Talk:Usage share of operating systems/Archive 6

Unusually high rank (highest in some countries) of Linux vs. Android on tablets
, I notice that Linux has ranked high for years now: http://gs.statcounter.com/#tablet-os-ww-monthly-201302-201509

[Windows is ranked 3rd, down in the noise after this mystery Linux.. I've never seen these high Linux numbers on say desktop, these are the highest Linux numbers, except for servers etc.]

Most pronounced in North America: http://gs.statcounter.com/#tablet-os-na-monthly-201302-201509

In fact Linux is a sizable fraction of Android (there), and I wander if it is Android, a fork of it, say Kindle Fire (yes, in the US, but find that unlikely for Africa..)? Or CyanogenMod etc. and just doesn't register as Android, but should be added to it? Could be both.. or either in different countries:

The increase in this Linux trend is pronounced in Africa, very recently and in fact:

In Senegal and Gabon (the only countries), Linux now ranks higher than Android (and thus highest). See here for older theory Talk:Linux comp.arch (talk) 09:26, 15 September 2015 (UTC)


 * That's quite interesting. Maybe we could ask the folks behind gs.statcounter.com what do they actually count in as "Linux" on tablets? &mdash; Dsimic (talk &#124; contribs) 09:45, 15 September 2015 (UTC)


 * I am thinking maybe OLPC identifies itself as x11 Linux tablet ? The other options are Cube i7 with Ubuntu, from China, or Tizen TV from Samsung. The fluctuations tell me not many people are using tablets in African countries, just like this 100% Desktop Linux spike in Antartica. --Ne0 (talk) 07:23, 20 September 2015 (UTC)


 * Thanks for the possibilities.. I had overlooked OLPC, as it is a laptop.. but they also had some tablet. I still doubt it is it, as I think not very successful.. I admit I was a little surprised by the African launch (was it first there..?) of $22,999 big screen, curved 88-inch [Tizen TV] screen and the African Cinema Mode. I guess it's a tablet :) it certainly isn't the usual desktop. I'm going with fluctuation until we see some more data and this Linux trend not going down. comp.arch (talk) 00:33, 21 September 2015 (UTC)
 * With combined Tablet and Mobile stats, GNU/Linux is sustaining much better in Gabon. GNU/Linux on Tablet & mobile in the continent of Africa is hovering at 4 to 5% for two months. So, I guess this is not just a fluctuation. The bigger fish is the 'Unknown' mobile OS in Guinea reaching 49% of all OS. --Ne0 (talk) 11:02, 28 September 2015 (UTC)

Use of Net Applications
Unfortunately, although it is widely-cited, Net Applications is (in my opinion,) an unreliable source, and should not be cited in this article, or any other including market share data, for the same reason that video game articles do not accept any use of VG Chartz.

This organization has Apple and Microsoft as "clients" (thus meaning that it has a presumed conflict of interest towards these firms), and has a history of having used skewed and questionable methodologies for obtaining its data. For this reason, I would like to request that Net Applications data be banned from use on Wikipedia. ViperSnake151  Talk  01:40, 4 October 2015 (UTC)
 * NetMarketshare adds weight to it´s stats, from traffic flowing through USA, monitored by CIA (NSA/CSS PRISM). In addition, NetMarketshare dosen´t seem to count Chrome OS in it´s statistics. So, I guess Netmarketshare doesn´t really pass RS. --Ne0 (talk) 18:32, 13 October 2015 (UTC)
 * It add weights to traffic data from different countries because the majority of the websites it uses for data are from the US, which skews worldwide results. StatCounter's results are actually misleading because, since they do no weighting, the results are biased towards the US and India, which together make up half of the traffic data.(citation)
 * I'm confused as why you'd mention PRISM, because that's not what it says at all. It takes its raw statistics from code placed on websites who use their analytics platform, and weighting information from the CIA World Factbook, which is available for anyone to read.
 * ChromeOS would be most likely counted as either Linux or Other. That they don't break it out doesn't mean they don't track them. -MarkKB (talk) 01:08, 26 November 2015 (UTC)
 * Direct quote from NetMarketShare: "The Net Market Share data is weighted by country. We compare our traffic to the CIA Internet Traffic by Country table, and weight our data accordingly."
 * NetMarketShare uses what it calls CIA data, I call it NSA/CSS PRISM data. Note that, CIA internet activity & Online Intelligence gathering has been merged with NSA, to create NSA/CSS, led by NSA. Hence all CIA internet data belongs to NSA. --Ne0 (talk) 11:06, 5 January 2016 (UTC)


 * ViperSnake151: That's an article from before NetApplications weighted their results, when they reported raw numbers. It describes the exact reason that led them to start weighting, and the very same problem that StatCounter continues to have. If that counts as a "history of skewed/questionable methodologies", then StatCounter certainly has one too.
 * As for "conflict of interest", Microsoft and Apple pay for their reports AFAIK. Them paying for NPD and Gartner reports doesn't stop us using those statistics. -MarkKB (talk) 01:08, 26 November 2015 (UTC)

Why no coverage here for server OS market share?
I keep hearing that Microsoft is growing faster than Unix/Linux combined, and that Linux is growing at the expense of Unix, but not taking share from Microsoft.

