Estimating the number of Linux users (the old estimates)

(or: why I think we're eighteen million)

This page will no longer be maintained. It's still here to show you how we've made up the old estimate. Links to proof the old estimate were once valid, like our estimate was in our eyes. It is possible they are no longer valid, like it can be possible that our old estimate isn't valid anymore. The page stays for historical reasons.

It is VERY hard to get any real idea of the size of the Linux community.

In a sense, it is because we want it that way: we do NOT want anything like centralized control, and that means that we don't want anyone to say authoritatively how many people we are, either.

The only number we can be sure of is "many".

The lower limit is something like what is counted in the Linux Counter, but even this is not certain; people may have stopped using Linux after registering, or may be registered under 2 different emails.

The Linux Counter also has several builtin drawbacks:

• Only people who run across mention of it on Linux enthusiast web pages are likely to hear about it, although several important Linux documents also mention it
• Only people with E-mail or Web access can register with it
• And most importantly: Only people who *care* to be registered can be counted.

There are many other pointers to follow:

• Red Hat's Sizing the Linux Market (this link is to the Linux Journal) was probably the estimate that was worked most heavily on at the time; the March 1998 official estimate was 7.5 million users.
  They also claim that approx. 750.000 Linux CDs were sold by US suppliers in 1997 (up from 450.000 in 1996).
• UNIFORUM's article from April 1996 which claims "slow but steady growth" for Linux
• A LinuxGram article from June 2001 quotes Dataquest forecasting 543.778 Linux-based servers shipped in the US market in 2001 - this was regarded as a pessimistic estimate by many.
• A July 2000 article in C|Net News.com quoted IDC as having counted 1.3 million shipments in 1999
• The Netcraft Web server survey noted that in June 2000, there were 6.1 million Linux-powered web sites in their survey, of which 2.1 million were "active".

The most optimistic claim I head came from Jose Daniel Ponz of Spain, who sent me this message in late 1998:

After reading your document on estimates of Linux users, I think there is a piece of information missing: if you do a very simple exercice by searching documents in Altavista using different keywords, you will get the following numbers:

  Keyword     Number of documents
  -------     -------------------
  Linux           1284800
  Windows NT       246985
  Windows 95       331332
  DOS              948860
  Solaris          247140

According to these numbers, Linux has an impact on WWW (and therefore popularity and perhaps users) of the same order as all MS systems together (1527177). This impact is more than 5 times larger than Solaris.

One could argue against these numbers in many ways, but the important point is that Linux is the most popular OS.

These numbers were slightly optimistic then; a slightly expanded search provided the following word counts on November 1, 1998, repeated on Aug 28, 2000:

Word	Nov 1, 1998	Aug 28, 2000
win95	1.668.740	5.915.270
windows 95	3.614.004	7.213.890
win 95	385.886	809.183
win nt	116.302	452.313
winnt	206.630	1.028.093
windows nt	2.574.740	6.411.093
solaris	762.314	2.655.022
linux	3.390.592	43.008.332
dos	7.209.982	24.217.299

Sum for Windows 95: 13.938.343
Sum for Windows NT: 7.891.499

So there is no doubt by now: Linux is the most talked about OS on the Net!
(note: the reason for the increase in mention of DOS is probably the incidence of Denial of Service attacks...)
(Notify me if you try the search and the numbers diverge very much!)

