By Jeff Gould, Peerstone Research / Interop News

How many IT departments still use servers designed when Lyndon Johnson was President and gasoline cost 30 cents a gallon? More than you might think. Depending on who you ask, there are somewhere between 7,000 and 10,000 mainframes still in operation around the world, about half in the financial sector, and the rest in a scattering of government agencies, utilities, universities and other big users. (By contrast, IDC says there are about 30 million Intel and AMD based servers installed in the world.) The vast majority of these big metal boxes have those three magic blue letters stamped on the front panel, which means they use an architecture that first entered service with the famous IBM 360 back in 1964. Of course, the 360s and the 370s and the s/390s are long gone. But their successors are still with us. Just last month IBM launched its latest refresh of this venerable line, the System z10, and many long-standing customers – though few if any new ones – are still willing to write big checks to get their hands on this gear.

The question is why? Why aren't these cantankerous old beasts as dead as diplodocus and triceratops? Aren't they outrageously expensive? Isn't it much cheaper to run even the biggest applications on Intel or AMD servers with Unix, Linux or Windows Server?

Answering these questions turns out to be not so simple. While there is no doubt that mainframes are outrageously expensive – a modest mid-range machine can easily run $1 million for the hardware alone, plus many hundreds of thousands of dollars more per year just for the operating system and basic middleware – there is controversy about the other half of the proposition, to wit the claim that the newer platforms really are cheaper. In one corner, you have a snooty IBM turning up its nose at the modern gear and sniffing that "when you include all the hidden costs such as labor, reliability, security, manageability, power and so forth, those scale-out commodity platforms are actually far more expensive than our elegant centralized machines", or words to that effect. In the other corner you have virtually every other major IT vendor on the planet – Intel, AMD, HP, Sun, Microsoft, Red Hat, Novell, Oracle, Dell, etc. – sputtering with rage and shouting "No way dude, our stuff is way cheaper!" (The execs in charge of IBM's own Intel/AMD-based server lines remain diplomatically silent during this debate.)

I can't wade through all the arguments and counterarguments here. Microsoft, HP, Intel and other interested parties have produced a boatload of case studies that purport to show mainframes cost more. It does seem that common sense is on the side of the "commodity" platforms. But defining and measuring the true costs over time of a complex set of enterprise applications running on varying hardware and software platforms in different organizational contexts is a very hard problem. There are just too many variables to yield any simple general rule.

One thing for certain is that the remaining mainframe customer base is not yet rushing to abandon its legacy machines. According to Gartner, IBM's mainframe hardware revenue plunged precipitously during the 1990s, dropping from $12 billion per year at the start of the decade to a measly $4 billion in 1999. However, revenue has stayed flat at this level since 2000, which happens to be the year IBM launched its current zSeries hardware. In fact, despite the big decline, the true size of IBM's remaining mainframe business today is far larger than the current hardware sales alone suggest, because the pricey machines are also responsible for bringing in a sizeable chunk of the firm's software and services revenues. A quick back-of-the-envelope calculation based on IBM's breakout of its Q4 2007 numbers suggests that 40% of its software business and perhaps a third of its services revenues are ultimately driven by zSeries hardware. That would put IBM's overall take from big iron at roughly $30 billion per year. Definitely something worth fighting for – and maybe even fighting a little dirty for – if you are sitting in Big Blue's boardroom.

But before we get to the fighting part, let's look at things from the mainframe customer's point of view. IBM's competitors usually couch their arguments for migration in terms of the readiness of their platforms to run mainframe-class applications and the lower total costs of ownership to be won by doing so. But I think there is one very simple factor these arguments don't address, which explains why so many customers are still reluctant to migrate despite all the obvious benefits of the newer platforms. That is the huge pile of legacy code still running on these big old boxes.

I have lately been perusing some fascinating legal briefs filed by feisty Silicon Valley startup Platform Solutions (commonly known as PSI) in the court battle it is waging with IBM for the right to sell zSeries-compatible "open mainframes" based on ingenious firmware embedded in Intel Itanium chips. According to these documents, legacy Cobol applications running on IBM mainframes are worth $1 trillion. But IBM itself gives a much higher number for the total value of all applications running on its machines, estimating their worth at an incredible $5 trillion. In other words, if it were a country, the mainframe legacy code base would rank as the second largest national economy in the world, surpassing the GDP of Japan! And these are not just any applications – they run the core business operations of the world's largest banks and insurance companies, airlines and government agencies. They started out life decades ago and contain by some estimates as many as 200 billion lines of code. The business value of this legacy is so enormous it functions like the gravitational field of a giant planet that grips and pulls in everything around it. To escape from the planet it's not enough to move fast, you have to move really fast – in the case of our own planet Earth, escape velocity is seven miles per second. Similarly, to justify the cost and risk of leaving the mainframe, you need to show not just benefits, but really big benefits.

To be sure, the lose-the-mainframe vendors are not suggesting that users simply rewrite or replace their existing applications. Instead, they typically propose a mix of new development (e.g. Java or .Net), packaged applications (e.g. SAP) and the re-hosting of legacy Cobol code on new platforms using suitable middleware. But that is easier said than done. From the point of view of the people sitting in the corner office, the most attractive alternative to undertaking a big strategic change whose outcome is uncertain is to do nothing at all. If you own a hugely valuable asset and someone comes to you with an ambitious but rather complex plan to improve that asset, you are going to look long and hard before you leap. Indeed, many people won't leap at all unless the house is on fire.