It was puzzling to not see anything here -- seems I came here about a year ago and the server OS market was covered then -- was it removed?

50.148.209.148 (talk) 19:15, 15 December 2015 (UTC)


 * There is the "Public servers on the Internet"-section, and it's easy enough to see into the past with "View history" (then it just didn't say "Public" but should have, and that is what is easy to get numbers for). There is also "servers" for web near the bottom, but older data, and should be updated (not sure if it is the same data). comp.arch (talk) 18:04, 8 January 2016 (UTC)

2016 Update
This whole article needs to be updated as soon as possible. Windows 10 is showing as only a few percents. With the automatic update, this should be significantly higher. Not representative of today's numbers for sure. — Preceding unsigned comment added by 198.103.184.76 (talk) 14:32, 14 January 2016 (UTC)

Android desktops
Yes, Android desktops are here, & I have two of them: Remix mini, & Shield TV(+launcher). We should create a new article for Android desktops. --Ne0 (talk) 11:52, 19 January 2016 (UTC)

Categories
The categories are not precise.

Category suggestion #1: Mobile/handheld/wearable

For the definition of mobile, look at https://en.wikipedia.org/wiki/Mobile_device#Types

Now that phone screens are up to 6" and mini tablets (iPad Mini) are 7.9", there is not much difference between the two. Same OS, same parts - just less than 2" bigger screen. How about collecting all tablets, phones under mobile/handheld devices.

There are lost of these devices and many of them are dominated by Linux or Linux is a big player. Phones, tablets, smart watches and other wearable, e-book readers, navigators.

Category suggestion #2: Embedded

The article could also mention development boards and other consumer/hobbyist embedded devices or embedded in general. Raspberry Pi and so on. Smart TVs often run Linux.

The other categories seem ok - desktop, server, mainframe, supercomputers, web (server/client).

2001:2003:F638:A000:0:0:0:32F (talk) 03:49, 20 January 2016 (UTC)


 * I try to just go with the sources, but some include tablets with smartphones (that are called "mobile"), while others include with desktops or as a separate category. That a phone has telephony additions, but otherwise everything else in common with tablets, isn't a defining thing – to me – as an operating system, but since sources say so, I go with it. That wearables/smartwatches (or TVs), happen to run say Linux or Android, doesn't mean that apps are compatible to or from. I think we'll just worry about this, when sources start counting these as categories.. I think, rightfully, they will not include all wearables into the same single category. comp.arch (talk) 14:05, 23 January 2016 (UTC)

What desktop operating system has quadrupled share from 1% about a year ago? [Highest in Asia at 5.73%, but could be more than one OS..]
http://gs.statcounter.com/#desktop-os-ww-monthly-201503-201602

Unknown is most popular in Asia then Africa, South America. Elsewhere, such as North America, I only saw "Other" high, but not this high.

"Unknown" is a mystery (I've already asked StatCounter, waiting for an answer..). In Feb 2015, six versions of Windows are shown in legend, plus OS X, since then one less Windows plus "Unknown". CVS-file already shows Linux and Chrome OS (and Other), so at would be an addition to that making Linux kernel-based 1/20 combined.. if: I'm assuming, this can't be any kind of Window or OS X (all known versions must be accounted for..), could it be Android desktops/laptops (Remix OS or other Android-derivative?)? CSV-file doesn't show Android.. maybe it's assumed that it can't fall in the desktop category? Or not all Linux versions are counted? Could it be something else? I doubt FreeBSD or Solaris this high.. comp.arch (talk) 14:35, 12 February 2016 (UTC)

Update: In India "Unknown" is up to 10% on mobile, of course not that high, still the second most popular OS.. On desktop it's almost 5%, but could be a different "Unknown".. I'm guessing this is the same, some kind of Android derivative, not yet counted. I'm assuming not Fire OS, as I'm sure it's not too popular in India, and also too known to not be counted (and not available on desktops). comp.arch (talk) 10:49, 16 February 2016 (UTC)


 * "Unknown" are derivatives of Android, listed as such as "a consequence of our Android device detection system. It requires frequent updates to keep on top of new models, but the volume required has been increasing. At the moment when a mobile device's vendor and model is unknown we also record the operating system as unknown with a default 'desktop' platform. Looking at the raw useragent strings for these hits, they are indeed predominantly Android useragents that haven't yet been accounted for in updates to our handset detection.


