Measuring Ubuntu's Market Share
Over the last few years, various figures have been put forth as measures of Ubuntu’s desktop market share, with some sources contradicting themselves and most sources contradicting the other ones. Below, we take a look at different attempts to put a number on Ubuntu’s popularity, and explore the various shortcomings and ambiguities involved in all of them.
It goes without saying that there are a lot of difficulties inherent in attempts not just to measure Ubuntu’s market share, but to define what “market share” itself means. For example, are you an Ubuntu user if you install it on an old PC and use it once in a while, or does it need to be your main operating system? What if you dual-boot? Since Ubuntu doesn’t cost money, can it be said to have a share of the operating-system market at all, or does it fit in somewhere else?
What Canonical says
The figure that seems to be most ubiquitous for quantifying Ubuntu’s user base is 8 million. This number dates to a December 2006 interview in which Mark Shuttleworth, Ubuntu’s founder, stated, “We know now that there are probably at least 8 million users.” Unfortunately, he didn’t explain how he arrived at that figure.
Curiously, Shuttleworth revised his estimate in October 2007, when he asserted that Ubuntu had “in excess of 6 million users,” again without outlining the methodology behind the approximation. And while “in excess of 6 million” could be compatible with the claim of “probably at least 8 million” made ten months earlier, that’s a pretty wide margin of error.
To complicate matters, Canonical representative Chris Kenyon again cited the 8 million figure in October 2008, admitting that market share “is a hard thing to count and there are lots of issues about methodology for counting but I have seen nothing that sheds doubts on that.”
Given the variance in popularity estimates put forth by Canonical and the absence of a transparent process for measuring market share, the only certain thing to take away from these numbers is that no one at Canonical really knows with meaningful precision how many people are using Ubuntu–or if they do, they don’t want to make an exact number public.
DistroWatch
Beyond Shuttleworth’s nebulous figures, a second source for estimating Ubuntu’s popularity is the venerable DistroWatch.com. The site’s “Page Hit Ranking” lists Ubuntu as about 1.5 times as popular as Fedora, at least among people who visit DistroWatch. Of course, no one knows exactly how many Fedora installations exist, but the project offers published statistics that can be useful for estimating.
Extrapolating Ubuntu’s market share based on DistroWatch’s numbers and Fedora’s statistics is an exercise left to the reader, however, because I couldn’t figure out any way to do it reliably or meaningfully.
Dell Netbooks
Dell’s sale of netbooks, one-third of which reportedly ship with Ubuntu, also sheds some light on how many people use Ubuntu. Unfortunately, Dell doesn’t say how many netbooks it sells, and critics have argued that many people may buy the Linux version of the Dell Mini to save money, and then replace Ubuntu with Windows when the machine arrives. (I suspect that such customers are a minority, since non-geeks don’t reinstall operating systems; they use whatever comes with their computer.)
Given Dell’s dominance as a PC vendor in most markets, and the fact that netbooks have been a hot item over the last year, Ubuntu’s popularity on Dell Minis would seem to point to a large user base. But it’s difficult to say anything more precise without more information from Dell.
Conclusion
It’s clear from the approaches above that putting a hard number on Ubuntu’s market share is impossible. Canonical seems reluctant to put forward any qualified figure, which is unfortunate–minimally, it would be nice to see an honest attempt at transparent analysis similar to Fedora’s statistics.
In any case, perhaps it doesn’t really matter how many people use Ubuntu, as long as those who do use it like it. Popularity isn’t a measure of quality. But it is a good indicator of progress towards solving bug # 1.
Here is another interesting source:
http://www.google.de/trends?q=ubuntu%2C+fedoraamp;ctab=0amp;geo=allamp;date=allamp;sort=0
Again, it’s not really “market share”, but more of “popularity” which should go along with market share 🙂
Great read. Particularly the link to the Fedora Stats. I think the most relevant number:
TOTAL UNIQUE IPs: 14,205,176
The discussion in Fedora’s seems questionable (to me, more people use dynamic IP’s than they are behind a firewall, and those reset SEVERAL times a year). But let’s assume this is the Fedora user base for a minute.
Here is a very interesting number you may want to add to the post:
http://www.google.com/trends?q=fedora%2Cubuntu%2Ckubuntu%2Credhatamp;ctab=0amp;geo=allamp;date=allamp;sort=0
Kubuntu alone matches Redhat, take them both out, and now Ubuntu is about 6 times more popular than Fedora.
So, here goes my lower bound estimate:
* Ubuntu is 1.5 times as popular as Fedora (DW stats).
* Out of each unique user, there two unique Fedora IP’s
Sorry, my last post got out by mistake, I followup here.
My lower bound estimate is then 7.5 million * 1.5 = 11 million
My upper bound estimate is 15 million * 6 = 90 million.
