isham research
(This discussion has been overtaken by events. The current version will be found here)
The z800 Model 0E1 completes a lineup started with the base z900 in October 2000 and continued with the low-end z800 in February 2002, and the z900 turbo announced on 30 April 2002.
So as predicted last year, the entry point is now the 2066-0A1 at 80 MIPS - the less powerful 2066-0E1 being priced at much the same level because of its standard Integrated Facility for Linux (IFL). But the 2066-0E1 and 0A1 are too powerful for thousands of IBM's mainframe users - even a secondhand Multiprise 3000 H30 is too large for most. In the 1970s and 1980s when licence charges were 'per system' and not 'per MIPS', this would not have been a problem - small users with modest growth were a good market for cascaded secondhand hardware. Since the adoption of graduated charges based on machine capacity, this is not the case - if a small company receives a larger system as a gift, its software costs can bankrupt them. The concept of 'white space' has been introduced to eliminate software charges for unused capacity in the MVS world but not for VSE.
A further issue is the huge number of older IBM mainframe models being taken out of support in 2003 - 65 systems, with many more also being unable to run the latest Release 2.7 of VSE.
New System/390 and zSeries systems of less than 80 MIPS are, however available - via emulation of the architectures on other platforms such as IBM's xSeries. At present, the only IBM-approved zArchitecture product is Flex-ES from Fundamental Software and its partners. This was briefly marketed by IBM itself as 'x/EFS' on the IBM NUMA-Q x430 systems, but the underlying ex-Sequent hardware platform was too expensive.
Flex-ES is now available in a 64-bit zArchitecture-compliant (ESAME - ARCHLVL=2) form. Independence of hardware Architecture Level Sets is a further advantage for emulation at the low end - and the ability to switch between, e.g., ESA mode and zArchitecture mode via a parameter opens new possibilities for regression testing across Architecure Level Sets.
It has been suggested that users could move legacy workloads to a low-end z800 at their current charge levels - a low-end form of IBM's 'white space' concept - but in the current environment ISVs are unlikely to co-operate. One business partner programme may have the reverse effect from that intended; a user with a smaller system can move to a z800 0E1 or 0A1 and temporarily retain the same software charge level - but with a Damocles' sword of full software charges after 48 months. The proportion able to grow into and justify the software charges for such a machine is likely to be small, and the reversion to normal charges after four years may prove a great incentive to leave the platform or downsize to an emulation solution before then.
Another idea would be to use a suitably partitioned Intel or AMD server such as IBM's x440 to provide a System/390 environment on one or two Intel processors, with native Intel Linux or other environments on other processors. Hitachi has open-sourced a Linux file system with potential for accessing legacy System/390 data stored on external DASD direct from Linux.
Flex-ES's greatest initial success was in the Partnerworld for Development market - via Cornerstone and T3 Technologies - this was largely because of software licensing issues. Emulation is now increasingly being adopted for production in many sites - especially good results being achieved with I/O bound workloads. Performance-limiting measures to qualify FLEX-ES systems for ESL licenses (up to 8 MIPS) are well proven - these could be expanded to provide performance points tailored to each of the lower Processor Groups, thus preserving users' current charge levels.
Such a system - an open-ended Intel-based server running a System/390 workload at a specified Processor Group performance level, combined with Linux (or other) workload(s) on the other engine(s), would reduce the pressure for early migration, allowing it to proceed in a more controlled manner or be deferred. Interim growth could be catered for by lifting the system to the next Processor Group, with an appropriate adjustments in software charges. It might prove easier to justify investment in such a system than in one dedicated to the System/390 environment. If the migration stalls, the Linux portion of the system could be upgraded with no adverse effects on software charges for the legacy workload.
IBM might benefit greatly. It currently has no viable new hardware below the z800-0A1, and thus nothing to offer 95% of its old installed base. However, a user transferred to a machine with the same performance and modern software support would be able to exploit many IBM products - mostly middleware - that are otherwise inaccessible. DB2 V8 requires z/OS 1.3, for example, and Domino V6 requires z/OS 1.2. Many modern solutions exploit such middleware, and it might be that signficant opportunities could be opened up in this way - not just middleware sales but also eventually real zSeries sales.
Emulation opens other possibilities. Not only does it lower the entry cost for ISVs wishing to develop for or port applications to Linux/390 and/or z/OS, it also allows a mainframe environment to be run on a laptop. Demonstrating System/390 applications has always been difficult; it is hard to do it effectively over a dial-up link - either the entire system has to be installed at the prospect's site or all of the customer's decision makers have to visit a demonstration data centre. The latter can be a little like herding cats. Only rarely can applications be demonstrated satisfactorily via dial-up links. Now, a humble IBM ThinkPad can run z/OS or Linux/390 under Linux - and fit in a salesman's briefcase.
Comments on the above ideas are welcome at philp@isham-research.co.uk - all input will be treated in the strictest confidence - no contact of any kind will result unless requested.
Older articles are in the archive