Almost unnoticed, the computer industry has split into two price-performance curves that are diverging at what must feel like the speed of light for those who are stuck on the wrong one. The high-price modest performance curve is occupied by the surviving proprietary architectures – led by IBM’s 3090, 4381, 9370, AS/400, DEC’s VAX/VMS, Unisys […]
Almost unnoticed, the computer industry has split into two price-performance curves that are diverging at what must feel like the speed of light for those who are stuck on the wrong one. The high-price modest performance curve is occupied by the surviving proprietary architectures – led by IBM’s 3090, 4381, 9370, AS/400, DEC’s VAX/VMS, Unisys A and 1100, ICL’s VME and the various GCOSes at Bull. The low-price, soaraway performance curve is occupied by RISC architectures plus the two surviving mass-market complex instruction set microprocessor families iAPX-86, MC68000, in all cases running one or other version of Unix. Initially it was only at the low end of the market that users began deserting proprietary architectures for Unix in droves, so that the companies that suffered most were the low-end minicomputer manufacturers, but the trend is now shooting up into the mid-range, which is one reason why the IBM 9370 has been such a complete failure as a new-user small business computer, and why AS/400 sales are beginning to slow: a good half of the System/36 base could well take two looks at the price to be climbed to to get aboard the AS/400 line and seriously consider a switch to Unix instead.
The switch has been a disaster for the likes of Data General, Prime Computer, Norsk Data, and it has now begun to lay waste the Wang Laboratories VS and the Nixdorf Computer 8870 bases. By contrast, Hewlett-Packard Co, which saw it coming early and planned for it accordingly, is flourishing. The problem for the pushers of proprietary architectures is not initially that they lose their existing customer base, but that they find almost overnight that their bases have stopped growing. Since no computer manufacturer has ever budgeted for no growth until the bailiffs are banging at the door, the sudden loss of new business has a disproportionate effect on profits, and the dismal routine of lay-offs, plant closures and cut-backs to try to downsize the business and get it back onto an even keel comes into play. The problem then is that the cuts mean that the existing customer base gets less attention than it expects, much less than its Unix-using neighbour on the block whose supplier is on a rising curve, and very soon the straitened means of the proprietary architecture vendor means that there are not the resources to bring the next generation of hardware and operating software out in a timely fashion, and so the loyal base begins to erode as well, at which point the merchants of death in the shape of the mergers and acquisitions pushers and the lethal leveraged buyout wreckers start massing like malignant Mafiosi.
Strikingly, DEC seems to have spotted the mortal danger to its core business just in time – and its recognition of the urgency is underlined by the fact that decided that it couldn’t afford to wait for its own designers to come up with a suitable RISC architecture, and it went out and bought one in from MIPS Computer Inc, a reassuringly inoffensive-looking Sunnyvale company that did not seem to pose a threat to anyone. That remains to be seen, but having bitten the bullet, DEC has leaped aboard the low price, high performance curve without a backward glance, and is just keeping its fingers crossed that the MIPStations will start delivering the enormous volume it needs to keep margins healthy fast enough to make up for the inevitable tailing off of the VMS line whenever that starts to bite. And early signs are that DEC may have made the switch just in time, but it seems highly likely that the forthcoming Aridus VAX CPU will be the last VMS machine from DEC that will be more powerful – at least for a month or two – than its most powerful Unix machine: there is a 60 MIPS ECL implementation of the MIPS Computer R-series RISC family in the pipeline. And even Tandem Computers Inc, the one company in the industry generally regarded as having a unique solution for high-volume transaction Zprocessing, is heavily hedging its bets and has signed to use the MIPS RISCs and is busily gearing up for its own plunge in
to Unix. The problem for IBM is that the industry-wide rush for Unix at the low end and the mid-range is starving it of future customers for its top-end mainframes, so much so that when the 4381 base has moved up to IBM’s next generation top-end mainframe line, there will be no source of new users to provide horizontal growth of the top-end base: all the available growth will come from upgrading the existing MVS base, which means that with the steady improvement in price-performance that characterises all architectures, whichever curve they are on, IBM’s mainframe business will go from slow growth through no-growth to decline in perhaps as little as 12 months.
And IBM could well be about to hasten the process: users are hearing hints that the company is thinking about adjusting its software licensing terms so that instead of attaching licences to machine model numbers it will attach thhem to physical processors, so that if a user partitions a 3090-500 into a 300 and a 200 and runs it that way, it will attract two software licences. IBM may have to think again on that one, because the rumblings of discontent about the sky-high cost of IBM mainframe software would almost certainly become a roar, and third party IBM software developers would start rushing to do versions of their applications to run under Amdahl Corp’s native 370 Unix, UTS, and the switch would be complete with 3090s being seen as plug-compatible Amdahls – Amdahl will license 3090 users to run UTS if they ask for it. Is it too late for IBM to retrieve the situation? It would cost the company two years of unprecedented pain and the switch would be almost impossible to manage convincingly. With its newest models in the RT line now over a year old, having been launched in July 1988, where – DEC launched its first generation MIPStations in January this year and has already come out with the second generation – IBM has lost the last shred of credibility it had in the Unix market, and it stretches credulity beyond breaking point to conceive of an IBM salesmen pushing an RT at any prospect that might even possibly be persuaded to take an AS/400. The genie is out of the bottle, and no amount of malevolent machination in the Open Software Foundation can now get it corked back in again