 * We're setting in place a more intensive and frequent update cycle for our handset detection, which should in the near future see a reduction in the 'Unknown' percentages." comp.arch (talk) 19:28, 20 February 2016 (UTC)
 * My Shield TV registers as unknown --Ne0 (talk) 14:37, 7 March 2016 (UTC)
 * How do you know? Maybe they register it as Android? Maybe they register it as mobile? You do not know their algorithm, that will hopefully be fixed soon. comp.arch (talk) 08:32, 8 March 2016 (UTC)
 * Shield TV(Android TV 5.1) registers as, Desktop, Unknown 5.1
 * http://gs.statcounter.com/detect -shows what Statcounter recognizes your current device as --Ne0 (talk) 21:51, 10 March 2016 (UTC)

Chrome OS and Linux
Is it necessary to separate Chrome OS and Linux operating systems? By definition, Chrome OS is a Linux operating system. If so, would it be better to say "Linux (excluding Chrome OS)"? Sega31098 (talk) 02:28, 20 January 2016 (UTC)
 * Linux as an engine refers to the Linux Kernel. Linux as a full system usually refers to GNU/Linux. Chrome OS, RHEL, & SLES are Enterprise Linux, NOT GNU/Linux. --Ne0 (talk) 12:21, 20 January 2016 (UTC)
 * Yes, Chrome OS (and Android) need to be separate from other Linux. What is key, is that different operating systems, run different applications. Android is pretty much separate from traditional Linux (e.g. GNU/Linux desktop or server type). Chrome OS is also different as that only runs web apps. That is happens to run on the Linux kernel doesn't matter much. You can include these (all) under a Linux-kernel based. It may or may not be too helpful.. In the long run, Chrome has been said it will merge with Android. "Enterprise" status does not enter into all this in my view and isn't contradictory with GNU/Linux in the cases mentioned above (except for Chrome OS, what has little to do with GNU). comp.arch (talk) 13:57, 23 January 2016 (UTC)

Some of the data on this page seems a bit biased in favor of some enterprises. GNU/Linux (see also GNU/Linux naming controversy) is Linux-based OS that is composed entirely of free software, most of which is under the GNU General Public License. Red Hat Enterprise Linux, SUSE Linux Enterprise Desktop and other commercial variants use GNU userland and most of them are licensed under the GNU General Public License and various other free software licenses. That's why they're also obligated to freely provide their source code. Most of these OS's are based on the open-source model, including the Android OS. The biggest difference between GNU/Linux, as you like to differentiate and call it, and the Enterprise Linux is the commercial support and consulting services. Aside from the commercial support, you can run most of the available software on both. --Ic1101ic (talk) 09:11, 7 April 2016 (UTC)

Wrong information on google
This page currently makes google say OSX is the most used operating system when given the query: Most used OS. Which I think is caused by some clever combining of OS's in that first graph to give a wrong impression about what the real most used OS is, which currently is windows 7. I hope someone who is more wikipedia savvy can change this. CoatThese (talk) 22:24, 17 April 2016 (UTC)
 * Nope, the most used OS is Android, worldwide as of December 2015. Most used Desktop OS is Windows 7 --Ne0 (talk) 20:50, 20 April 2016 (UTC)

"Unknown" is most popular OS in more than one country
Since late last year, "Unknown" has outranked Android (and others) in Madagascar. If "Unknown" is added to Android, which I think it is, it it makes for more share than all OSes combined.

Papua New Guinea has also just shown "Unknown" most popular. [French Guinea (in South America) is however with Windows 10 highest ranked, seems the first (only now) country in that continent.]

If only desktop OSes are considered "Unknown" is still on the radar, say in India. It's probably then also Android. comp.arch (talk) 16:19, 5 June 2016 (UTC)
 * I think most Android game consoles, Android set-top boxes, Android HDMI dongles, and Android Laptops are being recorded as "Unknown." (Both my Android boxes show up as "unknown", Shield TV as "desktop unknown") --Ne0 (talk) 10:42, 10 June 2016 (UTC)


 * Note that this hasn't been true in the past; see . So I would avoid jumping to conclusions: "Unknown" really should be treated as "unknown" in this case. --DanielPharos (talk) 22:02, 17 June 2016 (UTC)

#Worldwide device shipments - 2015/Windows
Why in the table (Usage share of operating systems), in the 2015/Windows column, does it say "thereof OS X is 21 million"? --Diblidabliduu (talk) 16:47, 21 June 2016 (UTC)
 * I can't recall, if it's the same underlying Garner source or some other. I can't locate now. "iOS/OS X" isn't one operating system, but what the source given gave me. comp.arch (talk) 15:58, 26 August 2016 (UTC)

