Hitachi took the mainframe market completely by surprise two months ago (Computer Weekly, 9 March 2000) by announcing its temporary withdrawal. The decision leaves IBM and Amdahl as the only two companies actively marketing System 390 mainframes today.
HDS Europe systems manager Chris Douglas says, "Hitachi has decided to shift resources to a new generation server, code-named Hercules. It will bring it to market in the second half of 2001." As for existing systems, "We are reducing the volume of Trinium and Pilot systems. We are continuing to market them to our existing customers, but we will not necessarily be doing market generation activities into the IBM base," Douglas says. Hitachi's other European distributor, Comparex, has taken the same position.
Hitachi's decision signals the end of the road for its Advanced Cmos-ECL (Ace) technology and for the Skyline and Trinium mainframe ranges based on it. The company is falling into line with all the other mainframe suppliers, both within and outside the IBM world, by turning to Cmos logic circuitry as the base technology for its next generation. Hitachi was always going to have to do this eventually, but the end for the bipolar systems has come sooner than the company expected.
The Ace story started in the early 1990s. At that time there had been a catastrophic decline in mainframe sales, fuelled by a worldwide recession, plus the ready availability of open systems selling at a much lower price than mainframes. A major reason for the price differential was that mainframes were constructed from technology purpose-built for high performance IT systems. The manufacturing costs were higher than for the commodity components used in smaller systems, and so the street prices were higher.
In 1992, IBM accepted that customers would no longer pay premium prices for mainframes, and that it had to change its cost base. So the company took the decision to abandon traditional mainframe ECL circuitry in favour of Cmos. Shortly afterwards, Fujitsu, Amdahl's main technology partner, announced it was doing the same.
Cmos offered three main advantages. First, it was much cheaper, because it was being used in a variety of systems and was being produced by a large number of chip manufacturers. Second, the pace of development promised to be faster, because so many more companies were investing in it. Third, it offered substantial environmental benefits - reduced cooling, heat dissipation, power consumption and floor space.
Against this there was, in 1992, one huge disadvantage. The Cmos technology of the day generated nothing like the same processing power as bipolar technologies such as ECL. When IBM introduced its first generation Cmos mainframes in late 1994, the uniprocessor was rated at 12 mips, while the last of the company's ECL uniprocessors offered 60 mips. Even at the much faster rate of technology development offered by Cmos, it was clearly going to be some years before a Cmos system could deliver the same power as an ECL system.
IBM and Fujitsu/Amdahl's decision to abandon Cmos at this stage was thus a bold and risky one. The other player in the market, Hitachi, adopted a quite different strategy. The company decided there was mileage left in ECL and embarked on a development project to produce circuitry which combined the speed advantages of ECL with the environmental and cost advantages of Cmos. This became Ace.
As a result, when Hitachi Data Systems and Comparex launched mainframes built from Ace circuitry as the Skyline and the M2000 respectively, they were alone in the market. When initially launched in 1995, a top-end Skyline delivered 780 mips, compared to the 480 mips of the top-end bipolar machine from IBM and the much smaller 160 mips of the contemporary top-end Cmos machine.
Hitachi also had a second line of attack - its own range of Cmos mainframes. Rather than pursue two separate development efforts, it bought Cmos processors from arch-rival IBM and mated them with some of the I/O technology from Skyline to produce the Pilot series (sold by Comparex as the C2000).
This two-pronged strategy was very successful. Throughout the mid-1990s Hitachi gained market share from both IBM and Amdahl, to the point where it ended up with well over 20% of the total market. Furthermore, because Skyline was unique, HDS and Comparex were able to command premium prices for it - at 10% to 15% above the rate for Cmos systems.
This golden period lasted until IBM's G5 systems arrived in late 1998, followed shortly afterwards by Amdahl's Millennium 800 in early 1999. These two ranges were the first Cmos systems to offer more power than the predecessor ECL machines, and eroded much of Hitachi's advantage. Only at the very top of the Skyline range did HDS and Comparex have a system bigger than anything IBM or Amdahl offered. At this point, Hitachi started to lose market share.
Nonetheless, Hitachi continued to invest in Ace and unveiled Ace II systems in early 1999, with general availability last September. HDS called them Trinium, Comparex M3000. They went up to 12 processors, but 16-ways were also promised, and HDS announced the availability of these systems as recently as February.
But Hitachi was continuing to lose market share. One estimate was that sales fell from 21% of the market by value in 1997 to 14% in 1998, and has since fallen further to a level below 10%. Not only has Hitachi not been selling as much, but many of the Skylines installed have been replaced.
In addition, the company suffered a setback with its Trinium systems, having to revise downwards the promised performance levels by about a quarter last October. Trinium has, in any case, not proved popular with users, mainly because the environmental disadvantages of ECL mean a significantly higher cost of ownership. Only four have been installed in the whole of Europe in the year since its launch. Pilot has also not sold particularly well. And Hitachi needed a new development strategy for this part of the range, as the OEM agreement with IBM ran out at the end of last year. All these developments led to the decision in March to put mainframe marketing on temporary hold.
HDS is promising it will come back in with Hercules in 2001. Douglas says Hercules will be based on Hitachi's Cmos-based technology, using the experience of Ace to get faster technology, but he is unable to be more specific.
He continues, "[Hitachi is] looking at functional requirements, especially multiplatform: not only S/390 but Unix and NT, running in separate logical partitions."
