By Jeff Gould, CEO & Director of Research, Peerstone Research

The following is a hypothetical case study of the market for enterprise-class computer systems. To the best of my knowledge the events described herein have not yet occurred. However, it is possible that they may do so in the near future. If they do come to pass, they are likely to be the subject one day of exhaustive case studies in America’s leading business schools. Like all good case studies, this one is open ended, because its outcome will depend on the choices and actions of the players involved.

Suppose you are the CEO of a computer manufacturer who has decided to present an exciting new business idea to your board of directors. You propose nothing less than to turn your company around by entering the lucrative multi-billion dollar market for high-end machines that run IBM’s 64 bit operating system z/OS. Some of your advisors may object that mainframes are a dying breed. But you will remind them that, according to no less a source than IBM itself, these mainframes are still home to an eye-popping $5 trillion worth of applications and 80% of the world’s corporate data. And yes, you will insist, that was “trillion” with a “t”.

It is the value of these applications that determines the potential value of the mainframe platform underlying them, you explain. That platform in turn consists of two parts, the mainframe operating systems, such as IBM’s flagship z/OS, and the mainframe hardware that runs the operating systems. The applications are wedded to the operating systems, and the operating systems are – at least for now – wedded to IBM’s hardware. If it were possible to move these applications for a reasonable cost to cheaper non-mainframe platforms based on commodity Intel/AMD microprocessors, then the value of traditional mainframes would plummet. It is because IBM has succeed in preserving important technical and legal barriers to such migration that its mainframe hardware continues to earn far higher profit margins than all other servers on the market, including even the most sophisticated high-end Unix machines. Those fat margins, you observe, are due in large part to the fact that IBM now has a share of the mainframe market estimated at 99.8% by one of its few remaining would-be competitors, T3 Technologies. This near total market domination is due in turn to the fact that the traditional IBM plug-compatible manufacturers such as Amdahl have thrown in the towel and would-be new entrants such as PSI have disappeared into a black hole.

But despite the barriers to entry, you tell your board, such a large and lucrative market remains a tempting opportunity for computer companies like your own. Your company desperately needs new sources of growth. Wall Street demands nothing less. Even after competition between your company and IBM has driven mainframe prices down to more realistic levels, the market will still be huge. If your company could only get its fair share of this pie, its stock price would soar.

Assume for the sake of argument, you continue, that IBM would agree to license its flagship z/OS to your customers at its usual extortionate rates if your company was capable of producing a machine that could run it. This assumption, you concede, might be counterfactual given that the Department of Justice has terminated the historic Consent Decree that for decades compelled IBM to play fair with competitors. But you ask your listeners to imagine just for a moment that some future antitrust authority in America or Europe will revisit the issue. After all, the DoJ and the Europeans have both said they expect Big Blue to play nice, and since it obviously isn’t, one can reasonably assume that sooner or later they will notice.

The question nevertheless remains: how can you build a computer equivalent to an IBM mainframe? This is no minor challenge, you candidly admit. The z10 microprocessor at the heart of IBM’s latest System z mainframe is a quad-core 4.4 GHz chip with no less than 991 million transistors. It is manufactured in IBM’s own factories using the company’s 65 nanometer fabrication process. Duplicating this technology would certainly require billions of dollars in upfront investment. Indeed, the extravagant cost of matching IBM’s 64 bit z/Architecture hardware was the reason cited by Hitachi and Fujitsu-Amdahl when they shocked mainframe customers everywhere by abruptly pulling out of the plug-compatible business back in 2000. The unavoidable conclusion appears to be that the challenge of designing and producing a billion transistor microprocessor is just too daunting for a company like yours to take on.

But things are not as hopeless as they seem, you tell the board. There is a way to duplicate an IBM System z mainframe without building a new microprocessor. An open source mainframe emulation program called Hercules, consisting of a mere 300,000 lines of C code and capable of running on standard Intel or AMD microprocessors, can be downloaded for free. Apparently it is shockingly easy for knowledgeable programmers to emulate the z/Architecture 64 bit instruction set, since Hercules was put together quite quickly by just a handful of coders working in their spare time. (Hmm, you privately wonder, how come all the rocket science engineers at Amdahl and Hitachi couldn’t figure this out, since they had been emulating IBM instruction sets in microcode for years? Did IBM offer them some incentive to look the other way? Let’s see now, shortly after Hitachi stopped developing plug compatible mainframes it scored a huge deal to acquire IBM’s disk drive manufacturing business. Oh, and Hitachi was actually commissioned by IBM to help develop some of the early z Series mainframes. (No, don’t go there…)

At this point your board is beginning to get excited and you carry on with your exposition. Benchmarkers report that a barebones $1,500 dual processor Opteron box loaded with the Hercules emulator can do useful mainframe production work at 30 MIPS or better. You conduct a little thought experiment for the board which demonstrates that, with a little effort, your company’s renowned R&D department could quite possibly build a far more powerful multiprocessor machine using the most recent quad-core x86-64 chips. You might even look into using Intel’s amazing new Dunnington chip with six cores and 1.9 billion transistors. That should be more than a match for the z10. It turns out that the Hercules emulator dispatches a new thread for each z/Architecture CPU it emulates, and does the same for each i/o process. This means it is highly scalable. With a suitably tweaked symmetric multiprocessing architecture, such a machine, although built from commodity components, might well be capable of generating hundreds or even thousands of mainframe-equivalent MIPS. Your company could sell it for a fraction of the cost of an actual System z mainframe and still make a generous profit.

There is only one obstacle that could prevent this scenario from coming to pass, you concede, and that is the legal barrier IBM has managed to erect around z/OS, which remains the indispensable platform for the world’s trillions of dollars worth of legacy mainframe applications. Without the Consent Decree to compel IBM to license z/OS to users who purchase non-IBM server hardware from companies such as yours, you are at an impasse. IBM seems to enjoy gobbling down its rich monopoly pie all by itself and has little incentive to share, other than the idealistic notion of preserving its good name. Mainframe customers are willing to pay the extra freight because they see no alternative. With luck the Department of Justice or the European Union might be persuaded to arouse from their regulatory slumbers and look into the matter. But that will take years. In the meantime your company is being squeezed between commodity servers on the low end and IBM’s protected Cobol behemoths on the high end.

Perhaps there really is no hope, you acknowledge to your board. By the time the regulators wrap up their case, your company may be driven from the server market and perhaps into the arms of a competitor, possibly even IBM itself. It seems that only IBM is allowed to make a killing in the mainframe market. With a sigh, you remember what your parents always told you: life isn’t fair.

Class discussion: put yourself in the position of this CEO and devise a strategy that will persuade or compel IBM to resume licensing z/OS to users of all brands of server hardware on competitive market terms.