What is Statcounter's "Unknown" category?
"As of April 2016, online usage of Unix-like kernel derivatives (Apple systems + Google systems + GNU/Linux) is now more than Windows online usage." When I crunch the numbers from the CSV file, I get a difference between Windows and Unix-like that's less than 3 percent, while the "Unknown" category is more than 3 percent. If we treat Statcounter's "Unknown" category as unknown, this means the difference is statistically insignificant, and thus the statement is wrong. In other words, there's a claim here that the "Unknown" category doesn't include Windows, or is mostly Unix-like. I cannot find a breakdown of the "Unknown" category on Statcounter's website, so can anybody here show me where the proof of this claim is? --DanielPharos (talk) 08:31, 19 June 2016 (UTC)
 * Using the CSV for April 2016, Win7+Win10+Win8.1+WinXP+Win8+WinVista+Windows Phone+Win2003+Win8.1 RT+Xbox+Win2000 = 46.44%. Android+iOS+OS X+Linux+Chrome OS+Samsung+Playstation+Tizen+LG+MeeGo+Bada = 47.85% (If you just include the first 4, it's still 47.1%). You don't need to include Unknown or Other for Unix-like's to outnumber Windows. Yes, it's a small increase (although I would argue still statistically significant) but comparing these numbers to other months, you get 47.32% vs 47.15% in March and 44.3% vs 49.74% in May. The May numbers are definitely statistically significant, and April remains the month where Unix-likes overtook Windows, at least via this metric. We can't make any assumptions about "Unknown", but the difference in the May numbers is 5.44%, and "Unknown" was reported as 3.35%, so it seems your concern is resolved anyway. --Skrapion (talk) 18:14, 19 June 2016 (UTC)
 * But let's say that "Unknown" is comprised mostly of Windows (through some browser that doesn't include the OS in the user agent). Then 46.44%+3.35% > 47.85%, and thus Windows is still larger than Unix-like. So you're making assumptions about "Unknown" not being Windows, even though you go on and explicitly state the opposite. In other words: the difference is NOT statistically significant, because you have to treat the "Unknown" as a measurement error. And thus the statement in the article is wrong: Statcounter showed no such thing in April 2016. --DanielPharos (talk) 05:58, 20 June 2016 (UTC)
 * "But let's say that "Unknown" is comprised mostly of Windows (through some browser that doesn't include the OS in the user agent)." Isn't that unlikely? Any browser you know of? All known are accounted for, and the numbers are high. We need to get an answer from StatCounter soon.. better yet, they just correct numbers.. comp.arch (talk) 16:21, 26 August 2016 (UTC)
 * Since we're talking about single-digit percentages here, something "being unlikely" isn't good enough. Browser: I already mentioned one down below: Lynx. And yes, I hope StatCounter can elaborate what their "Unknown" category is, exactly. Or at the very least how they've defined it. --DanielPharos (talk) 16:31, 26 August 2016 (UTC)
 * And I just looked at the W3Counter stats you just added (I can only find the Top 10 Platforms to work with?): the "Unknown" category I calculated to be 8.33%, with the difference between Windows and Unix-like to be less than 2 percent. Once again: not statistically significant, unless we start making assumptions about the "Unknown" category! --DanielPharos (talk) 06:03, 20 June 2016 (UTC)
 * Typically, the best assumption you can make about measurement errors is that they affect all measurements equally. If the current measurements still had the potential ambiguity you're worried about, I'd see your point, but we're not arguing over whether or not these sources show that Unix-likes have overtaken Windows, we're only arguing about when it happened. We're just looking for the crossover point. Would you really prefer 'According to StatCounter, Unix-likes probably overtook Windows in April, and definitely overtook Windows in May"? --Skrapion (talk) 04:52, 21 June 2016 (UTC)
 * "If the current measurements still had the potential ambiguity you're worried about" Nobody has showed me otherwise. "We're just looking for the crossover point." Which we probably cannot ever pin down to a single month (or even quarter), with such a large "Unknown" category.
 * But indeed, enough with the arguing, time for fixing! I'm 100% okay with your suggestion, although I'm worried it would be flagged for weasel wording. Maybe something along the lines of 'According to StatCounter, Unix-likes overtook Windows in during the period of April-May 2016'? --DanielPharos (talk) 05:54, 21 June 2016 (UTC)