Note that some times, a unique ID corresponds to several users, like if you have on home PC and several users (family members).
Another twist would be, if the relationship of users is really 6 to 1 (Ubuntu to Fedora), then in order to have around 9 million users we can afford to admit that in the Fedora stats, each unique user gives rise to 10 unique IP’s in the stats. This seems like a very conservative measure. An I really wouldn’t be surprised if Ubuntu had several time more users than Fedora …
Leo:
Follow the links on the Fedora stats page in my blog where I try to measure the affect of NAT (a client undercount effect) versus dynamic ip (a client overcount effect) by trending the Fedora MirrorManager logs against the smolt logs.
http://jspaleta.livejournal.com/42464.html
The take away is from my analysis of 16 months of ip connections is these two effects nearly cancel..and if anything Fedora’s unique stats are slightly undercounting the number of clients because the effect of NAT dominates slightly.
If you have a better idea..I’m all ears. I care about finding a counting methodology that makes sense and that all distributions can use to come up with a reasonable estimate of the amount of linux activity.
Even if the current methodology Fedora is using is not optimal… the simple fact remains, Fedora is being transparent about the data its collecting and giving you and everyone else the ability to scrutinize the methodology when Canonical is not. Canonical isn’t doing that, and that says a lot about Canonical’s commitment to transpart leadership of the Ubuntu community. How the data is being tabulated is far more important than the final number.
-jef
Chris:
You put way to much stock in Dell as a leader in netbooks. Both Asus and Acer are crushing Dell. In fact Acer alone is crushing all the Canonical partner OEMs together.
http://www.displaysearch.com/cps/rde/xchg/displaysearch/hs.xsl/090511_Mini_Note_Penetration_Nears_20_in_Q1_09_Dominated_by_Acer.asp
This report puts Dell at 0.4 million mini-note (aka netbooks) in 2009Q1. And Acer at 1.8 million mini-notes…and Asus at 0.9 million.
By those numbers Dell is selling 30%-40% Ubuntu netbooks that translates into 130 to 160 thousand Ubuntu netbooks in Q1. That’s great…but is it a leader? Not hardly. If Acer is only selling 10% of its netbooks with Linpus pre-installed they sold ~180 thousand of them in the same quarter…outselling Ubuntu on Dell netbooks.
-jef
Leo:
The google trends information is worth mentioning.
No one… not even Google…has validated Google trends as accurately measuring anything.
http://www.google.com/intl/en/trends/about.html
“We hope you find this service interesting and entertaining, but you probably wouldn’t want to write your Ph.D. dissertation based on the information provided by Trends.”
Google Trends as a methodology has not be rigorously validated, and its measurements are not publication quality. If you want to play around with it and have a good laugh comparing Coca-Cola to Pepsi or Toyota to Honda feel free. But you hold the results up as credible at your own peril.
-jef
The other key challenge: Virtualization messes up market share numbers. In the old days, it was relatively easy to track how many operating systems shipped with PCs and servers.
With virtualization, users are constantly adding and removing multiple operating systems from their PC and Mac hardware.
Also, instead of market share it’s becoming more of a time share war. Yes, I use Windows, Mac OS X and Ubuntu. So the three have 33.3 percent market share each in my house.
But if you ask me how much TIME I actually spend in each environment, the numbers shift heavily to Mac OS X and Ubuntu.
[…] final da tarde, eu encontrei este artigo no blog do Juanjo Martínez, contribuidor do Fedora e novamente tive a mesma impressão que tive […]
Regarding the notion that some Dell netbook buyers might buy an Ubuntu-loaded mini and then load Windows, I’d submit that the opposite is well established on PCs in general: Purchased with Windows (because that was ~only option), then loaded with Linux (many choosing Ubuntu). Most users will stick with the initially loaded OS (Linux or Windows), any swapping will be in favor of Linux for (1) historical reasons (only one way to go if ~all were Win at first), and (2) laptops and desktops sold still outnumber netbooks and #1 still applies (Win-gt;Lin is more prevalent than Lin-gt;Win).
What kind of numbers would we see if we based the estimate on the OS usage distribution from web survey sites (1-2% Linux, or as high as 5% if VERY generous), an estimate of the total number of systems in service, and a plausible % Ubuntu of total Linux users? Of course, each of these estimates is rough.
For those who believe Microsoft is always right, you could use their market share number for Linux – around 6%, larger (in Ballmer’s opinion) than Mac OS/X – and work backward based on your perception of Ubuntu’s share of the Linux market.
See http://www.osnews.com/story/21035/Ballmer_Linux_Bigger_Competitor_than_Apple.
