Using free and open source software (FOSS), advocates like to say, is not a popularity contest. It's about doing what's right. However, the Debian and Ubuntu Popularity Contest projects might disagree.
Using the Popularity Contest package, these two projects collect and post weekly anonymous reports about the software used on each system on which theyre installed. The result is a rare insight into the habits of FOSS users, including how often software is upgraded, what applications are used the most, and the differences between the expert-oriented Debian distribution and the more user-friendly Ubuntu distribution.
The main purpose of the project is to collect information for installation, so that the most widely used applications can be put on the first CD of an install set, or installed automatically. In Ubuntu's case, the collected information is also used for software ratings in Add/Remove Software.
Both projects explain that each host data source is identified by a random 128-bit hash, and that identifiers are removed after twenty days. However, those concerned about privacy can choose not to install it, or remove it without any difficulty.
Popularity-contest reports four main statistics for each package: The number of installations, the number of hosts on which the program in the last 30 days (or Votes, as the package calls this statistic), the number of hosts on which the program was not used in the last 30 days (or "Old"), and the number of hosts on which the package was upgraded ("Recent").
Unsurprisingly, core applications that are included in every system, such as dpkg or ncurses log the highest number of installs, and, as often as not, the highest number of Votes and Recent installs as well. The result, as some Ubuntu users complain, is that the popularity ratings in Add/Remove Software are skewed in favor of core applications, and newcomers can be mislead by the relative unpopularity of such useful desktop tools such as Inkscape.
However, if you look at the raw data on the project sites, you can easily ignore core packages and focus on whatever subject interests you.
Installations and upgrades
As you look at last week's figures, one of the most noticeable results is that, in both projects, the number under Votes and Recent are low compared to the number of Installs.
For instance, the Ubuntu results list 425,490 installations of Firefox 3.0, only 92,629 Votes and 63,034 Recent upgrades. In other words, although roughly half of Ubuntu installations reporting include Firefox, only about 20% have used it in the last 30 days, and only 13% have upgraded it.
With IceWeasel, Debian's non-branded version of Firefox, of 46,276 installs, only 25,797 (46%) are recently used, and 11,207 (24%) recently upgraded.
Percentages differ for other packages, but in almost every case, figures for use and upgrading are much lower than for installations. Probably this difference reflects how easy GNU/Linux distributions make the installation of software.
However, what is less obvious is why, when automatic upgrades are available on modern desktops, most people are not upgrading as soon as new packages are detected -- or perhaps not at all. Apparently, the average Ubuntu user and, to a lesser extent, the average Debian user, are far more cautious about upgrades than you would expect from their availability.
Another way you can use the popularity-contest data is as a measure of how comparable software is being used. For example, both Ubuntu and Debian show high numbers of GNOME desktop installations, as you might expect from such GNOME-centered distributions: GNOME is installed on 85% of Ubuntu installations and 50% of Debian installations, and has been used recently by 78% of Ubuntu users and 55% of Debian users.
The relatively low number of Debian users is probably explained by the fact that the distribution appeals to more advanced users than Ubuntu, and such users are more likely to choose one of the dozens of alternative desktops or window managers available, or even to use the command line.
This explanation gains credibility when you find that only one-eighth of Ubuntu users have KDE or Xfce desktops installed, despite the fact that Ubuntu maintains separate sub-distributions called Kubuntu and Xbuntu.
By contrast, the Debian results show only slightly higher percentages of KDE users than the Ubuntu results, and one-sixteenth of Debian installations include Xfce. The Debian results might be abnormally low for KDE if a large number of long-term users participate in popularity-contest, because at one point nine years ago, Debian seriously debated dropping KDE because of licensing issues. However, a more likely explanation is that Debian desktop usage is distributed more widely than Ubuntu's over other alternatives.
Doing an analysis on OpenOffice.org's applications is equally revealing. In Ubuntu's results, OpenOffice.org Writer is installed on almost every host measured by popularity-contest, while, on Debian, slightly more than half the hosts reporting have OpenOffice.org installed.
However, in both cases, the popularity of separate applications -- which can be installed separately -- is consistent. In each set of data, Writer, OpenOffice.org's word processor, is most likely to be installed and recently used. It is closely trailed by the Calc spreadsheet and the presentation program Impress.
In fourth place is Draw, and in fifth place the database Base. And, in the Ubuntu results, Base has about half the installations, and only about ten percent of recent uses of any of the other applications -- a result that may suggest that many desktop users continue to be intimidated by the idea of databases.
The popularity contest data also reveals how often equivalent programs are being used. For instance, on Debian, Evolution is the most commonly installed email reader, appearing on 50% of installations, but only a few percent have used it recently.
Meanwhile, 16% have KMail installed, and 15% have Icedove, Debian's non-branded version of Mozilla Thunderbird. You receive comparable results on Ubuntu, except that Evolution appears on 88% of installations and has a recently-used rate of about 5%.
These figures suggest that, although Evolution is installed with GNOME by default, it has been unable to obtain a high level of acceptance in its fragmented category. The same is true of KMail, despite the fact that most installations of KDE include it.
But perhaps the most interesting figures are those that compare FOSS packages with proprietary ones. Ubuntu installs the free nv driver for NVidia cards, but on over 86% of machines reporting, the proprietary NVidia driver is also installed. In much the same way, the free Radeon driver for ATI cards is installed on one-eighth of reporting Ubuntu machines, while just about as many machines use the proprietary fglrx driver.
But on Debian, the proprietary Nvidia driver has only one-fourteenth the installations of the free nv driver, and the proprietary fglrx driver one-eighth those of the free Radeon driver. Noticeably, too, although the Ubuntu statistics are roughly twelve times larger than the Debian ones, the number of installations of the free Intel video drivers on Debian is one-third that of the installations of the same driver on Ubuntu.
This sampling of possible inferences also seems to suggest that, while Ubuntu is based on Debian, the audiences for the two distributions are distinct.
In general, Debian users seem more eclectic in their use of software than Ubuntu users, and less likely to use an application simply because it is included by default. Debian users also seem more likely to be concerned to maintain a free installation than Ubuntu users -- a conclusion that is hardly surprising when you consider Debian's reputation for freedom, but is still interesting to see being supported by statistics.
The limits of analysis
To what extent last week's figures are typical is uncertain. Very likely, studying the figures over a longer period would produce different results. Possibly, too, those who participate in the Popularity Contests are not typical users of either Ubuntu or Debian.
For instance, the number of recent upgrades would undoubtedly depend on how recently the latest upgrade for a package was released, with upgrades peaking immediately after a release, and declining at the end of another release cycle.
Probably, too, any given set of figures includes some anomalies that would be evened out over time (which is one reason why I give approximate figures most of the time; citing the raw data would only give a false sense of precision).
It would be interesting to see how the results from Ubuntu and Debian compared to those of other major distributions -- and, conversely, to see Fedora's hardware-reporting Smolt ported to Ubuntu and Debian to give an even more fully rounded picture of user habits.
However, neither event is likely to happen in the immediate future. For now, Popularity Contest provides a rare view into the habits of two sets of free software users, confirming expectations in some cases, and also offering the occasional surprise.