I would say most of the 3% unknown is among the "others" being shipped, like the 520 million "others" shipped last year. Possibly includes Linux devices like Solus, C.H.I.P, Kano, Parallella, etc. Non-mobile Android devices are also sometimes counted as unknown. --Ne0 (talk) 05:28, 14 July 2016 (UTC)
 * I'm not sure what these Gartner's numbers have to do with this? Can you elaborate? --DanielPharos (talk) 08:24, 14 July 2016 (UTC)
 * According to Gartner, around 520 million devices with "other" operating systems were shipped last year. It is very likely that most of the unknowns in StatCounter are among these "other" Operating Systems, and NOT Apple or Microsoft systems. --Ne0 (talk) 09:56, 15 July 2016 (UTC)
 * I still don't see a direct correlation. Devices shipped does not equate to web clients, as the first includes devices that are never online. Additionally, the "unknown" category may include Windows/OSX/Linux devices that cannot be identified as such (for example by spoofing the user agent header), which explicitly don't fall into the "Other" Gartner category. I understand what you are trying to insinuate, but from the sources provided I cannot draw your conclusion. --DanielPharos (talk) 13:38, 15 July 2016 (UTC)
 * You may be able to spoof browser agent, but you cannot spoof device shipment: a Chromebook is a Chromebook, a Mintbox is a MintBox ! --Ne0 (talk) 10:50, 16 August 2016 (UTC)
 * Yes, but that's not what we were talking about, because StatCounter doesn't measure device shipments, but webclients. --DanielPharos (talk) 15:29, 16 August 2016 (UTC)
 * Yes, StatCounter gives statistics on webclients, and device shipments translate into device usage, and ultimately counts in the webclients. --Ne0 (talk) 05:08, 23 August 2016 (UTC)
 * I'll just quote myself as my reply: "I still don't see a direct correlation. Devices shipped does not equate to web clients, as the first includes devices that are never online. Additionally, the "unknown" category may include Windows/OSX/Linux devices that cannot be identified as such (for example by spoofing the user agent header), which explicitly don't fall into the "Other" Gartner category. I understand what you are trying to insinuate, but from the sources provided I cannot draw your conclusion." --DanielPharos (talk) 05:49, 23 August 2016 (UTC)
 * "Devices shipped does not equate to web clients, as the first includes devices that are never online." ...Then tell me, if those devices don´t eventually end up in customer hands, what happens to them ? do the brand new devices go straight to landfill ? --Ne0 (talk) 05:27, 24 August 2016 (UTC)
 * Not every device in customer hands becomes a web client. Servers often don't. Neither do some low-power devices, such as Arduino's and Raspberry Pi's. Or NASes. I'm not familiar with them, but I doubt Steam machines are counted by StatCounter, unless Valve's websites happen to be among the websites gathering statistics. What about most other consoles? They ship with operating systems too, so they are counted in the first category, but may not be in the second. You haven't shown whether or not the correlation between the two categories is strong enough to make the claims that you do.
 * Now that I think about it, people can also install their own operating systems. I think quite a few Windows-shipped machines end up being Linuxes. Probably not a significant effect, sure, but again, you haven't shown this to be insignificant.
 * You haven't shown that "user agent spoofing" is insignificant either.
 * In other words: "from the sources provided I cannot draw your conclusion."
 * However, let's stop arguing in circles, and get back to the original point: the StatCounter "Unknown" category. Can you show what that category is comprised of? (Showing with sources instead of merely suggesting, as that is not only not good enough, it's probably OR.) --DanielPharos (talk) 06:03, 24 August 2016 (UTC)
 * Both my non-mobile Android devices, SHIELD Android TV and Remix Mini show up as unknown on StatCounter. I would guess StatCounter unknowns include Android game consoles, Android mini-PC, GNU/Linux smartphones, and other obscure systems, such as non-branded OS like OpenBSD, FreeDOS, etc. We need more proof/confirmation on this. For example, at least 10% of computers in India are sold with some form of DOS, and DOS skills are highly valued in India, like Linux skills. But StatCounter does not show any DOS usage in India. My guess would be, it is counted as unknown. Also, I had used Windows for 15 years, and have never seen a browser on Windows not being recognized as Windows. Please do give examples, if you have any. --Ne0 (talk) 10:30, 24 August 2016 (UTC)
 * "I doubt Steam machines are counted by StatCounter, unless Valve's websites happen to be among the websites gathering statistics", you have this backwards. Clients, not servers are counted. Consoles, are in the noise, however, as I guess most are just playing games.. not online.. comp.arch (talk) 16:30, 26 August 2016 (UTC)
 * Sorry if I wasn't clear: I meant the Steam consoles; I never meant any servers. The Steam consoles are Linux-based, but their web-activity is mostly restricted to Valve's websites (i.e. Steam). In other words, they are shipped devices, but unlikely to show up in StatCounter's "unknown" category. (That is, unless StatCounter happens to be monitoring the Valve website.) --DanielPharos (talk) 16:33, 26 August 2016 (UTC)
 * Sorry, they do track "client websites using Javascript in the user's browser" (private communication) comp.arch (talk) 17:36, 26 August 2016 (UTC)
 * Not sure what you mean by that. What I was talking about is this: If Valve is using this on the Steam website, Steam OS would show up (probably as Linux). If Valve isn't doing that, most Steam consoles would probably not show up. --DanielPharos (talk) 18:11, 26 August 2016 (UTC)
 * I think you misunderstand. If Steam OS machines have web browsers, then they are counted on other sites. StatCounter seems to provide a service, so you can track who comes to your site. That is probably the same system they use to track all web clients, but Valve wouldn't have to set it up for the servers to make SteamOS count at StatCounter. comp.arch (talk) 19:35, 28 August 2016 (UTC)
 * But Steam *is* a browser,; it's a Webkit-based engine. Now, admittedly I'm not familiar with the Steam OS implementation, but I highly suspect it's the same: a webbrowser (plus some other features surrounding it) that is displaying the steampowered.com website. In other words: the main store interface is "just" a website, and thus all Steam OS devices are web clients. But they won't be counted in StatCounter at all, unless they visit StatCounter-enabled sites. And I doubt steampowered.com is one. Thus, most Steam console are not even counted by StatCounter. This is what I have been trying to say all along. Please correct me if I'm wrong with my assumptions about Steam OS! --DanielPharos (talk) 05:51, 29 August 2016 (UTC)
 * "that is displaying the steampowered.com website", not only, as I confirmed looking up on youtube. And on the WP page: "based on estimations made by the tracking website Steam Spy" [probably unrelated], "access a built-in web browser without having to exit the game" [yes, most will not browser much [arbitrary websites], also known that mobile apps are more used than the web on smartphones, so web use as a proxy for all use of a device is flawed] "users could share screenshots on websites of Facebook" comp.arch (talk) 08:42, 29 August 2016 (UTC)
 * So, conclusion: it *can* be a web client, but "most will not browse[r] much". And thus we once again cannot tell what percentage of shipped devices will show up as web clients, nor can be deduce how many Steam consoles are in the "Unknown" or Linux categories, without getting more concrete numbers from StatCounter. --DanielPharos (talk) 14:20, 29 August 2016 (UTC)
 * "cannot tell what percentage of shipped devices will show up as web clients" I think StatCounter could be counting Steam OS just as well as other consoles, or non-consoles. I just can't tell. The objective of this page, is to count "usage share", not just "web usage share". Always keep that in mind. They are not the same, but StatCounter could be fair for the latter. It could also be missing out on SteamOS, but at least they count other consoles. Keep in mind that the most popular console, PlayStation (at least what they count, could be all or one version) is at 0.09% and all consoles total at 0.15%. I doubt SteamOS is is more popular than PlayStation (or all combined), so it would at best account only for a tiny fraction of "Unknown". I guess they are all counted, just not classified right. comp.arch (talk) 12:16, 31 August 2016 (UTC)
 * "I think StatCounter could be counting" I have never denied that. In fact, I've explicitly said this. I was only questioning out loud whether it's likely for Steam consoles to show up in the StatCounter numbers. "The objective of this page" I'm glad we agree on that. "Always keep that in mind." I have never done otherwise. "so it would at best account only for a tiny fraction" ok, so it's probably not Steam consoles. Great! Glad we can cross another one off the list. Glad we agree on that (even if we came to that conclusion using different reasons). --DanielPharos (talk) 15:28, 31 August 2016 (UTC)

