Like every other mainframe manufacturer, IBM Corp is talking about the evolution of a high-performance transaction processor with Sysplex, concurrent maintenance and automatic recovery facilities. Arthur Parker, IBM UK’s Enterprise System manager, predicts that within five years, the mainframe will be an enormous data server handling a 20-fold increase in data, and incorporating technologies like image, voice and full motion video. Well, maybe. Some observers are less convinced about the mainframe’s future, saying that compression techniques and innovative technology at the workstation will erode the large systems base. Nonetheless, Amdahl Corp has said similar things in the past, and Hitachi Ltd reckons that the largest systems will be delivering up to 3 GIPS by the end of the 1990s. Hitachi believes that the move towards distributed computing and inter-corporate multivendor networking will create new roles for the mainframe. It foresees the host becoming a computing engine for user applications; a database manager for data integration; and a total resource manager.
Disparity
The demand for mainframe power has been growing at between 30% and 40% per annum, and there is no reason to suppose that will decline during the 1990s. However, uniprocessor performance increases by a mere 2.5% per generation, or at 15% to 20% per year. This disparity and the failure of hardware technology to keep pace means that Hitachi, and the rest, are reassessing issues like logic structure, architecture and software. In terms of components, faster circuits are emerging as alternatives to Silicon. ECL, CMOS and BiCMOS have dominated commercial applications, but despite economic feasibility, Silicon is nearing its limitation in speed and density. Although Hitachi’s next generation mainframes will be Silicon-based, the company is looking at Gallium Arsenide, compound semiconductors and superconducting devices. Superconducting devices, apart from being faster, will consume several orders of magnitude less power than Silicon, which will offer the potential for still higher circuit densities.
Mainframe vendors cite emerging technologies to persuade that they can build ever faster machines in 370 architecture and alleviate its manifold shortcomings with ingenious bolt-ons but are they confusing the medium with the message? Janice McGinn suspects that they are.
The switching speeds of superconducting devices are higher than those of ordinary semiconductor devices because they operate different switching mechanisms, and research is under way into different devices including the Josephson Junction device, superconducting transistors and the impressively-named quantum flux parametron. The Josephson Junction is a diode device, but the surperconducting transistor is a triode that can be used for various circuits and large-scale integration. Also, the use of electron beam lithography enables the fabriction of extremely narrow gates between superconductive electrodes which facilitates ultrafast superconductive switching. Gallium Arsenide and like materials enable electrons to travel much more quickly than in Silicon, and their ability to generate and sense light makes them excellent candidates for optoelectronic integrated circuits, essential prerequisites for optical computing. Although they currently lag behind Silicon, Hitachi believes that optoelectronic devices will make inroads into commercial computing applications during this decade. Nonetheless, it seems likely that the cost will remain prohibitive. Mainframe architecture is set to change over the next few years, and Hitachi is developing a range of special-purpose features to be integrated into the host. A sketch of the future mainframe might include an integrated array processor, more commonly known as the vector processing feature; an integrated database processor; application specific engines; hypercomputing facilities; cryptographic features; global shared files and memory; super high-speed channel networks; shared expanded storage; and much more powerful front-end processors.
VOS and U
nix
Hitachi – and it is presumably not the only one – has systems with most of those features, but they operate under VOS and Unix rather than MVS. However, Hitachi reckons that demand for mainframe systems will grow at no more than 5% per annum, with Asia growing at around 18%, followed by Japan with 14.8%, 10.9% in Europe and 9.8% in the US. The company is insistent that it won’t withdraw from the shrinking mainframe market, but it is keen to emphasise its Open Software Foundation credentials and is planning to focus on other areas like open systems, Unix workstations, software and leasing. The last, says Hitachi, will stabilise its revenues while matching user payout to income. Hitachi, like the other manufacturers, may be talking of leading-edge technology, but one indication that the company is not having the happiest of times with lower profits on flat revenues, is the time from ordering to taking delivery of a mainframe – down from six months to just two. Possibly a sign that Hitachi’s manufacturing and distribution system is super-efficient, but an indication, surely, that there is slack in the system and that machines are sitting around waiting to be bought.
