Valid XHTML 1.0!

prefetch enabled page — explanation


 

(last update 27. Jun 2006, final month and long time statistics up to Apr 2006 (finished))

rough (also graphical) long time totals

proceed to statistics and quick facts: half years

proceed to monthly statistics

special OS related (especially Linux) and portable browser statistics

Browsers and Operating Systems — a difficult Relationship

Generally speaking, a user agent string — if present at all — contains not only information regarding the user agent, but also about the OS where it is running. The drawback is, that the OS and the user agent (with the only exception of the most proprietary and standards offending browser in the world on the monopoly OS) have nothing to do with each other regarding that string, which is logged by most webservers, and that the user agent is solely responsible for that string — not too seldom it omits all or most of the useful informations about the OS. Even worse from a statistical viewpoint (but welcome to security and partly privacy reasons) is the fact, that there are proxies masking or faking user agent strings too. This means, that OS designations in user agent strings are nearly worthless, if you care to analyze web statistics thoroughly. Therefore the only mathematical sure implication from these statistics is nowadays trivial: M§ has a monopoly in the browser market and as well in desktop OSs. Well, really no breakthrough in knowledge!

Despite I would love to present my OS statistics too, this very fact makes it useless, so I don't offer this very unclear statistics (the number two browsing OS can't be determined by it, nor the next places for sure). Let us concentrate on the browser market share; because this is much more reliable and complete.

Some more Remarks regarding these Statistics

Other drawbacks on user agent statistics are: robots, which are crawling websites to feed their search engines with data, download utilities, which fetch entire sites or parts of them not discriminating between contents, anonymizing techniques of browsers themselves (fake user agent strings) additional to proxies and so on. The biggest problem are anonymized and fake strings, which can mask user agents mostly of browsers, but also of download tools, but it seems due to the numbers, that it is not twisting the results too much (opposed to the OS case). But keep in mind, that many “Netscape compatible” browsers are in fact anonymized or even faked user agents starting with Mozilla/?, where the question mark holds in most cases a 4 and often is followed by a dot and more digits, even if it is no old Netscape 4 browser.

Prerequisites for considerable Representative Statistics

A website and the log analysis have to fullfill a number of conditions, to get useful material for this purpose:
  • enough hits to get numbers high enough even for more seldom met user agents

  • a broad, not specific range of topics, to avoid effects by special topics attracting primarily special user agents, especially special browsers of not representative groups of internet users

  • a neutral web design, which enables all browsers equally to display the contents without suggesting special browsers and/or versions for viewing at all

  • a really careful log analysis, as I will show, neither analog nor webalizer deliver really useful results without control and support by the human analyzer

Now it should be clear, why most statistics in this area differ from each other and that most of those simply doesn't meet the mentioned preconditions. Luckily enough I have a site (more accidentally then tailored for such a purpose), which is well in accordance with those conditions. Despite there are sites even better suitable for it, I think IMHO opposed to many published statistics I can deliver a useful one, like only a few others… But you may look on my numbers yourself, to judge it.

Tools for Webserver Log Analysis

There are three approaches to tackle this problem used by me: the leading GPL log analyzer analog, the runner-up (also GPL) webalizer and the UNIX/GNU commands and shells, to answer more specific questions (and to control sometimes the analyzers' output).

Part One: Analog as long Time Tendency Indicator

So far I have performed four half-year statistics with analog, all with the essentially same configuration and program version (5.24 for the first three, 5.32 for the last). On the background of a slowly, but steadily rising traffic and a very constant top level domain distribution (I have it included, so judge yourself) the numbers become rather clear regarding this topic. The interval of half an year each seems useful, because it levels the one month fluctuations, which are not big, but notable, virtually completely and delivers enough requests to analyze together, but doesn't ignore the fast moving nature of the Internet. Current remark: to give you also a preliminary impression of the tendencies at now, I have put up after three months elapsed the long time statistics already, which will get updates each month until this half year span is also complete. Either study yourself or read my comments or do both in any order you like, I hope this gives you some insights (one percent at least is required for a user agent to be displayed in the pie chart, all requests but only page requests are used for sorting to suppress the annoying robot activity somewhat):

May 2001 to October 2001
November 2001 to April 2002
May 2002 to October 2002
November 2002 to April 2003
May 2003 to October 2003
November 2003 to April 2004
May 2004 to October 2004
November 2004 to April 2005
May 2005 to October 2005

November 2005 to April 2006

Quick informations about this (for some more complicated issues see below):

  • the monopoly browser has gone from 83.5 % at average through 86.2 % and then hovered at 89–90 %. The end of the monopoly already dawns! The begin of descent was already visible, it decreased to about 87% “only”, then (end October 2004) to just 83% similar to the begin of these statistics end to just 73% until end October 2005. On the other hand the substitution of the 5 and older versions by release 6 is easily visible (some time ago v6 overtook v5), despite on the Mac systems there is still no version 6 available and will never be; officially the support for the Mac IE has also finished with end of 2005 (as on the now terminated UNIX versions for Solaris and HP-UX, which were not used at all quantitatively). And in 2003 Apple presented an own, KHTML based browser called Safari (shares that ingredient with KDE file/web browser Konqueror), which gained already noticeable share (see monthly statistics below) and replaced largely M$IE 5 on the Mac, which is already finished, since Apple released OS X 10.3 with Safari as default browser and took already monopoly on the Mac, even when counting unsupported earlier Mac OS versions (up to 10.1).

