The consensus is that IBM will give its 3090 top-end mainframes a mid-life kicker sometime in the first four months of 1987. The earliest date we’ve heard is this month, the latest is April. Whenever it comes, it will be an almighty kick, booting the performance to as much as 115 MIPS on one forecast, as well as finally implementing those long-awaited 6 Mbytes-per-second channels, with the limit raised to between 128 and 192 depending on how many CPUs the company decides to couple together this time around. MVS/XA has been written to run over 16 processors, and on that basis, the four CPUs of the 3090/400 is positively parsimonious. Most observers expect a six-CPU 3090/600 in the next announcement, but some money is going on an eight-CPU configuration, if not this time around, then in the final major facelift before the launch of Summit, which is expected to arrive late in 1989 or early in 1990.
Main memory is also expected to be expanded substantially, shooting up to a maximum 768Mb, which on a six-processor configuration would imply a 128Mb uniprocessor. The second generation 3090s are expected to use significant new circuitry in a repeat of the process that produced the X models of the 3081; after pressure from existing users, IBM offered comparable X performance to 3081s in the field, but the upgrade did not actually make those machines Xes, it simply enabled them to deliver comparable performance. IBM certainly needs the fillip that the new machines should give its sales effort – even in Japan, which had been a bright spot in the general gloom shrouding the IBM empire until mid-year, after which business turned down so badly that a forecast of 18% growth that seemed attainable at mid-year, had slumped to no growth by year-end. Japanese users are MIPS junkies, so a big boost in raw performance should do IBM Japan no harm. But eyes are now lifting beyond Sierra and onto Summit, and William Husband, a consultant to Meridian Leasing of Dearborn, Illinois has been expounding his forecasts for Summit and beyond in both the Computer Economics newsletter and in InformationWeek. He looks for the first Summit processors to appear in October 1989, offering 1.5Gb of main memory (the 32-bit MVS/XA can address only a maximum of 2Gb), 256 channels running at up to 9 Mbytes-per-second, and a price per usable MIPS of $90,000, down from $150,000 for the current 3090s.
But in maximum configuration, he sees Summit being a 16 processor complex with external fibre optic channels capable of serving serveral processor complexes, organised as two groups of eight single-image processor complexes linked by an inter-processor control and communications processor, which itself will be a 3090-class CPU running substantial parts of MVS. By 1991, he sees Summit getting a midlife kicker, taking main memory to 4Gb, performance to 250 MIPS, channel speed to 12 Mbytes-per-second, and number of channels to 384. Husband has even looked beyond Summit, visualising a machine for 1993 that would initially come with 16Gb main memory, rising to 32Gb or 64GB, 384 channels rising to 1,000, 300 MIPS, rising to 400 MIPS, and a price-per usable-MIPS of $45,000. He also reckons that main memory prices, using 4Mbit chips, will have crashed to $700 per megabyte from today’s $8,500 per.With 32Gb or 64Gb of main memory, MVS/XA is so far out of its depth that its not simply treading water as MVS/SP-3 did with 24Mb on a 3033, it’s sunk without trace long ago. So Husband postulates a successor operating system capable of managing all the complex resources that will be assembled into Summit and its successor – and suggests that a move to sell some of the code of DOS/VSE outright to users is a forerunner of things to come – that in due course, IBM may well start selling MVS/XA code, which would be a clear indication that it was on its way to the dustbin of history, following VS1 and VSE. All discussion so far has implied a homogeneous processor complex, simply more and more of the same. But as long ago as 1982, IBM was talking quite clearl
y of including special function processors within the complexes managed by MVS. As well as specialised database and vector processors, these could also include specialised processors that ran alien code – and a System 38 processor would be the most obvious example, although by the early 1990s, major Personal Computer applications will be so numerous and so ubiquitous and necessary that it may well be convenient to integrate an 80X86-compatible processor within the mainframe complex, as well as one for the RT Personal, if that machine ever garners a critical mass of applications. As for the operating system that will manage all these disparate resources, the only sensible start-point for IBM would seem to be the oft contemptuously-treated but ever-willing VM.
The machines in Husband’s forecast are of course already in existence in some form. Summit must already be a substantial way along the design process, with IBM beginning to make the hard decisions about which of two or three alternative circuitries and packaging techniques to use, while early designs of its successor must already have been roughed out on IBM drawing boards or computer-aided design systems and their databases. Husband has used a perfectly valid technique for his forecasts, using experience of the past to project the future. The problem with the technique is that it is the same one which three years ago projected IBM to be a trillion dollar company in 2004. It now seems unlikely that IBM will achieve that mythical figure until decades into the twenty-first century unless inflation reaches South American proportions in the meantime. And what exactly is the history being projected? Opportunistic 360, the cynical 370 con trick, emergency 3033 in response to the threat from Amdahl,the 3081D false start, successful launch of 308X about a year later, the X mid-life kicker, leading to smooth progression into 3090. But halfway through the life of the 3090, IBM is in total disarray as all its plans for rapid growth founder on the rocks of a buyer’s strike by users, and big question marks over a whole swathe of strategies that had been cast in concrete. It is by no means certain that users will rush to order the second generation 3090s in the numbers IBM needs, and if they fail to do so, a complete recasting of the company’s future top-end strategy may well become essential.