How Hitachi Aced the Mainframe Market

This is dedicated to all you NT nerds and Unix hackers who think the mainframe is virtually dead. I recently visited Hitachi's mainframe plant down in Shibusawa, south of Tokyo, and learned the story behind Hitachi's remarkable success in tripling its sale of mainframes over the past three years. Meanwhile, IBM has announced new models of a low cost mainframe that break the 1,000 MIPS (millions of instructions per second) barrier in general purpose computing.

Back in the late '80s the trade press began publishing stories about the imminent death of the mainframe. The cool new computing paradigm was the client/server. Mainframes became branded as the big, fat slobs of the computer industry. They were stigmatized as monstrously costly to buy, run and upgrade, and as for getting a decent-sized application written and debugged, well, you could expect Rip Van Winkle to be up and running first.

Companies began ditching mainframes in favor of cheaper Unix servers, stringing PCs off of them to spread computer power into every nook and cranny of the enterprise. As a result, worldwide shipments of mainframes peaked in 1988 at just over 4,700 systems; six years later, shipments were down to a bit over half that figure. Big Iron was descending into a rusty grave. But in 1994 IBM switched strategies when it introduced a new type of mainframe based on CMOS (complementary metal-oxide semiconductor) processor units, the same silicon fabrication process used in making the microprocessors inside our PCs. Until then, processors had been based on ECL (emitter-coupled logic) technology, which was ultra-fast, but expensive to fabricate. CMOS processors are slower, but can be made more cheaply. IBM was able to compensate somewhat for the slower speed by adding more CMOS processors to a system and connecting them in parallel.

Fujitsu soon followed IBM and introduced its own CMOS parallel systems, as did NEC, which had already begun shipping low-end CMOS parallel mainframes ahead of the IBM launch.

The remaining mainframe maker was Hitachi. Prior to its competitors' announcements, Katsuro Wakai, general manager of the company's mainframe group, had begun experimenting with new processor designs that combined both ECL and CMOS. "We found designing such a chip was not so difficult, but manufacturing it would be, I thought," recalls Wakai. In 1993 he took the idea to Akio Anzai, head of the device center that makes silicon especially for the mainframe group. Anzai said fabricating such a chip would not be difficult at all, but surely designing it would be. "When I heard that, we shook hands on the spot," said Wakai.

And so a hybrid chip came about that used ECL for the fast processing of instructions, and CMOS for handling the storing of data. Compared to full ECL, the new chip reduces heat and power needs, and because fewer hybrid chips could be used to get the same performance as a high-end system with many CMOS processors, the hybrid system significantly impacted the total cost of ownership. Wakai dubbed the new chip ACE.

The timing of ACE was perfect. Yoshio Oshima, head of Hitachi mainframe development, had been surveying customers' future needs. He found that while the majority of customers were willing to switch to CMOS mid-range systems in order to reap the cost savings, there were still many large corporations who wanted very powerful machines with the minimum of chips inside. "In terms of total cost of ownership and response time, many customers still wanted machines with a few fast processors, rather than (having to manage) a system with many smaller processors," agrees Wakai.

Before Hitachi launched its ACE-based systems in 1995, it had between 5-7% of the world's mainframe market; today, it claims between 18-20%. Not all this sales growth comes from ACE machines. Hitachi has licensed IBM's CMOS technology to build its own CMOS mainframes, but in numbers of processors, ACE outsells Hitachi CMOS more than two to one. Since the ACE launch, IBM has been scrambling to boost CMOS performance, and has gained on Hitachi each year. Now IBM claims its latest machine peaks out at 1,014 MIPS, verses the 971 MIPS scored by Hitachi's top hybrid system.

That won't continue for long, declares Oshima. "My personal goal for next fiscal year is a new ACE system providing over 2,000 MIPS." If he delivers, that would double the performance of the IBM announcement and put Hitachi way out ahead again. With this kind of competition and performance increase, combined with the mainframe's famed reliability and security, not to mention the rise of hungry applications like data warehouses, data mining and e-business, mainframes are clearly not going to die any time soon.�

When John asked Santa for a Big Iron last Christmas, Santa left a 1950s steam iron in his stocking. If you think you can pun better than Santa, e-mail John at boyd@gol.com.

Back to the table of contents