isham research

Mainframe Emulation - the ball starts to roll and the bell starts to toll.

2004 was a milestone year for users of small IBM mainframes. Service for over 60 models was dropped in 2003 with no replacements available. IBM introduced the z890, a new medium mainframe, but its triple-kneecapped 26 MIPS is too much for many users, even if they could afford the ISV software and non-integrated peripherals. This leaves a big gap in IBM's offerings that no MCM-based system can fill - and it is likely to be 2008 before more flexible packaging is available. Sole reliance on the expensive MCM looks to have been a major strategic mistake.

"Below 60 MIPS the only solution is emulation."

Dan Colby, then head of zSeries, Plenary Session, Partnerworld 2001

And in mid-2005:

• As predicted exclusively by isham research, UMX Technologies has gone - apparently leaving only a fraction of its code (with the comments in Russian) in escrow and very little documentation. A support nightmare for its handful of customers. The product may return to the market, courtesy of Roundtrip Systems LLC, who have bought some of the assets.
• It is still impossible to license IBM software under the Hercules system and there is no prospect of this changing.
• Extravagant promises have been made by Platform Solutions, including General Availability during 2005 - but some of their claims are less than plausible.
• Scattered rumours suggest that IBM is preparing its own pSeries 64-bit architecture emulator for release in 2H06. Given that the primary reason for zSeries customer satisfaction (NSI) being at an all-time low (down 8% in EMEA recently) is a lack of support skills within IBM, it seems odd that it would enter a market requiring a significant additional support commitment at a time when it has major problems in precisely that area.

So - because of its own success and the unrelated failings of others - Fundamental Software currently finds itself the sole global supplier of viable low-end System/390 and zSeries systems, via its resellers Cornerstone Systems and T3 Technologies. For this reason the rest of this discussion perforce concerns itself almost exclusively with Fundamental Software's FLEX-ES and IBM's hardware offerings.

With FLEX-ES Release 7 now being shipped by its business partners, Fundamental Software has raised the bar. Some features had the longest gestation periods in mainframe history but are nonetheless welcome. The most notable is completely revised physical channel connectivity, allowing FLEX-ES to attach ESCON peripherals, share other systems' DASD and allow them access to its own internal emulated DASD and many other peripherals. FLEX-ES Release 7 can thus be a player in a multi-system datacentre. All of this functionality is available to existing users via a simple software upgrade and purchase of the appropriate channel cards.

This opens many new possibilities:

• Replacing an obsolete unsupported mainframe in a shared DASD environment.
• Adding some extra MIPS without increasing software bills on an existing system.
• Adding extra (cheap) DASD capacity by sharing the emulated DASD of a FLEX-ES system.
• Migration to the FLEX-ES platform with a minimum of disruption.

Dan Colby's "60 MIPS" [ob cit] is now perhaps 150 MIPS. Simplistic sums done on Intel clock rates would suggest it should have become more over almost four years, but physical channel connectivity to FLEX-ES is still limited to a degree by available PCI slots and IBM's z890 starts to offer configuration flexibility above 90 MIPS. Perhaps 60 MIPS was a bit optimistic back in 2001.

Latent demand exists. The limited physical channel support (with its onerous storage configuration restrictions) in FLEX-ES prior to Release 7 was a major inhibitor in data centres and many otherwise eager potential users were dissuaded.

