Recently the Desktop Linux Survey was published. It showed how the different Free Desktop components and distributions are rated among different users. While of course these numbers are by no means reliable it makes fun to use them for some simple calculations.
The last Desktop Linux Survey was interesting, but hardly surprising. But most important is that this was no real statistic: the sample was not representative by any means.
Also, as Distrowatch.com noted, there are inconsistencies if you compare these results with other surveys. For example the web server logs for Distrowatch.com give OpenSuse 5 % of the cake, while the Desktop Linux Survey lifts OpenSuse’s share to 20 %.
Still, these numbers give some opportunities to play with. And that’s something I like. 🙂
Fedora
Let’s take Fedora: in all three columns listed at Distrowatch it gets roughly 5 %. Let’s take this as true.
Now, according to Fedora’s statistic tool smolt there are around 125,000 registered Fedora users out there. If Fedora really has a market share of 5 % this would result in 2.5 million users.
But smolt is only an opt-in tool and is not offered during text installations (Servers, etc.) at all. According to Fedora’s statistics page the estimated ratio between smolt data and reality is 1 to 10. This would result in 1.25 million Fedora users – and therefore in 25 million Linux users.
Looks nice, especially considering the fact that Apple’s userbase was around 22 million half a year ago.
Ubuntu
There is another possible approach to these numbers: Ubuntu’s Mark Shuttleworth estimated 8 million Ubuntu users in an interview in late December 2006. And both the Linux Desktop Survey and the web server logs from Distro Watch gave Ubuntu roughly 30 % of the cake. This makes more than 26.6 million Linux users 8 month ago.
German traffic lights
The numbers mentioned above only include popular distributions. They do not include embedded devices or corporate installations like the one of the German drugstore chain.
Now Linuxdevices.com posted an article about the German company Signalbau Huber. Signalbau Huber produces and installs traffic light systems across Europe. And they mentioned now in this article that they are going to switch their software from some Debian based thing to Sysgo’s embedded Linux solution ElinOS.
Unfortunately the article doesn’t say anything about how many Debian installations are already out there, but I gues there are quite some across Europe. This is again an example for the fact that it is impossible to track all Linux installations out there – and that there are many more than most people expect.
/home/liquidat 
Nice analysis!
Well, more than playing with numbers than an analysis, but thanks anyway 🙂
I think most interesting in this regard is the fact that I get roughly 25 million users both times. This shows that it is quite realistic to assume that there are at least as many Linux users out there as there are Mac OS users.
Btw., do you have any numbers about the Windows users base?
Recently I was at a pub drinking and playing GoldenTee, someone passing by knocked the power cord lose on the arcade and the screen flickered for a moment. Then a bios flashed up and seconds later I was watching a Linux kernel boot up.
Just another instance of embedded Linux thats out there many people are probably unaware of, how many bars have GoldenTee machines in them? And I’d imagine there are loads more arcade systems out there running Linux in the back ground.
What I find interesting is that out of hundreds of distros only a dozen or so have significant userbase. People often complain that there is too much variety and it’s hard to make packages for all these different flavours. Well if you look at top 5 distros and their derivatives (for example Mepis uses Debian packages for most applications, Mint uses Ubuntu, CentOS uses RHEL) you pretty much cover 90% of Linux userbase.
This kind of shoots down the argument that there is lack of commercial software on Linux because it’s too hard to make packages for every single distro.
Mike, I like that kind of stories 🙂
voislav: To be exact you only have to cover two major package types: rpm and deb based types.
However, even producing packages for several versions of one distribution can be tricky, and the rpm distributions are lacking several standards to pack one rpm which could be used among all the others.
Of course, with a lot of knowledge you can produce cross-distribution rpms and cross distribution debs, but it is still very hard to do so.
Shuttle worth didn’t really guess 8M users. He can get a fairly accurate count of how many Ubuntu users there are by how many unique IPs hit his update servers.
Those distrowatch figures, I guess those are regular hits from users all over the world checking out distros?
Never forget that people on some specific distros (when the distro is bad, or when the user try out many distros) are more likely to check out alternatives. So x% hits for distro Y does not mean that there is a usage of x% for that distro, only that those guys/girls are checking out distrowatch.
(I mention this because I read somewhere that there were a lot of hits from vista users on distrowatch, which is not so surprising since vista users might look for something better :D)
“Shuttle worth didn’t really guess 8M users. He can get a fairly accurate count of how many Ubuntu users there are by how many unique IPs hit his update servers.”
Actually these numbers may be skewed either way. For example, one system using a dialup ISP may be given N different IPs, depending on loading when the system connects. This would mean one system could be counted as many…
Going the other way, I have several systems on a LAN at my house behind one firewall/router/DSL modem. So they would all look like one (IP) I get and keep from my ISP…
There really is no good way to get a count other than draconian “the system must phone home” with some unique ID such as the MAC address of its NIC… But even that would miss a lot of the embedded systems…
Fedora Core 6 userbase has hit 2 million in Q1 2006.
http://linux.slashdot.org/article.pl?sid=07/03/06/1437245
About the smolt statistics, please consider that:
– smolt is offered as an option in graphical installs and not everyone
chooses to report their machines
– smolt showed up in Fedora 7 first and although you can now install it
on Fedora 6 as well from FEdora Extras, not too many users bother to
actually install and setup reporting
You quoted the Fedora statistics wiki URL, just read it more accurately.
🙂
The yum statistics are more accurate with about 2.8 million users
(unique IPs) for Fedora 6. The oldest statistics seems to be collected
some time after releasing Fedora 7. Yum statistics for Fedora 7
showed that it has hit the 1 million mark this week.
Back in FEB this year, I spoke with sabdfl (Shuttleworth) regarding unique IP addresses downloading the Ubuntu Feisty Fawn. On their servers alone, the totals are way up there.
https://wiki.ubuntu.com/MeetingLogs/openweekfeisty/askmark2?highlight=%28estimated+numbers+for%29
Pieter, the intersting part about the distrowatch statistics is the bit where they extract the data from the user agent string. Of course that one can be altered, but it gives a first glimpse.
spr0k3t, nice data, I didn’t knew them. Although it is a bit hard to estimate reliable numbers out of these data 😉
Zoltán, well, before you advice me to read data properly you might consider reading the actual article you are posting your comments at, hm? 😉
I did mention right in my post that smolt is an opt-in tool. I also referred to the yum data (I mentioned the ratio between smolt and reality; this ratio is derived from the yum data and the smolt data). You seem to somehow missed the second paragraph 😉
Anyway, about Fedora Core 6: I excluded it on purpose because the users which moved from Fedora Core 6 to Fedora 7 are not filtered – they come up in both counts.
And: I read the statistics page accurately. So accurately that my posts about Fedora statistics are often mentioned at the Fedora Weekly News Issues…