isham research
An analysis of the legal actions is available
Platform Solutions' (PSI) approach to the market has been discussed before with a degree of scepticism entirely justified in the light of recent rumours. PSI made strong statements about being an IBM plug-compatible manufacturer rather than an emulation supplier - the sort of statements that might imply that hardware would soon be leaving a California facility with the company's logo on it - just as the real PCMs did it.
But it would appear that PSI's base system is a standard Itanium server with - a novelty in the IBM plug-compatible market - a Hewlett-Packard badge on the front. So far, PSI has not announced any hardware content of its own (Fundamental Software markets channel adapters, etc., for FLEX-ES) so PSI's hardware is 100% Hewlett-Packard. Meaning, by implication, that an existing system of the right specification could be used. Mainframe emulation shipped on a DVD.
This may change the situation somewhat. Instead of seeing PSI as a standalone supplier that happens to use Hewlett-Packard hardware - just as Fundamental's partners used Dell and Compaq before switching to IBM xSeries - it may be better to view PSI as another tool for Hewlett-Packard (and perhaps Fujitsu) to use in attacking IBM's low-end mainframe customers. Where IBM itself, of course, has had no viable product for well over a decade.
A tunnel-vision obsession with high-revenue customers that still neglects the very real need for a low-end critical mass for the ISVs supporting the architecture.
And, it seems, Fujitsu has reviewed its commitment, having being an early investor in Platform Solutions.
PSI's model-oriented (declarative) approach continues - a PSI 5315-130 is supposedly equivalent to an IBM 2086-150 at 26 MSUs, or around 170 MIPS. Such a system is scheduled for performance evaluation in North America in the near future.
It's interesting that PSI's system is not described in MIPS terms, but in MSUs.
"MSUs are used for software pricing and are not necessarily a direct indication of relative processor capacity."
SC28-1187-09 Large Systems Performance Reference Page 78
IBM opened Pandora's Box itself when it introduced the concept of charging MSUs, reducing the stated MSUs for the z990 by 10% in October 2003. It has since reduced them again for subsequent systems by varying percentages; but the principle is now established - what is to stop PSI reducing its own MSUs by any figure it chooses? A PSI 5315-130 declared at 26 MSUs might be a 170 MIPS machine - but equally, since IBM uses a variety of MIPS/MSU ratios, PSI is at liberty to use any ratio it chooses and it might be a 300 MIPS machine with the software costs of a 170 MIPS system.
If it is indeed performance-related, the positioning is interesting. The 2086 range (better known as the z890) provides three 26 MSUs performance points:
If the PSI 5315-130 achieves 170 MIPS, it is very unlikely to do so with a single engine. Fundamental Software's much more mature FLEX-ES was benchmarked some years ago at 65 Intel MHz per 1 IBM MIPS and has since improved. So a 3.2GHz Intel processor could achieve something close to or even above 55 MIPS. With the poorer MP ratios emulation systems achieve, 170 MIPS is just about feasible on a 4-way Superdome. But although 55 MIPS per engine is enough to build systems in the z890 entry level, it isn't enough to provide the kind of power that the z890's other engine types - zAAPs, IFLs, etc., can bring to the party at around 366 MIPS each. True - a Superdome can run Linux applications on native Intel much more efficiently than they can run under Linux/390 under emulation - but the connectivity is nothing like so good and the zAAP offload advantage would be missing completely.
There are rumours that IBM will produce its own mainframe emulation based on POWER5 - possibly as early as 2Q06 - and that its current intransigence with other suppliers is simply a way of building latent demand for its own forthcoming product. IBM is no stranger to emulating the System/360 and all of its successors - from the beginning (System/360 Model 25, System 370/115, /125) it has implemented its mainframe architecture on a variety of highly disparate microprocessors. It's also often necessary to support new instructions for software development purposes long before the hardware is built - but this type of emulation never has production-level performance requirements. IBM possesses machine-readable architecture definitions that should make such a job a complete snap, but rumours suggest that its latest attempt suffers from both performance and integrity problems - fixes for which are often mutually exclusive. If the most recent versions of IBM's zArchitecture definitions are used, its emulation will not support current versions of VSE and a 64-bit version will become essential.