"Both my non-mobile Android devices" How are you determining where a device ends up, and is there a way to do this without violating OR?

"with some form of DOS" Nice, I didn't know DOS was still so popular! Any indication of how many of those DOS-devices are web-clients?

"Also, I had used Windows for 15 years, and have never seen a browser on Windows not being recognized as Windows. Please do give examples, if you have any." I've already mentioned one way: user agent spoofing. Most modern browsers have this functionality build-in. Better, take Tor; it's lying by default: http://tor.stackexchange.com/questions/4890/tor-browser-user-agent-strings

But there's more. Here's the default Python 2.6 user agent: "Python-urllib/2.6". I don't see any OS tokens in there.

Here's pyCURL: http://www.useragentstring.com/pages/useragentstring.php?name=PycURL

libwww-perl: http://www.useragentstring.com/pages/useragentstring.php?name=libwww-perl

Many of these would be scripts, but don't those count as "web clients" too? And some of them are legitimate web clients; it's just that the program displaying it to the user is going through a Perl/Python/... library. These will probably all be counted as "Unknown"s.

More browsers you say? Here's Lynx: http://www.useragentstring.com/pages/useragentstring.php?name=Lynx And I suspect the new Cliqz browser that's being developed doesn't send any OS-token either. Sure, not many people are using Lynx or Cliqz right now, but we're already comparing single-digit percentage numbers, so "not many" isn't good enough an argument.