  • the monopoly browser of earlier times, Netscape 4 and before, has lost market share heavily due to it's dated and no more developed state (and the illegal behaviour from the Redmond mob): from 9.1 % over 6.7 % and 3.85 % then 2.35 % until just 1.6 %. It is virtually dead now, because it dropped well below 1 % recently: only 0.25 % remaining.

  • the commercial alternative Opera had increased its market share considerably: starting in the observed time at 0.80 %, after a strong rise through 1.16 % and 1.27 % over a peak (see above) of 1.6 % to recently 1.4 % it is visible since the second half year measured in the pie charts. At last it is dropping again from a peak around 2% to just below 1%: 0.9%. Obviously the substitution on the both counting OS Win and GNU/Linux of version 5 to 6 and afterwards 6 to 7 was rapidly taking place and has helped to increase the total share too. Due to the recent release of Opera 7.5x (on Windoze, Linux, Solaris, FreeBSD, and now also Mac OS X) and last of version 8 another push for its share may be expected — but in the end it loses percentage to the Firefox browser as all others (aside from the next mentioned) do. Though since v8.5 its is free to use and 9 brings another couple of new capabilities with it (for example x86 Solaris is supported too), this couldn't turn the tide against Firefox so far.

  • Apples Mac browser Safari, presented as beta in early January 2003 for Mac OS X 10.2 (only), became a stable 1.0 version in summer 2003 and advanced with the release of Mac OS X 10.3 to the default browser of the OS. Even in the first half year report interval available (appeared after the first half of the time span already elapsed) it gained immediately a share of nearly 0.4%, and the second (first complete availability interval) saw it already at more than 0.7% (in October 2003 this was even 1% for the first time). So it seems justified to list this browser already here, because it can gain several per cent of the total browser market, despite it is bound to Mac OS X 10.2 or later. The latest half-year numbers were already slightly over 2%, at now well at 3%, even clearly surpassing the all-platform browser Opera.

  • besides the below discussed Gecko browsers others like the Mac OS X endemic OmniWeb or (UNIX) text browsers like Lynx are playing no role at all quantitatively, despite my site is so very well suited for text browsers and not up to recent w3c standards supporting browsers too (for example I don't use Iframes and with one minor exception no frames too).

Finally the most complex topic in this development (aside from the at last very strong activity of GoogleBot, Googles web crawler) has to be discussed in more detail: the Gecko family of browsers. Firefox has now become the strongest of these, as you can see in the current statistics below. In above statistics five of these are noticable regarding numbers of requests: Netscape 6–8, Mozilla itself, Epiphany and Firefox, now named so, after formerly Mozilla Firebird (the last was originally called Phoenix). Afterwards Camino (formerly Chimera) was added too. In this run analog in the current version counted Mozilla and Firefox together, now does seperately (and Firebird-/fox as another ones too), Galeon is counted seperately and Netscape 6–8 as members of the Netscape family browsers, despite it is much more like Mozilla than like Netscape 4 and the even older ones. Not enough, because Netscape changed the version number again from 6 to 7, these two are counted seperately in the pie chart, falling prone to their split at times in the chart: both were just below the 1 % limit to be displayed, despite they had together about the same share as Opera, enough to be displayed together. In the second chart there was no version 7, so it couldn't take away requests from version 6, therefore Netscape 6 was displayed! Meanwhile Netscape 7 has exceeded the 2 % limit and 6 is clearly below 1 %. Mozilla and the new Firefox together are rising strongly, just at the 2 resp. 10 % limit and the newest numbers show, that they will be now above forever. In numbers that altogether reads like this in percent (remark: I have splitted now Mozilla into itself, Phoenix=Mozilla Firebird and Chimera=Camino due to the recently strongly increased usage):

time\browserNetscape 6–8MozillaFirefox/-bird (Phoenix)Camino (Chimera)GaleonEpiphanysum of all Gecko browsers
200105–2001100.700.13--below limit!-0.85
200111–2002041.200.20-00.04-1.4
200205–2002101.250.620.02below limit!0.07-2.0
200211–2003041.81.20.150.100.0903.2
200305–2003101.91.00.280.050.06below limit!3.3
200311–2004042.31.41.00.040.03below limit!4.9
200405–2004102.752.04.050.040.030.028.9
200411–2005042.21.7110.05below limit!0.0215
200505–2005101.71.4150.050.020.0218
200511–2006041.21.317.50.07below limit!0.0320

