the contemporary computer industry. It is also a good example of tight project management, clever risk management, world class computer engineering on a shoestring budget, and the fundamental value of critical argument in the management decision process.
Geoff Croker's comment aptly sums this up: "The engineering design of computers is not a team sport. What counts most is persistence, rigourous discipline instilled at a university level, and smart technical thinking. We need real achievers if we want to achieve real things."
Text © 1997, Carlo Kopp
Artwork © 1997, Graphics Computer Systems, Carlo Kopp
All trademarks, product names and company names are copyright their respective owners. Use of any of the material in this paper, in printed or electronic form, without the prior written consent of the author will be regarded a breach of copyright and damages will be sought.
A DISPLAY CASE CONTAINING PROTOTYPE HARDWARE FROM THE MX DEVELOPMENT PROGRAM IS NOW SITUATED OUTSIDE THE CSSE SEMINAR ROOM IN BUILDING 26, CLAYTON CAMPUS.