If Linuxes are ending up in the "Unknown" category, that's *more* cause for concern, because it introduces the concept of an error on the values of all categories given! You are literally saying that the "Linux" category numbers are flat-out wrong (probably close, but wrong nonetheless). Can the "Windows" category be trusted? Without some constraints on the size of this error, how can we trust any of the numbers at all?

Without any concrete breakdown numbers, it'll be very difficult to drop the percentage "Unknown"s to below the difference between Windows and Unix-like for these months, if that's even possible at all. I don't see how those breakdown numbers can be reliably obtained without StatCounter simply providing them, and I think they don't actually have them, hence them calling them all "Unknown". As soon as you start analyzing the StatCounter numbers, trying to get more out of them, all kinds of measure errors and secondary effects (like the error on the values of all the categories, or the fact that spammers often spoof their user agents; how are those handled?) will need to be estimated and taken into account. That's why it's best to leave the "Unknown" category just that, unknown. This is why I've been taken the "numbers at face value" approach from the beginning and formulating it as "according to StatCounter", because I really don't want to open that can of worms any further. --DanielPharos (talk) 16:28, 24 August 2016 (UTC)


 * How are you determining where a device ends up : gs.StatCounter.com/detect - and yes, it is OR because you yourself are going there, not a reporter/blogger.


 * Statcounter on Bots/Scripts : We make every effort to eliminate bot activity. We identify bots and prevent them from being recorded in our stats. We also monitor multiple key metrics in order to identify any potential problems with our data. We maintain communications with many of the main technology providers and we reserve a 45 day window to revise stats if/when necessary. In the event of a change or revision being required we flag this on our graphs by way of a note.


 * DOS: the only organization I have personally seen using DOS is Indian Railways, one of the World's largest employers. At any rate, a computer running MS DOS as a native OS should be even less powerful than today's "Cloud PC"(Chromebooks) ! Since the 'official' DOS (from Microsoft) is so limited, vendors of current generation of PCs tend to use some other form of the original DR-DOS, such as FreeDOS.


 * Unknown browsers : As of last month, StatCounter has identified 99.9% of the browsers - there's a 0.07% unknowns and 0.16 % others. So, most of the 4% unknowns are NOT unknown browser !


 * "Can the 'Windows' category be trusted?": All those "Surface Pro tablets" sold, and not a single blimp on StatCounter ? - This is conjecture but, Windows 10 Intel x86 tablets seems to be counted in the desktop category... so the "Windows 10 desktop" category may be over-bloated with Windows 10 Intel mobile devices; just like unknown category seems to be over-bloated with Android desktop devices. This conjecture comes from Gartner and other's statments that Desktop sales are falling, while Mobile(including 2-in-1) sales are rising, for every OS. So, where is Windows NON-RT(non-ARM) mobile devices in StatCounter ?
 * There is also React OS, which is built to imitate Windows ! --Ne0 (talk) 04:41, 26 August 2016 (UTC)


 * Statistics should always be taken with a pinch of salt, as there is always a chance for error, in this case the 4% unknowns could belong to any OS, but probably belongs to the more obscure systems not counted in StatCounter. Hackers and Darknet users are known to change User Agent, and also they usually use BSD or Linux derivatives. Also, the population surveyed(online users) may actually be a narrower sample than previously expected: if atleast 50% of the 1.4 million Indian railways employees are using DOS on IntraNet, then you got 700,000 DOS Desktop PC not counted, since they are NOT "Online" -- Just because majority of the population you surveyed says 2 + 2 = 3, you still have to report the answer as 3 ! Remember, not everyone is connected to the Internet, maybe even the majority of the computing devices are not connected to the Internet... we will never know ! --Ne0 (talk) 09:58, 26 August 2016 (UTC)


 * "but we're already comparing single-digit percentage numbers", in India, "Unknown" is currently at 11.32% (plus "Nokia Unknown" at 5.14%), so that is a clue (I'm guessing not Tor): cheap no-name Androids, since India (and Madagaskar highest ranked for almost a year at up to 36%..) and more is so disproportional with the developed world. I've looked into this before (now in [|archive 6]). comp.arch (talk) 16:59, 26 August 2016 (UTC)
 * It's all circumstantial at best. Without concrete numbers, how can be draw solid conclusions? As I've said before: "from the sources provided I cannot draw your conclusion." --DanielPharos (talk) 17:03, 26 August 2016 (UTC)