I think accurate market share stats could be a distraction. Everyones is happy if it goes up year after year, but what if it stays stagnant or declines? That’s not going to look good and may have negative impact and work against you. If I had a choice of knowing or having a rough estimate, I’ll choose the later. Keep everyone guessing and talking, I say. 🙂
I agree with Zac here. I think when Ubuntu has finally gained a meaningful marketshare, it will be blatantly obvious. Until then, it doesn’t really matter.
[…] Measuring Ubuntu’s Market Share It’s clear from the approaches above that putting a hard number on Ubuntu’s market share is impossible. Canonical seems reluctant to put forward any qualified figure, which is unfortunate–minimally, it would be nice to see an honest attempt at transparent analysis similar to Fedora’s statistics. […]
[…] http://www.workswithu.com/2009/07/24/measuring-ubuntus-market-share/ […]
Analysis paralysis . . . numbers will mean only whatever you want them to mean. All this measurement merely proves theory that may not have been based on fact in the first place. Estimates of estimated data, samples with dubious validity, extrapolating invalid sample data to unrelated populations. There is no way to establish correlation between say a non-internet install base and internet desktop install, or market share and how much people like your software: Take Mac vs PC in terms of market share vs likeability vs ease of use. Yeah I do find the commentary interesting and pretty much everyone is in agreement that its kinda what I used to do in my analysis days, which we used to call intelligent BS. Another thing when it comes to analysis is that, as Warren Buffett says, “In the business world the rear view mirror is always clearer that the windshield.”
Many problems have been pointed out in relying on web user agent statistics from various sources, but if they were done properly, shouldn’t they give an accurate proportion of Linux in web traffic? Granted, that’s not the same as installed base, but it would be a meaningful number. So how about some good web statistics that are not biased towards the US/Europe/technical users/a particular search engine?
FUD is FUD, no matter who does it.
Whether it is flyboy taking numbers out of his butt or the US centric results that were pimped in the media last month which showed Linux at 1% after being in the 2-6% range a few years before.
People use numbers to say what they want. To push their agenda.
And using Distrowatch as ANY kind of measure about how many people use a distro is laughable.
Ive visited the Microsoft site a few times, usually by clicking on a link, it doesnt mean I installed Vista.
Also, I have installed Linux on about 20-25 different computers over the past 2 years for friends, family and coworkers not too mention LUG installfests.
How do you quantify those?
You dont.
Why all the fuss about Ubuntu, there are better distros out there a lot better than ubuntu these that never get a mention It is becoming a ubuntu Linux world, Do any of you ever try other distros? my personal opinion is all commercial backed community distros, Ubuntu, Fedora and Opensuse are being held back by their masters, Take sabayon and Linux Mint for instance, they blow the commercial backed distros out of the water, none come up to the scratch of these two community distributions, they work out of the box, with everything set-up. Has for the number of downloads Sabayon has of today : 9.607,513 Downloads, if only Unbutu, Opensuse and Fedora had a download counter on their web sites at least one would have a general idea of the numbers Linux is used by all distributions.
Fictitious numbers from Ubuntu is like MS stating they have 95% of all world wide users and 98% of business web servers, when we all know different. May I suggest you have a look at the not so well known distribution before giving Ubuntu the golden cup for the best distribution.
Carling:
Fedora has compiled statistics concerning downloads on the Fedora project statistics page from both the bittorrent server. Caveats apply of course. Fedora tries to be reasonably conservative with regard to the exposed stats. If multiple downloads have happened from the same IP address, they are only counted once in the tabulated direct download statistics. There is also no way to get stats on Fedora downloads directly from public mirrors that have no used a redirection URL from the fedoraproject control pages.
I’d be interested in knowing the Sabayon methodology for counting downloads. Do the Sabayon download numbers represent the number of initiated download connections include those attempts that were aborted? Or is that number only completed downloads? Do multiple concurrent connections from the same ip address get counted multiple times?
-jef
[…] Quelle: WorkswithU […]
I’m often frustrated by how Ubuntu is more conservative with some of the packages it ships. By the OS itself, I do believe I’d be happier as a Fedora user.
But that’s irrelevant.
What makes an OS is how easy it is to get apps, support, etc. I use Ubuntu because it’s time for the death of the ‘linux desktop’. There never was any such thing, it only confused developers. Apple and Microsoft make their money by marginalizing us, it’s a pretty effective strategy to reduce customer confusion, and I employ it liberally myself.
What about this?
http://www.statowl.com/operating_system_market_share_by_os_version.php?1=1amp;timeframe=last_6amp;interval=monthamp;chart_id=4amp;fltr_br=amp;fltr_os=amp;fltr_se=amp;fltr_cn=amp;limit%5B%5D=linux
Sorry. This is the working link:
http://www.statowl.com/operating_system_market_share.php
Select Linux share to see detailed statistics for Ubuntu…