isham research
During its battle with the plug compatible manufacturers IBM aggressively moved its mainframe revenue stream from hardware to software. This had the desired medium-term effect - both Amdahl/Fujitsu and Hitachi decided that there would be insufficient profit on zArchitecture hardware.
But it leaves behind an anomaly - IBM's mainframe revenue stream is now perhaps irrevocably skewed from hardware to software. A curious situation; mainframe software is IBM's most profitable activity (with gross margins sometimes around 90%) yet it has no obvious means of expanding its customer base.
Mainframe users don't grow on trees. More prosaicly, they don't spring fully formed out of the ground and buy 10,000 MIPS systems. They have to be captured and nurtured. And it isn't happening.
Why not? Because although sales of medium and large systems are relatively stable and perhaps even growing, IBM has no entry level mainframe product. Worse yet, it is even turning its back on 4,000-plus established small mainframe users, some of which might (given equitable software pricing) have been induced to expand their use of the platform.
IBM used to pay a hefty bonus to a salesman who won a new mainframe customer. Perhaps today's executives should be fined the same amount for each one they lose. Although it doesn't happen that often, small users do sometimes grow into big ones.
But no more. They're being shown the door - Hewlett-Packard et al are simply drooling. Because it hasn't introduced a new small system since the Architectural Level Sets started, IBM is in a curious position. Thousands of users have systems for which there is already no supported operating system - yet IBM is still charging license fees for the unsupported ones and also charging for hardware maintenance.
Within the next few months - possibly in July 2005 - IBM is expected to announce the end of hardware maintenance availability for most of the remaining small systems at the end of 2006. These systems include the G3 and G4 mainframes - neither of which have the G5-equivalent level set critical for most software - and perhaps the Multiprise 3000, which uses the G5 chip set. Rumours of the Multiprise 3000's demise may be the factor behind a resurgence of interest in second-hand z800s in recent weeks.
Oddly, the Multiprise 2000 may survive a year longer despite having little supported software available.
With so many sites currently under threat, even 5% of them starting to grow would provide more "new" sites than the whole of IBM's sales force has recently managed.
There are no signs of any superseding systems coming from IBM. Rather the reverse; IBM's next midlife kicker for the z990 - also expected in July - is rumoured to have a uniprocessor power close to 600 MIPS; not a good start for addressing the sub-40 MIPS market. By 2008, uniprocessor performance might well be over 1,000 MIPS. Not only that, but low-end systems depend on inexpensive peripherals - integrated storage, communications, etc. - the classic mainframe peripheral park is far too expensive.
IBM attempted to throw its most loyal customers (fifteen years without software price/performance improvements and seven years without a new product completely redefines the concept of loyalty) a bone with the x/430 but the product was simply over-engineered (and thus too expensive) for the job.
It's interesting to remember what IBM said about System/390 emulation on Intel at the time:
The NUMA-Q EFS [Fundamental Software's FLEX-ES] solution is intended to satisfy users who:IBM's FLEX-ES Announcement, Letter 100-362
- Need a technology upgrade or incremental performance increase that requires less than the capacity delivered by the Multiprise 3000 Model H30.
- Are replacing older systems:
- S/370 (TM), 43xx, 30xx, 9221, small 2003/9121, older P/390 and R/390, integrated servers, and similar machines
- Competitive systems such as:
- Amdahl ECL processors and Millennium 400/500 models
- Hitachi ECL, GX, and EX Processors
- Are offloading selected small projects and/or applications.
- Are involved in downsizing, rightsizing, mergers, or acquisitions. These users can just load images of their current environment and run.
- Are looking for an interim re-hosting platform.
- Need a multiple-use, multiple architecture disaster recovery platform.
- Are growing their UNIX or Windows 2000 workloads faster than their S/390 workload.
That, of course, dates from 2000 and the performance positioning below the Multiprise 3000 H30 reflects the NUMA-Q's use of an already obsolete 700 MHz Intel processor. Over the last five years a lot has happened in the Intel processor space and emulation can easily achieve 50 MIPS per processor - making a 120 MIPS system very easy and cheap to construct. Such a system would address the needs of all but the largest Multiprise users very well indeed.
But somewhere there seems to be a problem. A joke doing the rounds in IBM runs:
"How do you startle a zSeries executive?"
"Let off a firecracker behind him. But if you really want to see him clear a building, just whisper 'Platform Solutions' in his ear."
It's hard to see IBM's problem. The software revenue from 4,000-odd customers must be considerable - why let it go?