How the Server Ecosystems Stack Up (Continued)

<<< Click to return to the first part of this story

According to IBM, the worldwide IT ecosystem for Linux--which is a measure of the hardware, software, and services purchased each year--was about $25 billion in 2003. As it turns out, the Linux market is accelerating still, and it broke above $5 billion in server sales in 2004. Using the same 8:1 ratio to value the ecosystem (which fits based on IBM's data), then the Linux server ecosystem was about $40 billion in 2004. (Back in late 2003, when I was asked on a consulting gig to estimate the size of the Linux ecosystem, I said that if current trends continue in 2004 and Linux just grows linearly, then $4 billion in Linux servers will be sold in 2004 and the Linux ecosystem will grow to about $32 billion. I also said that if Linux was at the tipping point, as many people--including myself--think it was, then its use in data centers and departments could skyrocket in 2004, pushing Linux server sales up to $5 billion and the Linux ecosystem to $40 billion or so. I think this happened.)

There is still more X86 iron out there running in the world aside from those servers running Windows and Linux. There are probably somewhere between 2 million and 2.5 million NetWare servers still out there running on Pentium and maybe even some 486 servers. These things persist. NetWare is a well-regarded print and file serving platform, and it has been updated with modern Internet capabilities. But the weight of Windows and Linux has crushed it, and it was therefore no accident that Novell paid $210 million to acquire commercial Linux distributor SUSE at the end of 2003.

Yet another large installed base is out there on the X86 platform: the SCO OpenServer and UnixWare variants of the Unix operating system. Before Linux became popular, SCO's OpenServer was the de facto alternative to Windows and NetWare and was very popular for a whole array of turnkey systems for small- and medium-sized businesses that were never even aware they were running Unix underneath their application software. This installed base, which numbers about 1 million machines, persists precisely because those applications do.

It is hard to value this SCO ecosystem, precisely because it is targeted at small customers that only spend money on systems and applications once every three, four, or five years. SCO certainly does not make very much money selling its Unix software, but that doesn't mean the SCO Unix ecosystem is valueless. About six years ago, SCO Unix accounted for as much as 20 percent of total X86 server operating systems. The advent of Linux, the increasing dominance of Windows and, to a small degree, SCO's legal wranglings with IBM over Unix-Linux intellectual property have probably hurt that ecosystem substantially. Nonetheless, it is still probably on the order of several billion dollars--and could even be a great deal larger.

No one knows for sure how many FreeBSD, OpenBSD, and NetBSD Unix servers are out there on X86 iron--only a few tens of thousands of users are registered, but the BSD base could number in the millions for all anyone knows since registration is not required. The BSD ecosystem is miniscule, probably considerably smaller than that of SCO Unix, but BSD is a great product and highly regarded. (We use FreeBSD on some of our servers, in fact.) In a just world, BSD would have made Linux absolutely unnecessary. But it didn't.

For the past decade, the RISC/Unix server market has been the juggernaut of the IT industry, and despite the rapid adoption of Windows and Linux, Unix still accounts for about a third of worldwide server revenues. This is true because so many companies have mission-critical applications running on Unix iron. Having made the painful jump from proprietary systems to Unix in the late 1980s, early 1990s, or even only a few years ago, companies are loathe to do it again. So they continue to spend money on Unix machines, which have excellent reliability, security, and expandability features, even well after X86 alternatives with much better price/performance are available. Making the quantum leap from Unix to Windows is a huge pain. If Linux takes off anywhere, it will be among the vast numbers of companies who have Unix today but who want to be on cheaper Linux iron tomorrow. Linux is, in effect, a poor man's Unix and the skillsets and administration experience for Unix can be almost directly related to Linux. The gap between the two is small. This, if anything, is the real threat to Unix system vendors and Microsoft alike. In any event, there are probably around 1.6 million or so RISC/Unix servers installed worldwide that are still doing useful work. A lot of these machines are installed as front-end servers in the financial services, telecom, and service provider industries. In fact, my best guess is that only around 250,000 Unix machines installed worldwide have eight or more processors.

