Historians of the computer industry may someday describe this as the year the disk wars began. Until now, companies in the disk business had battles. But the marketplace was divided. Skirmishes in one sector could scarcely affect conditions in another. For disk makers, those were the good old days. The storage subsystems being designed today […]
Historians of the computer industry may someday describe this as the year the disk wars began. Until now, companies in the disk business had battles. But the marketplace was divided. Skirmishes in one sector could scarcely affect conditions in another. For disk makers, those were the good old days. The storage subsystems being designed today for every type of computer – from the one on your desk to the one in the glass house – are based on very similar disk drives. The capacity of these drives spans scarcely more than one order of magnitude. The other salient characteristics of the disks, such as average access time and data transfer rate, are even more closely grouped. The reliability of all the emerging disk drives is amazingly high and once again the difference between the least costly and most exotic units is very small. Academic understanding of the means by which groups of disks may be combined to yield data libraries of enormous size and near perfect subsystem reliability is widespread. Practical machinery based on this knowledge is readily available for moderately demanding environments, such as network servers and midrange systems, and can be obtained from a large and rapidly growing group of vendors. Although the technology used in disk array subsystems cannot so easily be scaled up to meet the demands inherent in very large computer systems, it is nonetheless clear that data storage in the mainframe world will soon undergo a transition to vastly improved technology. Up to a point, this pattern of change resembles that in the processor business. The similarities persist from the bottom of the market up to the mainframe class.
Desktop machines with main processor boards that require only a few chips are the norm. Mid-range machines require more components per central processor, but they are increasingly based on a single complex circuit or a small set of circuits. The chip makers may not have solved all the mechanical problems (particularly heat dissipation) that come with advances in circuit complexity, but the semiconductor industry seems assured of fulfilling all its promises without undue delay. Advances in memory, which can cost as much as a basic processor, have kept pace with the other semiconductors used in computers. But the technology that would enable the makers of large processors to take advantage of the kinds of circuitry used in smaller systems remains elusive. It is software technology. To date, the industry has had a much easier time building a machine with many engines than creating a way for such a machine to be used in commerce. As a result, the mainframe world remains isolated from the rest of the computer business. And mainframe customers only get the full economic benefit of advances in other industry segments when the vendors fear revolution is imminent. As such, revolutions are discussed by customers considerably more often than they are carried out, the mainframe makers are not in any danger of having their processor businesses collapse in the near term. They are threatened, but they still have time to invent new ways of building large machines that will keep them both profitable and competitive with alternative computing strategies. Still, these same companies (and the specialist organisations that make storage subsystems) do not have any chance of avoiding a potentially fatal collision with progress. The companies that make disks for the merchant market efficiently produce drives by the tens of thousands. A large and growing chunk of their production is devoted to disks with capacities from 500Mb to 3Gb. Customers that buy lots of these disks pay in the vicinity of $1 per Megabyte for the machines.
By Hesh Wiener
Within a couple of years, predicts Jim Porter of Disk/Trend, the selling price of a Megabyte of storage capacity will be well below 50 cents. And the capacity of commonly available disk drives in the 3.5 form factor (like those used in personal computers) will range to 5Gb and beyond. The speed and reliability of miniature disks are already more than ad
equate for mainframe applications… and they get better all the time. For disk makers, including AdStar, there will be no respite. The volatile mix of high demand, fickle buyers and industry standards has fostered a relentlessly competitive market. At the same time, there have been a number of technological developments that could be even more important to mainframe customers. Some of the companies that develop microprocessors and the other complex circuits used in computers are working on special processors that can manage disk arrays. Motorola, to name one supplier, has a chip that enabled Digital Equipment to announce a very nice little disk array last month at a list price of $5 per Gigabyte. Such circuits will soon evolve into mainframe class chip sets, as RISC chips have. While AdStar is fiddling with 10.8 disks for 9021 customers, its empire is burning. The new 3390-9 disk brilliantly addresses the market for medium performance, high capacity disks and the 3990-6 controller announced with it is an unquestionably better machine than the 3990-3. So what? EMC is gobbling up IBM’s best customers by providing a high performance disk subsystem that would also be very cheap if anyone else had the gumption to compete. The competition that counts isn’t likely to come from StorageTek and AdStar. They are rushing to perfect fail-safe disk arrays.
But both may find that their business plans are a lot more prone to failure than their inventions if, as we suspect, they think customers will pay a premium in cash and a penalty in performance to get the benefits their products promise. While every big customer would happily pay $10 to house a Megabyte of data in utmost security, few will pay anywhere near that rate to store the Terabytes that need protection. Even customers with data worth that cost will not be cheered if they are told that RAID 5+ arrays deliver sluggish response unless they are augmented by gargantuan controllers. We don’t think EMC’s current products are the best the industry can do, but the company’s sense of realism is remarkable and refreshing. The most impressive result of EMC’s success to date is the company’s destruction of a belief whose time had gone. That belief – once true, now merely a vanquished myth – is that mainframe customers would not buy a competing product unless it was a near-exact imitation of the corresponding IBM offering. EMC’s Symmetrix may look just like an AdStar subsystem to a channel or systems program, but that appearance is merely a cheap (thankfully) illusion created by clever technology. Henceforth, companies after a piece of the mainframe market can imitate IBM when IBM offers the best practical solution (which is quite often the case) or merely pay proper homage to IBM’s products when an alternative product design is superior. IBM will not so easily shake off rivals who focus on the essence of a product requirements rather than the particulars. And customers, getting an education as well as some interesting options, seem quite pleased that somebody has finally culled the wheat of their requirements from the chaff of IBM’s offerings. (C) 1993 Technology News of America Co.