Counting Linux In

- by Paul Murphy, -


In management you get what you measure; in volume sales, you get what the press reports. In the case of Linux installs, what the press reports is license sales numbers --and letting that continue would be a terrible mistake hurting everyone involved with open source products like Linux.


If you sell products, measuring those sales in terms of dollars during some reporting cycle like a quarterly or annual period makes perfect sense: it's dollar's you're interested in and so dollars you measure. That's not true, however, for the open source community; if you give away the product, then usage is the only measure that counts.

Unfortunately no one provides a widely accepted measure of open source installations or usage and the measures that do exist consistently under-estimate the actual install rate.

Consider, for example, the impact of IDC's recently released comparisons of its own server and desktop shipping statistics for the years 2002 and 2001. To quote from the press release: (because I can't quote from the report, at $3,000 I haven't seen it.)

Microsoft's share of worldwide server operating environment (SOE) new license shipments grew from 50.5% in 2001 to 55.1% in 2002, while the company's client operating environment (COE) new license shipments inched up from 93.2% to 93.8% of the worldwide COE market. By comparison, paid shipments of Linux SOEs captured 23.1% of the market, and Linux COE paid shipments accounted for 2.8% of the market total in 2002.

To an editor working in the mainstream media this is a "dog bites man" story of little interest to the public. After all, there's no news value to a story confirming what everyone already knows: that Microsoft is the invincible leader in this market. Thus the typical editor will minimize the exposure given the story even while being personally affected by it --because this confirmation of Microsoft's market dominance reduces that editor's interest in pursuing stories about Unix alternatives like BSD, Linux, MacOS X, or Solaris.

Imagine, in contrast, a comparable press release from a credible agency putting the combined Unix usage close to, or ahead of, Microsoft server usage. This would be widely seen as a a "man bites dog" story and thus be likely to attract major mainstream press attention. Given the realities of the marketplace, widespread press attention like that would almost certainly lead to significant new interest in understanding and exploring the Unix alternatives to Microsoft.

In other words, if real usage numbers are in fact much larger than license sales, then the open source community would have much to gain from getting and publizing those numbers.

It is, of course, impossible to predict what the real numbers would look like but we can think about three factors likely to significantly affect the difference between license numbers and installation numbers.

The first of these is quite obvious. Most people understand that Microsoft's Enterprise 6 licensing and related support contracts motivate companies to buy more licenses than they actually install while Unix licensing terms allow for multiple installs and thus motivate people to install more copies than they buy CDs and support for.

The second cause is much more subtle but probably more important. Few Microsoft installs last more than two years without significant change and the Microsoft sales cycle, for obvious reasons, aligns nicely with the annual reporting cycle. In contrast, Unix installs tend to last many years without significant change --people don't fix things that work-- and the open source installation cycle therefore tends to be largely unrelated to the annual reporting cycle.

We don't know how big an effect this is. It's easy to find people running Solaris 2.7 or Red Hat 6.0 but there's no real information on the numbers involved and little, if any, on the ratio of upgrades versus new installs among those who order new software CDs.

Consider, for example, the rather special case of Apple's desktop role. IDC's numbers suggest a 2.9% annual market share, amounting to around 3.5 million new licenses, for desktop MacOS X. If these were all upgrades, a typical five year cycle would suggest a real Apple usage share on the order of 12%; conversely, interpreting the number as all new sales would suggest significant market growth but say nothing about the installed base. Thus, although IDC's basic 2.9% number could reasonably lead the reader to conclude that the the Mac is almost irrelevant as a desktop competitor, both extremal cases support the opposite conclusion: that Apple is a growing desktop competitor. In reality, Apple's usage is probably nearer the middle indicating both a loyal customer base in the 8% range and significant new growth-- but the point is simply that the annual sales percentage is of very limited value as a guide to actual usage where the upgrade cycle is largely unrelated to the reporting cycle.

The third cause is likely to be the least important now and among the most important in the future. In the IDC numbers as reported in the press release, a license is a license: meaning that a Linux server used to run samba for 80 users counts the same as a Windows 2003 Server license bought for an NT server linking 25 people. Most egregiously, IDC's numbers put "combined Unix" --a term which excludes the three most popular Unix variants-- at 11% of shipments, directly equating a 5 CAL, $799, Windows 2000 Server license to the Solaris license shipped with a 15K serving 5,000 users.

Overall, therefore, IDC sales numbers almost certainly understate the real role of Unix by a considerable margin while overstating the role of the latest Windows Server and thus distort the public's perception of the server OS marketplace. Since "nothing sells like success;" i.e. because the public's perception of relative success affects sales, these discrepancies act to constrain Unix growth while boosting Microsoft's sales.

The open source community in general, and the BSD/Linux groups in particular, need to get together to fix this by collecting and publizing real installation numbers. One of the industry consortia or other organizations could, for example, ask every open source developer to put an anonymous registration option in the install procedure and then set up a shared data collection and analysis operation to receive and count the installation notices. That's what happens with Pine now; and yes there are some issues, but overall their numbers are among the most credible in the industry.