RAID and the collapse of IBM's mainframe storage business.

What were the technical and non-technical aspects of the collapse in IBM's DASD (mainframe disk) business from 1990 to 1995, from ~$9B and 80% market share to $1.5B and 35%?

Previously: a selected timeline of events, some Lessons on the causes of the collapse and some thoughts on RAID++, the next evolution in Data Storage.

Broadly, IBM DASD's business didn't fall alone, between 1991 and 1993, IBM lost $16 billion, due largely, but not entirely, to problems with their Mainframe business. They were to exit the PC market 10 years later.

IBM in the early 1960's had "Bet the Company" on a radical new line of computers, the S/360, which went on to dominate the market and become known as "mainframes". At this time, the company was well over 75 years old at the time and already the subject of anti-trust consent decrees, having it's business regulated by the US Government. This wasn't a play by a start-up or an ambitious up and coming technology business: this was an astute and bold gamble by one of the giants in the field.

It worked and they, and the shareholders, reaped 3 decades of explosive profits and growths. IBM always paid managers and salesmen (mostly Men), handsome bonuses - they got rich as well.

There were two forces at work that IBM first managed to ride, then managed hold back for around 15 years, postponing the inevitable "Disruption" but at the price of creating a competitive "overhang", that, once triggered, leads to a rapid and profound collapse in the market:
  • high Gross Margins, 60+%, were the norm for IBM to maintain their very expensive and successful sales and marketing techniques, and
  • Bells' Law, accounting for the formation, evolution, and death of computer classes based on its technology type. A new computer class forms roughly each decade, establishing a new industry.
Bell's Law, initially developed in 1972 to forecast the evolution of DEC's PDP & VAX computers:
A computer class is a set of computers in a particular price range defined by: a programming environment ... a network; and user interface for communication with other information processing systems including people and other information processing systems. A successful class usually establishes a horizontally structured industry composed of hardware components through operating systems, languages, application programs and unique content ... that serve markets through various distribution channels.
While IBM's R&D had been increasing capacity & performance while reducing the cost of its DASD and from 1956 onwards (see timeline for examples), it's market wasn't the only one developing disk technologies.

The minicomputer and PC markets had been roiling and boiling away since the 1970, growing in volume, revenue and capability every year. No one vendor, like IBM in mainframes, had been able to dominate the market, allowing them to extract massive premiums for their products.

DEC, first with the 16-bit PDP-11 and then the 32-bit VAX lines, was the benchmark and did exceedingly well, but was unable to capture the market and prevent new entrants.

Part of this was due to the arrival in 1969 of UNIX, an Operating System (O/S) from a company barred from selling computers, AT&T. UNIX became transformative because it solved a number of hard Software Engineering problems, especially "scaling up" an O/S. Microsoft failed at this in 2003, just as Multics did in the mid-60's and IBM almost did with OS/360. The IEEE POSIX standard created the formal base needed for the industry to write cheaply portable software.

UNIX was small, well documented and possible to learn independently, portable and performed very well. Linus Torvalds went on to reimplement UNIX from scratch, creating Linux, the O/S adopted by Google for Android. The BSD UNIX distribution fuelled the evolution and development of the system in the days of VAX's and super-minicomputers, ending by becoming the base for OS/X and iOS, the most successful Apple systems ever.

This means that now the most popular Operating Systems on the planet, by units and revenue, are POSIX compliant, i.e. Unix-derived.

First UNIX powered the minicomputer revolution, then Microsoft with MS-DOS then MS-Windows, powered the Intel PC revolution. Without an Operating System and Application Software, hardware isn't useful or interesting to customers. Software sells Hardware and portable Software allows customers to substitute products for others. Software drives competition by vendors and utility for customers.

In 2007 (iPhone) and 2010 (iPad), Apple leveraged their portable POSIX environment to create two whole new Computer Classes, smartphone and tablets, that have seriously impacted the existing PC market and caused financial challenges for Microsoft.

IBM is a remarkable company, not because it failed to see a wave of technological change about to break-over it and destroy one of its core businesses, but because it survived the "market restructure" of mainframes (and PC's) when no other computing business has survived a similar Market Disruption. There is an incredibly long-line of once-great computer companies that no longer exist: DEC and Control Data are two on my list. Burroughs and Sperry-Univac, in the guise of Unisys, are still sold, but the company is around 10% of the size it was in 1990.

By the time of the 1991 crash, IBM could trace a lineage in Data Processing and Storage back 100 years, through a large number of changes in technology: punched cards, mechanical calculators & tabulators, vacuum tubes, transistors, Integrated Circuits, Large Scale Integration and System-on-Chip (SoC) CPU's like Intel's 486.

The collapse of IBM's DASD business can't be viewed in isolation. It was as much financial as technological and market-restructure based and similar collapses affected many of IBM's business lines. The change in the mainframe market cause the second Dept. of Justice anti-trust suit of 1968 to be abandoned in 1982. It took another decade for the full effects of both in-market competition (plug-compatible vendors like Fujitsu/Amdahl) and Bells Law out-of-market substitution (PC's and superminis) to wash through.

Gordon Bell comments like John Mashey, creator of the MIPS chip, that CMOS, the low-power technology used in microprocessors, started in 1971 on its path to replace the high-power ECL technology used in super-computers like Control Data and Cray's, and mid-power Bipolar used in mainframes and high-end superminis. The Cray-3 design was killed off in 1989 because of the impact of CMOS and the collapse in demand for super-computers.

The 1987/8 Berkeley RAID paper was a trigger event, but for anyone who'd worked with both mainframes and minicomputers, the pricing difference was well-known and stark.

EMC because the new dominant player through better business practices and marketing strategies. In 1990, the obvious "Killer App" (or device) for RAID was going to be the StorTek "Iceberg", later taken up by IBM as their RAMAC offering.

No comments: