How the Server Ecosystems Stack Up

by Timothy Prickett Morgan

While the various consultancies like IDC and Gartner do an admirable job tracking server shipments each quarter, they do not talk much about installed bases and they never talk about ecosystems. But box counting only tells part of the story, as we all intuitively know. Gauging the size and strength of an installed base in terms of its aggregate spending--what I am calling a server platform ecosystem--is a very tough challenge. To be quite honest, the data to do this is quite thin.

That for the past two weeks in this "OS/400 Ecosystem" series of stories I have been making so many estimates and assumptions. This is more like cosmology than applied physics, which is why I don't think anyone tracks IT spending this way. To be quite honest, it is too hard, just like tracking all of the individual positions and velocities of atomic particles is to describe some macroscopic behavior of an object in the real world. I make no apologies for making estimates, no more than a cosmologist does. It is my hope that by framing the discussion of server ecosystems in the light that I have, it will lead others with a lot more resources than I can bring to bear--that means IDC, with $2.46 billion in annual sales and 13,000 employees, and Gartner, with $894 million in sales and 3,700 employees--to start tracking IT spending by ecosystem groupings, not just by boxes.

Having just said that box counting is not sufficient, it is, however, the first step in calculating a server ecosystem. And it is very hard to do. For one thing, the users of servers themselves often are not doing a very good job of tracking their server assets. They buy a reliable midrange Unix or iSeries machine, and they forget about it because it just does its work. They know they have lots of Windows and Linux servers, but they can't keep track of them all. They know exactly how many big OS/400, Unix, and mainframe boxes they have because they are so awfully expensive; many companies with small OS/400, Unix, or NetWare servers with turnkey applications on them had no idea they even had a particular brand of server.

Companies have machines in remote sites or in other company divisions, and they aren't sure just how many, partly because of a lack of controls, but also partly because servers have become so inexpensive that they can now be acquired with discretionary petty cash in company departments. No one has to go to the IT department to get an entry level server any more, and many people in charge of departments do not. Of course, this causes all kinds of costs and security risks, because sooner or later, something goes wrong and it is the IT department that is called in to fix the server and its applications, whether they knew about it or not.

So, talking about the server installed base worldwide involves a bit of guesswork and witchcraft. Just to make things complicated, throw in this idea: it is absolutely unclear how many of the 760 million PCs installed in the world as of the end of 2003 and probably the 850 million to 900 million PCs probably still installed at the end of 2004 were functioning as servers in peer-to-peer networks in small- and medium-sized businesses or within larger company departments. There are probably a huge number of small companies that have not yet bought a "real" server--meaning one that has been designed as one, with at least some of the expandability and reliability features that real servers have, such as ECC main memory, RAID data protection, a fair number of peripheral slots, expansion bays for tape drives and so forth. Exactly how many is unclear, but the odds are that this is on the order of tens of millions of machines.

To prove my point, consider this. Brad Anderson, the general manager of Hewlett-Packard's Industry Standard Server division makes a habit of tracking the market share of HP servers at Guild Companies. Two years ago, when we set up our data closet, it was 100 percent--four 1U ProLiant servers and a bigger 3U ProLiant box in a half rack. But today, after coping with noise and heating issues with these servers, which are rock-solid and give me no grief other than heat and noise--I built my own "lean, mean, green machine" servers using motherboards from VIA Technologies; we have four of these servers that we use as a load balancer, two Web servers, and a hot spare. Two of the ProLiants have been turned off and are kept as warm spares. At our tiny company, a box counter like IDC or Gartner could see the ProLiants (across all customers, of course, not in our data closet), but it could never see the servers I made myself. As far as the box counters are concerned, HP still has 100 percent market share. But in reality, that market share is only 50 percent.

It would be an understatement to say that coming up with accurate server installed bases by server platforms and then derive some server ecosystem statistics from is tricky. But we all like understatement, so let's just dive right in.

The "Real" Server Installed Bases

Despite all of the limitations I just brought up, I now want to limit the discussion to real servers--ones made by vendors and sold to end users. We have to start there. And we really ought to start with the X86 platform, since it dominates shipments.

By my estimates, there were probably around 16.5 million X86 servers running in the world in early 2004, when I first started thinking about server ecosystems and piecing together data. Many more millions have, of course, been thrown into dumps and scrap heaps in the past two decades. Most of these machines use processors developed by Intel, but a few percent of the installed base are using X86-clone chips supplied by Advanced Micro Devices, VIA, and a few other niche players.

About 9.5 million of these X86 machines were running one variant of Windows or another in early 2004, with many running Windows NT and Windows 2000. About 2 million of them were running Windows NT, according to various industry estimates. (IBM has cited this number in its Windows-to-Linux migration effort.) In 2003, new Windows licenses accounted for 4.4 million units, according to Gartner, but many of these were for Windows 2000 upgrades that were forced by Microsoft's draconian pricing that went into effect in July 2003. There was no new net footprint for a lot of these upgrades--exactly how many is subject to debate. It is hard to imagine that more than about a third of all Windows shipments in 2003 were actually for Windows Server 2003. IDC was reckoning that Windows 2003 would account for 29 percent of shipments in 2003, which means about 1.3 million units. That leaves a whopping 6.2 million units running one of the variants of Windows 2000 Server at the end of 2003.

Now, the Windows platform accounted for about 70 percent of all server shipments and 35 percent of revenues in 2004, according to Gartner. The question in terms of the installed base is how many of those were for new machines and how many were taken out. A big portion of the Windows NT and Windows 2003 installed base has certainly been upgraded, but I think the net new Windows installed base at the end of 2004 was not appreciably larger--maybe 1 million net new units. But Windows Server 2003 is probably about the same size as the Windows 2000 Server base on servers. I will be working to try to figure out how the Windows base looked at the end of 2004, but right now, I don't have any data on this.

According to an amalgam of Gartner and IT Jungle estimates, about 3.8 million new physical Windows servers were sold in 2003 (the remaining 600,000 units were upgrades of existing machines, presumably). Windows servers comprised about 33 percent of the $45 billion in worldwide server sales in 2003. Given a factor of 8:1 to roughly calculate the value of the Windows ecosystem (based on past statements Microsoft made in early 2001 about the size of the Windows ecosystem and other things IT analysts say about the relationship between a Windows server sale and all of the drag-along revenue it generates), that would put the entire Windows server ecosystem (servers, storage, software, services, and reseller markup) at around $125 billion in 2003, and probably at around $137 billion in 2004.

My best guess is that there were around 2.5 million Linux-based servers running in the world in early 2004. By this, I mean a real server chassis that has Linux on it--not a desktop machine tilted on its side running a server instance of Linux. The vast majority of these "real" Linux machines are running on 32-bit X86 hardware, but there are many thousands of Linux servers running on IBM iSeries midrange and zSeries mainframe boxes. At the current run rate, Linux is pushing about 20 percent of the mainframe MIPS that IBM sells in the zSeries line, and a zSeries engine running Linux costs around $100,000 compared to around $400,000 for a full-blown mainframe engine capable of supporting MVS or z/OS. Linux shipments on servers are growing at a healthy clip--close to 700,000 units in 2003 and just over $3 billion in sales. In 2002, Linux servers accounted for $2 billion in sales, and in 2001 it accounted for about $1 billion in sales.

Click to continue reading the next page of this story >>>