One inevitable consequence of IBM announcing its own zSeries emulation is that it will also have to provide a standard software licensing framework. Previous attempts to restrict certain license types, such as Growth Opportunity License Charge (GOLC) to its own hardware have uniformly failed (vide the Amdahl Omniprise) so the new structure is almost certain to be available to other emulation suppliers whether on Intel or other platforms. The withdrawal of ESL licensing at the end of 2005 may be part of the preparations for this.
If IBM also announces the z890's successor - using the z9 processor chip - at the same time, it will probably permit little or no performance overlap between the two. Otherwise, a pSeries system might take a zSeries sale - and that would never do. Other emulation suppliers lack these inhibitions. It is more likely that IBM's emulation will be positioned as a replacement for the Multiprise 2000 rather than the Multiprise 3000 - users of the latter are supposed to move up to "real" mainframes, no matter what the impact on their business. Interestingly, IBM has recently started referring to the z890 as a "medium" rather than a "low-end" system - preparing the ground?
It's inconceivable that IBM's emulation would be announced without 64-bit capability - it must, after all, have a range of currently supported operating systems and middleware. From this point onwards, it will be legally impossible for IBM to prevent other suppliers shipping 64-bit zArchitecture emulators.
And it's also entirely possible that such a double launch might be accompanied by a move back to direct sales for mainframe products - recent changes to the incentives paid to the channel for mainframe sales suggest this is in IBM's thoughts.
Other competitors - most notably Sun - have tried to gain entry to IBM's low-end mainframe base. But most required some degree of human intervention during a partially automated conversion process. It worked for CICS transactions but fell down dramatically on some types of applications, e.g., turn-around documents. And, of course, it required an irrevocable up-front decision to adopt Sun's UNIX architecture.
The plug-compatibles such as Amdahl and Hitachi never made the slightest attempt to convert their customers to any other architecture. This meant the loss of a customer system to a PCM was far from the end of the world for IBM - with no changes made to the software, it might easily win back next time. And it often did - and has done in very many cases since Amdahl and Hitachi Data Systems left the market.
And even PCM adopters continued to pay IBM huge sums for their software licenses and peripheral maintenance.
Any decisions to run IBM mainframe applications on a Hewlett-Packard Superdome under HP-UX would be similarly reversible. But only at first. The temptation to move them to native HP-UX may prove irresistible - once there, they will have more than an order of magnitude better price/performance and be lost forever to the IBM mainframe world.
With IBM's software license revenue stream (85% gross margin) and peripheral maintenance income gone forever, too.
All of which makes it hard to understand IBM's attitude to Fundamental Software. In the above context, FLEX-ES is a plug-compatible product. No software changes are made - Fundamental has no incentive to lock the customer in to a proprietary platform of its own. In fact, since all of Fundamental's partners use IBM xSeries hardware, IBM doesn't even lose access to the hardware. Should the user start to grow, or should IBM bring out a future system with greater flexibility and a lower product cost, the customer can easily be won back to "real" hardware. The xSeries previously running the emulation can then be recycled for almost any purpose. Perhaps even as a cost-effective platform for WebSphere. And, of course, the software revenue stream is never interrupted.
Curious that IBM has so negligently opened a door to permanent high-margin revenue loss.
Given that IBM has had no "new architecture" product down at this level for well over a decade, the next ones up are prohibitively expensive, and IBM didn't do too well with its last attempt to market mainframe emulation, this sector could turn into a Fundamental Software versus Hewlett-Packard and/or Fujitsu game, with IBM standing on the touchline. With its supposed commitment to treat all emulation suppliers equally, it can hardly aspire to be on the field. In such an event, the telling factors are likely to be FLEX-ES' features, installed base and track record versus the sheer market penetration and credibility of industry giants like Hewlett-Packard or Fujitsu. The cost savings compared with IBM's overpriced z890 systems and especially its software and peripherals are so enormous that price is unlikely to be an issue in any such face-off - the arguments will be feature/function - FLEX-ES' much greater functional richness such as FakeTape versus HP-UX's UNIX extensions.
Complete loss of control of its low-end mainframe environment looms for IBM, thanks almost entirely to its lawyers. And with the Intel performance curve blowing IBM's mainframe performance curve into the boondocks, what affects the low end today affects the top end by 2009. Even today (late 2005) 1800 MIPS is perfectly viable under emulation on Intel.
