isham research
Somewhat overtaken by events - as of late 2005, IBM is indeed planning to proceed with the mainframe.
What Hitachi said on 6 March 2000:
"Consequently, we have made a decision to accelerate our product development in heterogeneous enterprise servers. These servers will employ both new technologies and production techniques. In the meantime, Hitachi Data Systems will continue to deliver servers in support of existing server customer requirements. We will also deploy additional resources to enterprise storage and business solutions, details of which will be finalized in the next 90 days."
What Amdahl said (internally) on 15 October 2000:
"Based on marketplace requirements, Amdahl has made an investment decision to focus on its current System/390 business rather than develop and deploy a 64-bit architecture S/390 server."
Although service commitments and feature deliveries continue in both cases, eventually both companies - the last of more than most people realise - will depart from the System/390 plug compatible processor market.
(And now they've duly gone.
Why? The reasons are partly complex and partly simple - but at the end of the day, the game is no longer worth the candle. Skyline/Trinium was an expensive technology for Hitachi to produce - although Fujitsu/Amdahl stuck out a little longer with their less expensive CMOS technology, projections showed hardware selling prices falling below production costs within the lifetime of any such 64-bit machine. Some insiders blame analysts (especially one major US-based one) for continually bombarding customers with expectations of lower hardware prices, but the real truth lies within IBM.
When System/360 was announced (17 April 1964) it was the world's first 'universal' computer. In those days, this meant it handled both scientific and commercial workloads; or could, if you rented all the features - you couldn't buy them. In practice, this meant synergy - the CADAM system could use floating point arithmetic to design widgets, and interface with the decimal-based payroll system to work out how much they would cost to build. This "one pot for everything" idea remained current until July 1988, when IBM lost the plot completely and introduced Graduated Monthly License Charges - probably the biggest single mistake the company has ever made.
Graduated charges (in all the various forms) were fundamentally flawed - they priced software not according to the use a customer made of it, but according to the size of the machine it just happened to be running on. The evil refinement was a degressive charge rate - software became proportionately cheaper on larger machines. This placed the user in a dilemma - consolidate system images to save money on common software such as the operating system, or retain small images so that small applications could run on them and pay appropriate charges.
In the event, most users chose to consolidate and shed smaller applications to other platforms. Tens if not hundreds of thousands of applications were moved from IBM mainframes to vastly inferior 'open systems' or - even worse - Windows NT on the basis of entirely artificial business cases justified overwhelmingly by saving graduated charges that should never have been imposed.
Each application lost represented only a few MIPS. But the loss of potentially synergistic applications and their cumulative requirements for reliability and availability has critically weakened the platform.
The fact is that there's a cuckoo in the nest. Although IBM's Server Division doesn't regard the mainframe as a cash cow, the Software Division does - and it's squeezing it. Those who are tied to the zSeries mainframe are regarded as easy targets to maintain Software Division's profits. It is clear that the Server Division lost the battle to reform software charging on zSeries - the Workload Level Charging scheme is too little, too late - and may prove insufficient to generate significant user migration to the 64 bit z/Architecture.