Linux is threat to Unix, not Windows

Unless you've had your head under a rock for the last several years, you know that Linux has been gaining a considerable amount of attention and even market share. Through the use of shallow thinking and logical fallacies, many conclude, therefore, that Windows is losing market share to Linux. In fact, Linux is only rarely in competition with Windows. The real threat from Linux is to the establishment Unix versions, principally Sun's Solaris.

Everyone knows that in the desktop market Linux has no traction. There is an honest question of what it would take for Linux to gain some desktop market share, and the failure of companies like VA Linux and Dell to sell Linux desktops is a pretty clear sign that few people currently want to buy them. Certainly there are lots (depending on your definition of "lots") of people who do run Linux on desktop systems, but they're a teeny minority. I have my own theories for why they are likely to stay a teeny minority, but I'll get into that some other time. For the issue at hand though, ironically, Linux is not much of a desktop threat to big company Unix versions because those too have little in the way of desktop market share.

Much more interesting is the state of the server OS business. It's harder to get a feel for this from just talking to people and the publicly available research seems a little less trustworthy to me. One key thing I look at is the platforms for which third parties are selling products, and those are mostly Windows, Linux and Solaris. Almost everyone with a commercial server product to sell offers a version for Windows NT/2000; most of them have a Unix or Linux version, possibly both. You can't glean much from this other than that there are a lot of people running Windows and Linux/Unix servers.

Common sense can be a nifty tool at times like this. Linux is a Unix clone and is therefore architecturally similar to the more commercial Unix versions like Solaris, though there are some differences between them. Windows and Linux/Unix, however, are quite different. So if you've got a Windows server, migrating to a Linux/Unix server is a major and difficult event. The same would go for migrating from Linux/Unix to Windows. Even migrating from one Linux/Unix to another is difficult, but nowhere near as hard as switching to Windows. Since Unix users would have a much easier time migrating to Linux, it's no surprise that they constitute the bulk of those migrations.

All of this is verified by the most recent Web server survey from Netcraft. This is a rather famous survey, usually misrepresented to indicate Apache's overwhelming dominance in the Web server market. The famous survey, which in this month indicates Apache at 59.51 percent, IIS at 27.46 percent, has always been a survey of domains, not servers. In other words, the survey has never necessarily indicated that 59.51 percent of Web servers are running Apache, but that 59.51 percent of Web sites were running on Apache servers.

The reason for this phenomenon is that Apache and Linux are overwhelmingly popular with Web hosting services, for the very good reasons that they work well and are free. Hosting services can run thousands of small, low-bandwidth home pages and brochureware sites on a single Linux/Apache server. Whether IIS is better or worse for this market technically is irrelevant--IIS costs money. (In spite of this, you can still get hosted IIS Web sites for reasonable, even low prices. I'm not sure how these companies can compete.)

The latest Netcraft survey goes the extra step of attempting to measure the number of physical servers on which each product runs. This isn't easy to calculate, and Netcraft assumes a fairly high margin of error (less than +/- 10 percent). But the numbers are still compelling: About half of all the servers surveyed run some version of Windows. About 30 percent run Linux. Solaris and BSD come in at 7 and 6 percent respectively, and everything else is spare change. This is seriously at odds with the common perception of the Internet, in no small part because of previous misrepresentations of the Netcraft surveys.

Of course this survey only attempts to measure HTTP servers, meaning Web servers. It says nothing about the numbers for database servers and file servers and other miscellaneous servers, and it definitely says nothing about the amount of revenue taken by vendors for systems; some Sun systems cost millions. But I would have assumed that if Linux and Unix had an advantage anywhere, it would be on Web servers, where Linux software is mature and well understood. My best wild guess is that Windows does even better in other areas.

Netcraft has followed these numbers over time, and it turns out that the gains in Linux have come at the expense of other Unix versions--not Windows--which makes sense for the reasons described above. And a recent analysis by D.H. Brown Associates indicates that at least some Linux versions are gaining in quality and scalability.

For many years, the overall market was growing so much that Linux could grow and not be so apparent a threat to Unix. After all, Linux wasn't really a big-time OS capable of running on high-end systems and didn't have big, stable companies behind it. But even if there are lots of systems for which Linux isn't yet a realistic alternative, the number and share for which it is a feasible option has grown. As the credibility of Linux increases in the Unix world, the number of users who will abandon their expensive and proprietary Unix systems for Linux will grow as well. And now even IBM and the new HP-Compaq combo seem to care more about Linux than about their own proprietary Unixes.

Both Sun and Microsoft would point out that the cost of the actual operating system is not a significant part of the overall system cost, especially if you look at total cost of ownership (TCO). But in the case of Unix, Linux also brings the option of moving from proprietary hardware to off-the-shelf x86-architecture PCs. Netcraft also saw this as part of the trend. Once again, this isn't an issue for Windows users (who have always been on that hardware).

In the long term, it looks like there will be two operating systems that matter: Windows and Linux. I would guess that some day something will change to make migration costs less onerous, and users will then be able to move from one to the other more easily. But in the meantime, I wonder when the rest of the Unix business will begin to take Linux for the death threat that it is.