In terms of server counts, Sun Microsystems has the largest installed base of Unix machines because its UltraSparc iron was the preferred platform during the ISP, telecom, and dot-com build outs in the late 1990s. Hewlett-Packard hitched its star to RISC/Unix in the late 1980s, and quickly developed the broadest and deepest ecosystem in the Unix market with its PA-RISC servers. IBM and its partner, Bull, were late to the Unix game in 1990 and were pretty much also-rans in the Unix racket until the 1997 launch of the 64-bit Apache PowerPC RS/6000-AS/400 platforms. From this moment forward, IBM began pushing Unix very hard and eventually created a business that rivals its mainframe line and severely undercut the growth prospects of its proprietary midrange line.

My best guess is that about 60 percent of the 1.6 million Unix server installed base is comprised of Sun machines, which comes to about 960,000 servers. HP's share of Unix shipments is much smaller--it has ranged from 20 percent to 30 percent of Sun's shipment rate in the late 1990s and early 2000s--and so it stands to reason that HP's share of the Unix server installed base is much lower. A reasonable estimate is that HP has about 240,000 Unix servers of one vintage or another running out there in the world, mostly at midrange shops doing the kind of ERP work that proprietary midrange and mainframe machines do. IBM has probably sold about 520,000 RS/6000 and pSeries Unix servers since the introduction of its Unix server line in 1990, and there are probably around 310,000 of them still doing useful work. IBM sold a lot of modestly powerful machines and some very large machines as adjuncts to mainframes or outright replacements; the company also sold lots of Unix clusters for government research and data warehousing. That leaves another 90,000 or so RISC/Unix boxes from a plethora of other vendors still out there, working away, around the globe from Fujitsu Siemens, Silicon Graphics, Cray, NEC, NCR, and a few others.

In terms ecosystem, HP says its HP 9000 ecosystem (including all hardware, software, and services contracts) came to about $20 billion in 2003 inside HP. It is probably double that once you take into account the HP Unix channel and third-party software and services. IBM and Sun are less forthcoming about the ecosystem value of their respective Unix platforms, but it will come to about the same ratios. In 2000, Sun was outselling HP in Unix by a factor of roughly 3:1, with Sun having about 50 percent of revenues and HP having about 20 percent. By 2001, HP had come within spitting distance of Sun, and IBM was only a bit behind. IBM tends to sell many more machines than HP, but still only managed to match HP's revenue share in 2000 and it fell behind HP in 2001. If the value of the ecosystem of a platform scales with the revenues of that platform at the vendor level (and it should, very roughly), then IBM should also have an AIX ecosystem worth about $15 billion a year inside IBM and about $30 billion outside. The ecosystem value for the Solaris Unix platform is not as large inside Sun since it partners for so much software and services, and it has certainly declined as Sun's sales have plummeted between calendar 2000 and 2003 when Sun's sales have shrunk from close to $17 billion to down to $11 billion. Sparc servers and related products probably account for 90 percent of that revenue, which means the Solaris server ecosystem is probably worth about $10 billion inside Sun. Outside of Sun, that ecosystem balloons much higher than its does for either HP or IBM because of Sun's partnerships and reseller channels--perhaps the total Solaris ecosystem is worth as much as $30 billion. This means that the Unix ecosystem of the top three vendors was about $100 billion in 2003, and there was probably another $20 billion for the other Unix players. That brings the total value of the Unix ecosystem--all the RISC/Unix players and SCO on X86--to about $125 billion in 2003, or roughly on par with the Windows on X86 ecosystem at the same time. I have a hard time believing that the Unix ecosystem grew in 2004, so I think it probably stayed about the same.