The sales pitch of the migration proponents says "life will be better on the other side of the hill." But the reality is that many mainframe customers won't want to move until life becomes intolerable on this side of the hill. What could make that happen? Two things, I believe: (1) the archaic and inflexible nature of legacy mainframe applications architecture; and (2) overreaching by IBM as it tries to make up for low growth and declining margins by squeezing as much extra profit as it can out of its captive mainframe customers.

Archaic and inflexible: Gartner has a good deck of PowerPoint slides reviewing what it calls Legacy Modernization. According to the analyst firm, the IBM mainframe applications architecture:

• Is monolithic
• Uses [outdated] procedural programming styles
• Sacrifices flexibility for performance
• Fails to separate data, business logic and presentation

In the long run, these are the things that will kill the mainframe. Remember, these attributes are intrinsic to the applications themselves and can't be remedied by IBM's endless stream of ingenious tweaks and innovations in zSeries hardware or system-level software such as z/OS. Until now IBM has bumbled along by persuading customers to surround legacy apps and data with expensive Java middleware implemented by even more expensive professional services. But inevitably the time will come – probably sooner rather than later – when this strategy chokes on its own complexity and cost.

The big squeeze: No company in the IT industry has more experience extracting monopoly rents from its customers than IBM. This long-standing historical fact about Big Blue has somehow been forgotten in the post-Y2K era, when IBM has engineered an amazingly successful make-over of its traditional "Big Blue Bully" image by touting its dedication to openness. While Linux and open source software may be insignificant parts of IBM's actual revenue stream, they are hugely important to the new "Gentle Giant" face it presents to the world. The reality of the company's business practices is somewhat different, as can be seen in its ruthless attempt to squash would-be mainframe competitor PSI with a court case alleging patent and copyright law violations.

This is the part where I have to remind everybody that I am a hard core believer in free markets, and I therefore believe that IBM actually has a perfect right to refuse to license its proprietary technology to competitors who want nothing more than to undercut its prices and entice away its customers. IBM has spent billions developing z/Architecture and z/OS, so it is only natural that it doesn't want PSI selling Itanium boxes that do the same job for a fraction of the cost. By the same token, I think Microsoft ought to have the right to keep the inner workings of Windows secret, Red Hat ought not to be compelled to unbundle RHEL software updates from pricey support services, Oracle should not be forced to port its E-Business ERP suite to non-Oracle databases, and so on.

Having said that, I am amazed by IBM's effrontery in the PSI case and by the naïveté of the members of the open source world who take the giant's promises of openness at face value. Big Blue would have us believe that its attempt to preserve a de facto 100% share of its market is merely a defensive measure on the part of a poor beaten-up underdog. A $5 trillion underdog. At least Microsoft is good enough to let other people manufacture the hardware that Windows runs on. IBM owns everything – the hardware, the operating system, and the middleware. In the case of PSI, IBM is trying to shut out a tiny competitor who is merely asking IBM to sell its mainframe OS (at the usual exorbitant price) to customers who buy PSI's alternative mainframe hardware.

Consider this interesting illustration of the duplicity to which IBM is apparently willing to stoop to preserve its mainframe monopoly. Until February 2006 the company published the following very noble declaration on its web site:

"IBM has an open approach to patent licensing for products in the Information Technology (IT) field and is generally willing to grant nonexclusive licenses under reasonable and nondiscriminatory terms and conditions to those who in turn respect IBM's intellectual property (IP) rights."

You won't find these words on IBM's site anymore, but thanks to the miracle of the WayBack Machine's web archive you can still see the original page here. By November 2006, a few months after this passage was deleted from its site, IBM was saying something quite different in the Amended Complaint it filed in Federal Court against PSI:

"IBM has refused to agree to license its patents to PSI and its copyrighted operating systems and other software for use on PSI's emulator systems despite PSI's demands because, among other reasons, IBM has no obligation to do so..."

Well, that's clear enough! In the same document IBM flatly states that it "will not license z/OS on PSI platforms." The court filings recount a long series of letters and meetings from 2000 to 2006 where PSI asks and then practically begs IBM to let it license z/OS. Reading through this story you feel the mounting desperation of the PSI execs as they gradually come to realize that, contrary to its 25 year practice with the former plug compatible manufacturers Amdahl and Hitachi (now withdrawn from the mainframe market), IBM has no intention of granting PSI a license. This episode reminds me of the movie Independence Day, where the President of the United States asks a hostile alien, "What do want us to do?" The alien's reply: "We want you to die."

By sticking it to PSI, IBM is also sticking it to its mainframe customers, the majority of whom would doubtless remain faithful to Big Blue even while benefiting from the more competitive pricing environment that PSI's entry into the market would bring. Again, I believe a company has a perfect right to behave in this take-no-prisoners fashion. Whether it is wise to do so is another question. Newton teaches that for every action there is a reaction. In a remarkable and rather hilarious turn of events, it seems that IBM is rediscovering this universal law of nature the hard way. For PSI has recently filed a so-called Article 82 action before the European Union against IBM for its refusal to license z/OS. Article 82 is the very same law that IBM and its mostly American allies in the misnamed European Committee for Interoperable Systems successfully leveraged in the recent EU court case that forced Microsoft to open its Windows file server APIs.

These days people sometimes formulate Newton's action-reaction law a little differently. I believe the expression is "what goes around comes around."