That reveals a slow down and even finally reversal in rise of Netscape 6 and 7, but Mozilla had accelerated its fast ascent even more! But even together, this is the strongest rise by more than a factor two seen in all of these browser statistics so far. Recently the Firefox (formerly Firebird, originally Phoenix) browser, a browser-only derivative of the original Mozilla (suite) browser, has considerably gained percentage as you can see: now already slightly over 10 %! (in April 2005 alone these were even already 14%, at midth 2005 already 15% at average). This makes it newly by far the strongest Gecko browser and the strongest contender of the monopoly. On the other hand Netscapes total due to analogs way of summing up is clearly decreasing, because the newer can't compensate for the predecessors decline. Final remark: with rather low numbers I can see too the CompuServe (AOL) Win browser, the AOL OS X browser as well as the Beonex (Win, GNU/Linux, Mac OS X, FreeBSD) and even the K-Meleon (Win only) and new GNOME browser Epiphany, but those Gecko based browsers are not (yet) important at all, therefore not shown (with exception of Epiphany, which is visible since a short time). The others (there are even more Gecko based browsers around, see browser overview) I have never seen in my statistics.

Part Two: Monthly Statistics by Webalizer

These can't be strictly compared to the above analog statistics for the following reasons: in one month notable fluctuations occur especially in the more seldomly used browsers, the categories are of a more global nature, i.e. I count all Gecko derivatives of mozilla.org together (Mozilla, Netscape 6–7, Phoenix/Firebird, Galeon, Chimera=Camino, …) and no subcategories of browsers are counted, so for example Opera 3, 5, 6 and 7 all are counted as Opera without difference.

The tendencies in the market shares of different browsers become more clear either way and you can view already another month, before another half-year interval is completed. Especially when you compare the first and last months available, the decrease of the dated Netscape 4 and before becomes even more impressive as well as the rise especially of the Gecko browser family, but Opera is growing too pretty fast. Important: keep in mind, that Netscape herein contains also Netscape compatible user agent strings, which are no true Netscape 4.x or older. This is even more valid as for the analog statistics above. Changed evaluation: due to the decline in the old Netscape I decided with begin of the fourth long time statistics interval (since Nov 2002) to seperate browsers not listed as special ones, but claiming to be Netscape compatible, from the true (?) Netscape browsers — so it is better comparable to analog and reveals the rapidly decreasing share of old Netscape more clearly, because it is not hazed by others, which are more constant in share. Surprising news: the coming out of the new browser Safari, Apples own one based on KTHML like Konqueror (KDE browser, mostly met on GNU/Linux therefore), had a major impact on these statistics, as you can already witness even in the incomplete current month statistics! (despite it appeared only on 7. Jan 2003) It seems, that Safari will replace the dated M§IE on Mac OS X 10.2 and higher rather quickly, grapping a total of 0.5% instantaneously as beta… For the interested: here are the entries relevant for the mainstream browsers and for some other user agents, which have to be appended thereafter, to get the statistics right: the order is always extremely important in the webalizer configuration file! Now watch yourself:

Month (YYYYMM):

Hint: in January 2003 I witnessed a strong peak of hits on the German no TCPA page, which increased the percentages of Opera and the Gecko based browsers above the average tendency. So you have not to worry about that (this special visitor traffic with a less than average monopoly structure has considerably decreased again, is now less important, so it looked like Opera and Mozilla would have lost share compared February to January 2003).

(German flag An die Heise-Besucher: bezeichnend ist, dass fast ausschließlich Leute an diesen Statistiken interessiert sind, die ein Gecko-Derivat oder Opera verwenden, wie ich am Morgen des 19. Dezember 2002 fest gestellt habe… M§IE kommt da so gut wie nicht vor, was u.a. meine Meinung über die nicht repräsentative Zusammensetzung dieser Leute hinsichtlich Browser bestätigt, auch GNU/Linux ist überdurchschnittlich vertreten)

Even more Special Statistics

Finally you may look additionally at these three special statistics of my own (also updated 9. Sep 2005, including the same time intervals):

GNU/Linux and Solaris browser statistics
“speculative” Mac OS statistics
Portable Browser statistics

Related Links of Interest

Upsdells selected example browser statistics
Googles Zeitgeist: some more statistics (changes often; sometimes with browser statistics)

 

browser resources  browser overview  GNU/Linux browsers  Mac OS X browsers  back to main

remarks, questions etc. to: stefan.urbat@apastron.lb.shuttle.de

(URL:  http://www.lb.shuttle.de/apastron/browsWar.htm)