If, indeed, the PSI product turns out to have significant backing from Hewlett-Packard then the world may change considerably. Fundamental Software built its sales channel slowly, persuading one IBM Business Partner after another to market the product and almost imperceptibly growing to second in the market without the backing of an industry giant. Hewlett-Packard has salesmen in every town, very deep pockets and a lot of lawyers; Fujitsu still has a lot of the ex-Amdahl support skilld. Once the green light goes on after PSI's pending tests, things could move very quickly indeed.
Hewlett-Packard's involvement may have other implications. IBM has so far refused to permit the use of its intellectual property to support 64-bit zArchitecture in commercial emulation products. As well as this being short-sighted - creating spurious off-platform migration pressures - it is also only a matter of time until ISVs such as Computer Associates or CompuWare wishing to ship products exploiting 64-bit versions of IBM middleware (e.g. DB2 V8) come to regard it as a restraint of trade. The problem for IBM is that there are so many of them - a potentially very complex legal environment could be created overnight. But Hewlett-Packard has deep pockets and good lawyers - it stands a much better chance than any of the emulation suppliers so far of prising 64-bit zArchitecture out of IBM. And if IBM resists, we may revisit Consent Decree or EU Undertaking territory all over again. A dangerous game for IBM.
Yet more potential for loss of control. Strategies, it would seem, designed to generate as many adversarial players as possible.
In this context, it's interesting that IBM is completely reorganising its legal department. The new structure places an outsider (Bob Weber of Jones Day) as the direct report to Sam Palmisano (CEO) with all of IBM's legal structure reporting through him. Perhaps this move is related to suits against IBM (working conditions in semiconductor plants, pollution etc.) or but it may yet have agreeable fallout.
With an obscene sucking sound, software license verification raises its wart-encrusted, slimy, stinking, weed-festooned, one-eyed, lop-sided rictus leer from the ice-cold steaming swamp just beyond the outer fringes of reality, mumbling something about "charging MSUs" and "indicator of performance". If PSI's emulation subsystem does, in fact, wind up being shipped on DVDs to existing Hewlett-Packard Superdome users, it may be difficult for IBM to ensure - on an ongoing basis - that its Ts&Cs are being observed. As suggested above, there seems to be some pressure within IBM demanding equal treatment for all emulation suppliers - and de facto because of its installed base, that means regarding Fundamental Software as the precedent. IBM's path of least resistance - time being of the essence - is probably to purchase or license Fundamental's dongle technology and make it a prerequisite for all suppliers; in the case of purchase licensing it back to Fundamental.
One odd coincidence. Back when the Hercules Open Software effort (note the many endorsements from IBMers) was beginning to impact the zArchitecture world, one Hercules user applied for and received a license to run z/OS under Hercules. This was not repeated - and IBM even issued a statement to this effect. PSI, when it placed its "evaluation machine" at LL Bean at the turn of the year, also applied for and received z/OS licenses - but may not receive any more, unless Hewlett-Packard's lawyers prove better than IBM's. Not hard, actually, as those with recent experience might attest. Odd - that each new implementation receives only the first license applied for. IBM claims that these licenses were granted in error; but if it was actually testing the implementations - and the act was thus deliberate - it may have created precedents. A priori, the latter looks the case.
Incredibly, Jay Maynard of Hercules notoriety continues to suggest there may be some legal way to run z/OS under Hercules, despite IBM's license agreements (example: SMP/E quite explicitly banning copying to PCs without paying extra license charges.
One could speculate about IBM's next "mid-range" mainframe. Although not appearing in public since the z800, Hitachi is still IBM's second source for high quality MCM substrates. A z890 successor might be built by Hitachi once more, and would have to meet certain simplicity requirements; so a single z9-style "book" with eight chip locations, each housing a single-stack chip that has not met the stringent z9 timing standards is a likely option. The z890 used a very innovative and attractive approach to delivering performance over a wide range - hopefully, IBM will repeat this with the new machine. One processor must be reserved as a SAP, and it would be prudent to reserve one as a apare - so six processors would be configurable. IBM has typically degraded top of line processors by 70% (z900 to z800) or 80% (z990 to z890) so a middle figure of 75% is not unreasonable. Another adjustment of software MSUs to 8 MIPS/MSU seems equally likely.