Lately there has been a kerfuffle in the blogosphere between partisans and opponents of the view that the Linux server boom is over. Supporters point to ominous IDC and Gartner data showing that growth in Linux server revenue has plummeted from over 50% in 2003 to a little over 10% in Q3 of this year. As you might imagine, these market share numbers have Linux fans in a tizzy. Responses run the gamut from anger (HackFUD and Linux.com), to denial (Linux Foundation), to acceptance (Matt Asay).

While it's true these server market numbers measure factory installed systems only and thus miss the many copies of Linux installed by users themselves, it's still hard to avoid the conclusion that Linux server sales growth is cooling off. I feel reasonably qualified to make such a judgment, because way back in 2004 my firm, Peerstone Research, was the first to publish actual hard data proving that use of Linux as a platform for core enterprise applications such as SAP and Oracle was surging.

Mind you, 10% growth on an annual revenue base that certainly exceeded $5 billion in 2007 is not exactly tepid. A few lucky server vendors took home $500 million more last year than they did they year before thanks to Linux. But compared to the glory years from 2003 to 2006 when Linux was narrowing the market share gap with Windows Server by leaps and bounds every quarter, the new growth rate could be seen as a disappointment.

But I don't see it that way. Why not? Because I think that Linux vs. Windows Server is an apples-to-oranges comparison that is stacked in Microsoft's favor.

The real comparison should be Linux + Unix vs. Windows Server. Consider these IDC market share numbers for Q3 2007:

The picture painted by this table looks quite a bit different from the simplistic view that Linux is falling short of Windows in the data center. What is actually happening is that the enterprise server market is consolidating decisively into two vast continents of approximately equal size and rate of growth, one based on the shared heritage of Unix and Linux, the other based on Microsoft's two-decades long effort to extend the operating system tree with an alternative branch derived from Dave Cutler's Windows NT (and ultimately from Digital Equipment's VAX-VMS). Together these two continents now account for just over 85% of server sales, up from 80% in 2006.

It is a pretty sure bet that the domination of these two operating environments in the enterprise IT market is only going to increase. Sure, IBM's mainframe z/OS and its mini-computer OS formerly known as OS/400 are no doubt going to hang around forever, but it's pretty obvious that they aren't going to be growing their share any more.

It is equally safe to say that neither Linux + Unix nor Windows Server is likely to displace or marginalize the other. The media, the fan clubs and the vendor PR departments like to focus on head-to-head competitions where customers do a feature-by-feature comparison before deciding which OS to buy or whether to rip out an existing OS and replace it by another. But this is a completely distorted image of the way the world really works. Real users in real companies buy applications, not operating systems. It is true however that most applications have a de facto affinity for some particular OS family, and thus often end up dragging along with them a specific stack.

The nature of the affinity between applications and OS families varies considerably. The vast majority of Unix applications can be ported quickly and easily to Linux because of fundamental similarities in the programming model. So it is quite natural to find that many traditional Unix applications – for example, Oracle databases and Java application servers – have shifted to Linux in a big way. Moving from Unix or Linux to Windows, however, is not always so effortless, because there are key architectural differences between the two families (e.g. between threads and processes). But in view of the size of the Windows market and the amount of money to be made there, some vendors of Unix or Linux apps have decided to bite the bullet and make the effort anyway (Oracle is a good example). Microsoft, on the other hand, has so far chosen to withhold its star applications such as SQL Server and Exchange from Linux for purely strategic reasons (despite persistent rumors to the contrary which seem to crop up at least once a year). So here the affinity is involuntary.

Whatever the nature of the affinity, it works. The market researchers have found very little evidence that customers readily switch from one OS continent to the other, except when there is a big price difference. From 2000 to roughly 2005, for example, a significant migration away from Unix occurred because the traditional Unix vendors (Sun, IBM, HP) foolishly allowed Intel and AMD based servers to open up a serious price advantage over their proprietary gear, inducing many Solaris, AIX and HP-UX users to switch to the then rapidly-maturing Microsoft stack running on cheaper hardware. But with the rise of truly enterprise-ready Linux distros such as RHEL or SLES, cost-conscious Unix switchers became far more likely to stick to their natural OS family (as the Peerstone 2004 survey cited above clearly demonstrated). Now that the Unix vendors have aggressively revamped their offerings to reduce the price gap between their wares and Linux running on standard Intel or AMD boxes, the outflow appears to be ebbing.

It should be clear by now that I don't accept the partisan viewpoint which draws a sharp distinction between "good" open source Linux and "bad" proprietary Unix. This distinction is pure politics and marketing spin, with no deep technical foundation. As I argued in my last post, all the credible open source players are in business to make a buck or are subsidized by someone who is. A few open source fanatics hate business and have developed the bizarre view that all software "just wants to be free". I suspect that most of these software socialists have secure day jobs (or trust funds). But the more sensible commercial open sourcers have merely leveraged the politically correct halo of their innovative distribution models to differentiate themselves from competitors who sell very similar products. Take RHEL and Solaris. Red Hat and Sun will cite you chapter and verse about how their operating systems differ, but when you compare them to Windows Server you can clearly see they are as alike as two peas in a pod.

How much more market share Linux can take from Unix will depend in large measure on which camp moves faster to adopt the other side's innovations. After letting Linux eat its lunch for years, Sun is now busy copying Red Hat. First it rolled out cheaper AMD and Intel hardware, then it switched from closed source to a compromise open source license that is sufficiently GPL-unfriendly to prevent migration of its source code to Linux. Now the venerable Unix system vendor is moving Solaris to a two-speed business model essentially identical to Red Hat's: a fully supported enterprise distribution (Solaris 10, equivalent to RHEL), and an unsupported but free "rolling beta" distro (forthcoming binary version of OpenSolaris, equivalent to Fedora).

While Linux and Unix will continue to duke it out for share as they grow closer – may the best OS win! – there isn't going to be so much back-and-forth between the Linux-Unix and Microsoft continents. When equivalent amounts of compute power cost the same and similar (or identical) applications are available, the customer's incentives to switch to a wholly different OS continent tend to fade away. We have firmly entered the dual-stack era of IT history, and we're not going to be leaving it any time in the foreseeable future. True, despite the equality in server spending, there are certainly more individual servers out there running some version of Windows than Linux or Unix. This is because Unix servers are still on average bigger and more expensive, with lots of processors or – in the case of Sun's "Niagara" machines – lots of cores per box. But if you reason in terms of effective server compute capacity rather than merely counting boxes, things even out again.

The obvious consequence of this bi-polar world is that management software which can function effectively across the two OS continents – and even occasionally join them together – will be at a premium. Administering and securing each continent with its own duplicate set of tools will be hugely inefficient. Nobody will want to run two separate directory and access control systems, two separate backup systems, two separate file server architectures, two separate security systems, two separate virtualization infrastructures, two separate provisioning and patching systems, let alone two separate management consoles for each and every routine action that system administrators need to take.

My top prediction for 2008 is that the ability to manage both Microsoft and Linux-Unix stacks seamlessly will become the single most important competitive differentiator for all utility and management software offerings. Microsoft itself has made some noise about wanting to move in this direction. However, Redmond isn't usually the first to market, so there are bound to be plenty of opportunities for other vendors as well, both large and small. The server OS wars appear to be winding down into a lasting stalemate. Let the interoperability competition begin.