Or consider the following lines, which I did back in August of 1994, when the Linux Counter stood around 10.000 users:

• Back in the OLD days (1993), 1842 people voted on the split of the comp.os.linux newsgroup. This was considered an ENORMOUS number at the time. Let's say that 10% voted at the time, and that the community has expanded by a factor of 10 since then.
• Somewhere before October 1993, the German magazine iX asked 1 % of their readers (260 people) what operating system they used. 29 % used Linux at home! The circulation should be around around 26.000, giving 7.500 Linux users among iX readers in Germany alone. Today, German users are 1018 of the 7347 users (13%); one may guess that 1 out of every 10 Linux user reads iX? Added to the same growth factor of 10, one can multiply the 75 (?) respondents by 7.700, giving 577.500 people.
• One guy asked the CDROM publishers how many Linux CDs they had sold. (around June 1994). The two that answered added up to something like 30.000 CDs. Only 75 of these are among the 545 who have reported to the counter where they installed from (as of Fri Jul 22); also, it is likely that a CD will be used for more than 1 installation, since this is explicitly allowed. On the other hand, some people buy newer CDs after a while. One would then assume that there are at least 6 users who get Linux by other means for each who gets it from CD, giving something like 210.000 users in toto.
• Linux 1.0 dates from March 14, 1994, according to nic.funet.fi file dates. Usage seems to have greatly increased since then.
• The comp.os.linux.* groups are among the busiest groups on Usenet. In the most recent (August 11, 1994) posting, the busiest Linux group was:

     Number Est #   Count  %of readers
     40     140000  1359   2.7%  comp.os.linux.announce
  
         

  (from a quoted Usenet message in news.lists) So let's call the base number 1359, the percentage of Usenet users sampled 1%, and say that 2 out of 10 Linux users on Usenet. That's 679.500 users.
• Some time in 1993 (?), Bill Jolitz of 386BSD claimed that "386bsd has over 30,000 verified , knock on wood, users!". This number was derived from the mail-in registration; at the moment, interest in 386bsd seems somewhat smaller than interest in Linux. A factor of ten, perhaps?
• The October 1994 issue of Linux Journal had a print run of 25.000 issues. If we assume that 1 out of 20 Linuxers read the Linux Journal, we get 500.000 Linux users.
• My news spool, as of August 26, 1994 showed that comp.os.linux.* had 3507 different "From: " lines among the 8402 messages. On August 19, 1995, the same numbers were 10485 articles with 4649 different "From: " lines, as measured by Morten Welinder. If 99 out of 100 Usenet-aware users didn't post this week, and 2 out of 10 Linux users have Usenet access, we might guess a factor of 200 here.

So, one could set up a series of counted numbers, and relate them to the "real number" of Linux users, like this:

Event	Base #	Factor	Estimate
comp.os.linux newsgroup split	1842	100	184.200
iX magazine survey	75	7700	577.500
CDRom publishers	30.000	6	210.000
386BSD relation	30.000	10	300.000
Linux Journal circulation	25.000	20	500.000
comp.os.linux.* postings	3.507	200	701.400
newsgroups readership survey	1359	500	679.500

Of course, the only thing really shown here is that if I am allowed to pick any number, and multiply by any factor I want to, I can get any number I want to get!

Some ways not to estimate

Some measures are lower than others.

One in particular that is low is estimates based on the number of browsers identifying themselves as Linux - for instance, Statmarket identifies Linux as low as 0.21% of browsers (Oct 26, 1999). (It has since gone commercial, so I don't have newer figures). Problem with this kind of measurements is that many sites expect a version of Microsoft Internet Explorer or Netscape before you can use the site. So many people, using other browsers like Galeon, Opera or Konqueror, had to change the user agent settings faking they were using Internet Explorer or Netscape, to enter these sites. Not changing these settings again the Konqueror, Opera and other browsers are then counted as being Microsoft Internet Explorer or Netscape browsers.

One answer is that stats may be confusing - you can never tell what you measure unless you say how you measure it.

Another is simply that a lot of Linux users use a different OS for browsing - a quick peek at the Linux Counter, which is about as diehard Linux as they come, showed that more than 60% of those registering as Linux users used a Windows browser to do so.

One possible conclusion is that the "zealotry" of Linux users is not quite so big as it's claimed..... check out the continuing browser evolution from my Webalizer statistics.

Still not satisfied?

Well - there's only one thing to do.
Get out there and work on getting better numbers!

Please mail any new ways to get estimates to me!

Have fun!