But this is not all there is to the server market, which comprises about 18.5 million machines worldwide. There are, according to IBM, about 20,000 or so IBM mainframe footprints. This was roughly the number of footprints prior to the Y2K rollover, and high-level IBMers in charge of the mainframe platform say this number, according to Gartner, is consistent. Notice the precision in that wording. IBM didn't verify the 20,000 footprint number, but merely said that Gartner says it is accurate. I happen to think that the actual mainframe base could be as low as 10,000 machine, and is probably closer to 15,000 footprints. Five years of server consolidation within the mainframe, 14 years of pressure from the Unix market and a decade of pressure from the Windows market cannot have allowed that total number of footprints to stand. IBM's mainframe sales were a staggering $13.1 billion in 1989, when the company did not yet have a Unix business and the AS/400 was only 18 months old. By 1993, mainframe sales had slipped to $6.5 billion, a level that IBM maintained for a few years as it made the transition from bipolar to CMOS mainframe engines. Sales continued to slip in the late 1990s to around $3 billion to $4 billion a year, the level that IBM has only been able to maintain because it has Linux driving some sales on the box. Without Linux, IBM's mainframe server sales would have been under $2.4 billion in 2003, and probably around $3 billion in 2004. That said, IBM mainframe shops buy huge amounts of hardware, software, and services, so even though they only represented about 7 percent of total server sales in 2003, the value of that mainframe ecosystem is very large--perhaps as high as $50 billion. In effect, mainframes drive twice as much money as you think they should based on how Unix servers do. This has a lot to do with IBM essentially having a monopoly on both mainframe hardware and software, and to a large extent, a lot of control over services relating to mainframes thanks to its Global Services unit.

IBM's largest customer base is without dispute the OS/400 server market, which has a worldwide customer base of about 215,000 unique customers and an installed base of about 450,000 machines worldwide--even after years of neglect and relatively poor price/performance (in terms of initial hardware and software costs). The OS/400 ecosystem has always been somewhat diminished since the very idea of the box--which has the operating system, database, compilers, and other system administration tools tightly integrated into a seamless system--means that there is less money for companies to make on it. As I said last week, it is hard to guess at the size of the OS/400 ecosystem, since OS/400 shops do so much of their own coding and they are not huge consumers of services. They do, however, buy third-party software and tools when it suits their needs. As I explained in the past two stories, I think the OS/400 ecosystem was about $6.5 billion in 2004, but if you consider the value of the people doing the coding and administration of those boxes, the ecosystem balloons up to about $65 billion. (The same would hold true, to a smaller extent, for all server platforms once you add the people costs into the equation. But the OS/400 ecosystem is, as I explained last week, dominated to a larger extent by people costs since OS/400 shops tend to invest in their own programmers, not best-of-breed, third-party applications.)

Rounding out the worldwide installed server base is another 375,000 or so HP VAXen and AlphaServers and another 150,000 machines including the HP 3000 proprietary minis and other gear from Unisys, Fujitsu, Siemens, NEC, Hitachi, Cray, and myriad other niche players and companies who have long since gone out of business. Old servers persist in the oddest of places.

Now, let's sum this up with two main points after all that data.

First, Windows and Linux are the only platforms that are showing revenue growth at the box level, and soon Windows server growth will halt and Linux will be the only growing server platform. This would seem to imply that, over the long run, Linux will be the only growing server ecosystem, too. The others will stabilize or decline.

Second, the other server ecosystems--OS/400, Unix, and mainframes--will be around for a long, long time. I think that all three have reached relatively stable points at this time, in fact. And there is a decent shot that the OS/400 ecosystem could see an upswing now that the iSeries is a more inclusive box with a little more IBM marketing muscle behind it.

Next week, I will wrap up this discussion of server platform ecosystems with some thoughts about the future.

RELATED STORIES

How Big Is the OS/400 Ecosystem?

The OS/400 Ecosystem, Part 2

TFH Flashback: Critical Mass, November 1993 issue