Douglas argues that this is a logical continuation of the policy followed with the company's 7700 disc sub-system. "Two years ago 7700 was entirely S/390: now 50% of 7700 sales are for open systems. Hercules follows the same philosophy".
In the meantime, IBM and Amdahl can hardly contain their glee. Both companies have been enjoying good sales recently, apart from a hiatus around the turn of the millennium because of the Y2K freeze. The mainframe market is not growing as fast as the Unix and NT markets, but is still growing at something like 60% a year, which is more than enough to be viable.
For Amdahl, the Hitachi withdrawal is particularly good news. The firm slipped badly in the mid-1990s when it went three years without producing any new models, and has been playing catch-up ever since. Now Amdahl is back in the position it enjoyed for 20 years and built its name with: that of having the most powerful mainframe on the market. The top-end 2000E (the 16-way 2168E) is rated at 2,007mips, compared to the top-end IBM 12-way 9672-ZZ7 of about 1,600mips.
For the future, according to vice-president of server business and marketing Carol Stone, "We will have a mid-life kicker at the end of this year, which will increase performance by 25%. Then there will be a completely new machine for the end of 2001."
All of this is underpinned by a substantial development effort. Amdahl gets its base technology from parent company Fujitsu, but nonetheless employs 300 engineers at its California headquarters dedicated to System 390 work. Amdahl director of system architecture and performance Mark Gehoon says, "The things we develop are partitioning software, system management software and control software, plus some connectivity hardware. We develop firmware for that. With firmware and system management software we are much closer to customers. Fujitsu's expertise is in the bricks and mortar."
IBM, meanwhile, is beginning to focus its attention on the next big step forward, the move to 64-bit addressing. That will come with a system code-named Freeway at the end of this year, probably November. Support for it will be built into the next release of the operating system, R10, due in September.
Beyond that, says IBM System 390 consultant Peter Norris, will be "the microcode component of Freeway, which will include lots of new functions and features I can't tell you about, affecting the PR/SM and sysplex layers of the machine. That is what will turn the machine into Freeway".
Beyond that again, is the move to virtual 64-bit addressing, about which IBM is currently saying nothing. That is a bigger change, as it will involve changes to all the subsystems like DB2 and Cics. To judge from the length of time it took IBM to "sysplex-enable" these subsystems, full 64-bit capability will not be delivered until 2003 or 2004.
That is the challenge Amdahl will have to meet. Gehoon is relaxed about it. "It's business as usual," he says. "That will be the third major architectural change in the past 12 years, after XA and ESA. By the end of the year, we will respond on 64-bit capability. Around 3,000 mips is the limit at which 31 bits runs out of gas. We will be under 3,000 mips till 64-bit arrives. We don't know the specifics, but we do know how it has to be put together. We did that with XA and ESA. So we have it 90% now, and we'll produce the other 10% when we see the specs. Software takes longer than hardware, which affords us a window of compatibility lag."
The move to 64-bit is a challenge Hitachi will also have to meet, and it will be harder for the Japanese company as it is trying to achieve other objectives at the same time. There must be doubts about the ability of Hitachi to produce a system such as Hercules by the end of 2001, and the absence of detailed information about it does not inspire confidence. The company has a major marketing challenge in convincing the System 390 user base of its continued commitment to that market if it is not to disappear from sight altogether.
The slow move from ECL to Cmos
1992 - 4Q:IBM decides to abandon development of bipolar technology for mainframes
1994 - 3Q:IBM delivers first range of Cmos mainframes, the 72
1995 - 3Q: HDS announces first Ace technology mainframes, the Skyline series
1996 - 3Q:Amdahl delivers first Cmos mainframes, the Millennium 500 series
1998 - 3Q:IBM delivers first Cmos mainframes that exceed the power levels of the old ECL systems, the 9672-G5
1999 - 1Q:HDS launches second generation Ace mainframes, the Trinium range
1999 - 2Q: Amdahl brings out current Millennium 2000 C and E models
2000 - 4Q: IBM scheduled to launch Freeway, its first 64-bit mainframe
2000 - 4Q:Amdahl promises 25% performance improvement over current products
2001 - 2H:Hitachi to return to mainframe market with Hercules, a multiple operating system
2001 - 2H:Amdahl scheduled to launch its first 64-bit mainframe range
The name game
IBM's next generation of mainframes is currently known by the code name Freeway, or G7, and first release of the operating system that will run on it as OS/390 R10. But don't expect them to be called this when they actually see the light of day.
IBM is currently going through a mind-bending exercise deciding what to call them. The powers that be have decided that, as we had System 360 in the 1960s, System 370 in the 1970s and System 390 in the 1990s (they were obviously asleep in the 1980s), IBM needs a new name for a new millennium (but they can't call it Millennium, because that's what the Amdahl range is called). OS/390 will get a new name at the same time.
Web-serving on a mainframe
IBM has been arguing for some time that a System 390 mainframe is the best option for users providing e-commerce services based on data already held on mainframes.
Iowa-based Principal Financial Group has moved a major part of its Web services from a Unix environment to an IBM S/390 mainframe. "Performance has improved dramatically - our typical response time is only two seconds now," says Dale Ward, assistant director for information services. "With our core data already residing on the IBM S/390 and the back-end transactions already taking place there, making the same S/390 our Web server allows us to cut out several steps, eliminating potential points of failure along the way."