Open Source

(LinuxWorld) — Evans Data Corporation hired me to help out with a research report focused on Linux developers. They surveyed a broad range of developers, including VARs, consultants, developers working for ISVs and IT-developers for companies of every imaginable size. It turns out there are so many jewels in the survey results that it's difficult to decide which ones to put on display here at LinuxWorld.

After much consideration, I chose a nearly flawless diamond. It replaces the cubic zirconia otherwise known as the axiom that Linux is taking more market share from Unix than from Windows. I had long suspected this was a fallacy, and Evans' data confirms my suspicions.

Of the developers surveyed, more than 50 percent who now develop primarily for Linux used to develop primarily for Windows. Only 30 percent used to develop for some other Unix or Unix derivative. In case you missed it, the operative phrase is "used to." In other words, this is not a prediction of an emerging trend. It is cold, hard information about what has already transpired, and it withstands the scrutiny of a jeweler's eyepiece.

If you want to talk trends, however, it is instructive to look at the shift in focus that developers plan for next year. This year, Windows commands attention of 50 percent of the developers. Roughly 40 percent focus primarily on Linux. These priorities will switch places almost number-for-number next year. Actually, more than 50 percent plan to focus primarily on Linux and less than 40 percent on Windows, so the switch favors Linux. But because the differences are within the margin of error, it is essentially a symmetrical reversal of fortune where Linux will take priority over Windows starting next year.

Why the misconception

There are dozens of reasons why people have underestimated how quickly Linux has been grabbing Windows' market share, but the Evans data confirms one of my pet theories. Windows market share is usually estimated by the units of Windows Microsoft claims to have shipped. This figure is already skewed, because it includes every unsold box of Windows XP sitting on shelves at Best Buy or Circuit City. More significant, however, is the fact that it includes every PC with a pre-installed version of Windows.

Linux market share, on the other hand, is usually estimated based on surveys, number of commercial boxes sold and the number of downloads.

The actual market-share shift from Windows to Linux is obviously more complicated. When someone purchases a PC with Windows pre-installed, and then overwrites that pre-installed Windows with Linux, nobody subtracts "one" from the installed base of Windows and then recalculates the Windows market share. So Windows starts out with a false boost and maintains its illusory market share even as it gets replaced by Linux.

Alternative methods are still inadequate

One might suggest that a survey of operating systems currently in use could resolve this issue. It would, indeed, provide a snapshot, assuming companies could assess accurately what operating systems they have installed. Personally, I think this number would be highly unreliable. Unless the company is small enough to keep track of what, where and when software is installed, people tend to take a wild guess unless they have the threat of an audit hanging over them. If it was easy to track installations, companies wouldn't be as inclined to give in to Microsoft audit threats and buy into Licensing 6.0.

Assuming you could get a reliable snapshot of Windows and Linux machines installed, there is yet another complicating factor that renders this less-meaningful. Linux runs on far more hardware platforms, including those that outperform Intel-based systems. Even on Intel, Linux outperforms Windows. Let's say one company installs two Linux servers, but the company next door chooses five Windows servers to handle the same load. In this case, the market-share growth for Windows and Linux would be equal in one respect (the amount of work performed by the servers) but skewed to favor Windows with respect to the way market share is represented for public consumption (the actual number of units in use). Once you include Linux on non-Intel hardware, which can mean anything up to a mainframe, the disparity becomes even more dramatic. The irony here is that Windows gets an unfair market-share boost because it is inferior to Linux and requires more installations to do the same work.

On the other hand, one might claim that distro-hopping might skew numbers in favor of Linux. Distro-hopping is when someone buys or downloads multiple distributions, then eventually commits to the one they like best and discards the others. In the first place, I can't imagine how that could begin to offset the unknown numbers of copies of their favorite distribution. In the second place, Windows has a similar anomaly that inflates Windows market share even more: upgrades. When you upgrade your server from Windows 2000 to Windows XP, I am not aware of anyone who subtracts "one" from the number of installations of Windows.

The phantom shift

Let's get back to the invisible market-shift. One needs to ask the following questions of the Evans data to understand the stealth process with which Linux is displacing Windows market share:

1. Are people really installing Linux in a way that is difficult to estimate from downloads and sales?

The answer is a resounding yes. One misleading factor is that Red Hat Linux has by far the largest share among the developers surveyed. Red Hat is a commercial brand of Linux, so it is tempting to assume that one can estimate the number of Red Hat Linux installations based on the number of copies of Red Hat Linux sold. But Red Hat, like most other commercial brands of Linux, does not require you to purchase a separate license for every machine upon which you install their Linux (Red Hat recently implemented new policies to encourage people to purchase separate licenses, but the company does not yet require separate licenses). Therefore, every copy of Red Hat Linux sold can actually represent anything from zero Linux machines to every box in the company... and anything in-between.

2. If Red Hat is the most popular Linux, doesn't that mean we can get at least a rough idea of the starting point by counting Red Hat sales?

The answer is a resounding no. Almost 40 percent of developers say they get their Linux by downloading it over the Internet. This figure includes those people who say they favor Red Hat, so a significant portion of those Red Hat installations is likely to be based on downloads, not packaged boxes. This means that any Linux-distribution sales-figures from Red Hat or any other company are relatively meaningless.

This 40-percent figure is important in another way. Only commercial boxes of Linux are likely to carry any license restrictions. It is safe to assume that a downloaded copy of Linux does not carry per-machine or per-user license restrictions (Note: I know of no exceptions to this assumption, but although I will allow for the possibility, it is extremely unlikely). With this in mind, each download can represent virtually any number of Linux installations.

3. But are people really installing Linux over Windows?

The answer is a resounding yes. More than 70 percent of developers say they install Linux over whatever OS happens to be pre-installed on the machines they buy. Could that pre-installed OS be another flavor of Linux? Sure. In that case, Linux gets counted fairly, since one uncounted copy of Linux replaces a copy that is counted by being pre-installed. But does anyone want to place a wager on which OS is more likely to be pre-installed on those machines?

The survey includes many other interesting facts and trends, some of which I may address in future columns. It also raised a lot of questions it did not answer, but that is to be expected. I was recommended to Evans by a very good friend of mine, Esther Schindler. You may remember Esther as one of the few bold proponents of OS/2 back when OS/2 had a chance. Esther warned me about the "occupational hazard" of research surveys: when you begin to analyze the survey results, the information takes you down roads that make you want to go back and ask more questions. Of course, it's too late for that round. I hope I get the opportunity to resolve these issues in a future survey — although there's no doubt that those answers will raise even more questions for the survey after that. Regardless, this was one of the most difficult and fascinating exercises I've been through. Among other benefits, I'm glad to see that Evans can now debunk a popular myth about Linux market-share growth.