FLEX-ES already has impressively rich functionality in other areas:

• FakeTape emulates tape operations on cheap UNIX/Linux disks, from where FakeTape files can be remotely mirrored very cheaply - a number of users have completely eliminated physical tape mounts. Traffic on the FLEX-ES mailing list suggests pervasive exploitation of the feature. The scripts to do this are being shared around the FLEX-ES community just as code fragments were shared among early System/360 users - though instant FTP access is a little more responsive than the old six-monthly GUIDE GOODIES tape and developments are accordingly more rapid.
• Architectural co-existence is close to perfect. On one and the same Intel processor, one FLEX-ES instance can be running in the ancient ECPS:VSE mode while another runs in 64-bit architecture mode - ideal for "rescuing" applications from long-obsolete environments. No hardware product will ever be able to do this - nor will IBM's rumoured future emulation product.
• Industrial strengh networking capabilities that exploit both the IBM operating system's capabilities and/or those of the host UNIX/Linux platform.
• Complete remote operation and diagnostic capability. This is useful where an organisation wants to consolidate its IT operations but is forced to leave a physical system in one or more remote locations because of, e.g, local unit record or cartridge tape requirements. Move the staff, leave the system under a desk somewhere and use FakeTape plus remote mirroring for backup.
• Although UnixWare is still the preferred platform for commercial FLEX-ES systems, Linux is increasingly employed in special situations (e.g., Partnerworld for Development systems and certain special commercial environments) and will eventually take over. Most FLEX-ES systems are 2-ways - one processor for the emulated environment and one for housekeeping and general functions. Make a Linux 2-way into a 4-way - still with only one processor enabled for emulation - and you have an awful lot of spare Linux power in the system at a fraction of the cost of an IBM IFL. And outside the IBM software charging environment, zAAPs are irrelevant anyway.
• The fact that major changes (such as zArchitecture and the new channel support) can be implemented in FLEX-ES systems purely by reloading software alters the approach mainframe users can employ to get new investment. Changing to, e.g., z/VM 5 can produce real benefits for both the user (increased functionality) and for IBM in terms of assurance of its revenue stream and ease of support. But the change is all in software - little or no hardware needs to be acquired; and in organisations with long-term moratoria on mainframe hardware investment, this could be the only way to preserve the software revenue stream.

Like the PCMs, Fundamental Software has its own "value add" in the functional richness detailed above. NAS/HDS offered inboard vector processing and faster parallel channels long before IBM. Amdahl had its Multiple Domain Feature and Extended Channels. To win back these users, IBM had to develop its own equivalents. Fundamental has now opened up quite a lead and its users are starting to drive the process even harder, making suggestions and even providing sample code. It is extremely unlikely that IBM can come up with an alternative of similar functionality, even were it not frozen into sitting on its hands by its own internal politics:

• Hardware solutions don't have this capability. Even if IBM were to give away a future hardware system free of charge, providing the same functionality on such a hardware base would cost much more than licensing a FLEX-ES system and using off-the-shelf PCI cards.
• It is hard enough to justify kneecapping a 450 MIPS processor down to 26 MIPS, but the common architecture also requires at least one SAP per complex and IBM is unlikely to want to market a system without at least one IFL and zAAP as well. Kneecapping a 600 MIPS processor would be even more difficult to justify. One potential solution would be vertical partitioning of one processor to provide the characterisable engines.

Fundamental Software claims not to be (or want to be) a plug-compatible supplier, although the new channel functionality in FLEX-ES Release 7 takes it a step along that path. Since IBM has no viable small system, users still using the small mainframes obsoleted by IBM last year (or even earlier) would otherwise be lost forever from the mainframe environment. In a real sense, FLEX-ES is a lifeboat for their IBM and ISV software investment.

The penny has still not dropped at IBM. It still refuses to countenance ESL license terms for z/OS, despite the fact that it would create another 180 or so z/OS environments for Software Group to sell middleware into within a year. It also refuses to permit 64-bit operation under FLEX-ES in commercial environments - in spite of the inevitable loss of face when this is eventually forced by the ISVs.

In fact, a system "lost" to FLEX-ES is not "lost" at all, as it would be to Sun or HP. All of Fundamental's partners use IBM xSeries hardware and license IBM and ISV code to run on it. IBM's relationship with its customer barely changes - except for its local engineering manager losing a few headaches.

FLEX-ES is not the best way to preserve the lowest end of the mainframe market from now onwards - it is now the only way.

Home Page