 * "Statcounter on Bots/Scripts": So no concrete numbers then? No indication if it's sub-single-digit percentages?
 * "DOS": So no indication how large a fraction of those are web clients, only that it's "probably small".
 * "Unknown browsers": I don't see how this applies to OS detection. Again, there's probably a strong correlation, but as I demonstrated with Lynx, it's certainly not 100%. I suspect you're mixing the two up? The "Unknown" is about OSes, not browsers.
 * "Can the 'Windows' category be trusted?": Your conclusion seems to be that we can't trust those numbers either. And ReactOS is indeed another good problematic case! Hadn't even thought of that.
 * "in this case the 4% unknowns could belong to any OS": Ah, so you agree with my original point of treating the "Unknown" category as an error on the OS-determination. Great!
 * "but probably belongs to the more obscure systems not counted in StatCounter.": Sure, but without breakdown numbers from StatCounter, there's no way to know for sure.
 * "Remember, not everyone is connected to the Internet,": Yes, so focusing on web clients at single-digit percentage levels is a flawed premise from the start. Great, does that mean we can now stop doing that? Especially since you've just agreed that we should be treating the "Unknown" category as the measurement error? I mean, if that's the conclusion we both are drawing, then that wraps up our discussion here! --DanielPharos (talk) 14:04, 26 August 2016 (UTC)
 * I WAS "treating the ´Unknown´ category as an error", until User:Comp.arch got verification directly from StatCounter:
 * "Unknown (are) a consequence of our Android device detection system. It requires frequent updates to keep on top of new models, but the volume required has been increasing. At the moment when a mobile device's vendor and model is unknown we also record the operating system as unknown with a default 'desktop' platform. Looking at the raw useragent strings for these hits, they are indeed predominantly Android useragents that haven't yet been accounted for in updates to our handset detection."
 * "We're setting in place a more intensive and frequent update cycle for our handset detection, which should in the near future see a reduction in the 'Unknown' percentages." --comp.arch 19:28, 20 February 2016 (UTC) ...In short StatCounter can´t keep up with the rate of device release, especially in Asian markets Ne0 (talk) 15:27, 26 August 2016 (UTC)
 * Ooh, that's certainly interesting; they may not be breakdown numbers, but surely it's as close to that as we'll every get, I suppose. Is there any source/reference for that StatCounter statement? --DanielPharos (talk) 15:39, 26 August 2016 (UTC)
 * Never mind, I see you included a link. And I've now also seen the "private communication" ref. How is that not OR? I didn't know such a ref is even allowed on Wikipedia! --DanielPharos (talk) 16:27, 26 August 2016 (UTC)
 * And I predict the rate of unknowns is set to increase...
 * Mozilla/5.0 (Android 6.0.1; Tablet; rv:48.0) Gecko/48.0 Firefox/48.0
 * Firefox Android app on my Chromebook Pixel 2015 shows up as "unknown 6.0, desktop" on StatCounter. So, Next year, as 2-in-1 and convertible Chromebooks are released with Android running in a container, we will start getting more and more unknowns ! The conundrum with this is, how do you identify a vendor or model from the user-agent in a browser running in emulated Android ? you can´t ! --Ne0 (talk) 16:58, 26 August 2016 (UTC)
 * Yes, so estimating the usage share of operating systems from web clients is a horribly inaccurate method to do so, we've already established that. What are you proposing we do about it? --DanielPharos (talk) 17:00, 26 August 2016 (UTC)
 * Maybe get StatCounter to drop the Vendor and model requirement for mobile devices ? At any rate, even if you are using an emulation of Android, you are still using Android, and not the underlying OS. The 2-in-1 and convertible Chromebooks are meant to be used as both Tablet and Netbook. Having a browser built for a touchscreen OS(Android), on Chromebook makes it much more easier to type on a Chromebook device in Tablet form. --Ne0 (talk) 17:14, 26 August 2016 (UTC)
 * I suppose that would reduce the size of the "Unknown" category, yes. But I don't think it would solve the underlying problems. However, I'll take any improvement we can get; can you see if StatCounter is interested in making that change? --DanielPharos (talk) 17:21, 26 August 2016 (UTC)
 * "I've now also seen the "private communication" ref. How is that not OR?" It is, but not disallowed.. I've only "private communicated" (in extreme cases).. and seems I, or someone, needs to again (I can't make them fix it (just say it's alarming) but you can ask what "Unknown" means). WP:V is the major rule in my book, and OR to confirm is ok. It seemed ok, to ref "private" communication (read e-mail, but I'm not sure about quoting private e-mail), better than nothing, as anyone can also e-mail StatCounter. [I've seen refs to "private communication" in academic papers..] The problem is, they do not know, but they seem to know it's not Windows at least. I was hoping they would just resolve (or reduce scope of) the "Unknown" situation, so this thread would have been avoided.. before the "Unknown" situation got out of hand. At least I brought this up in Talk.. comp.arch (talk) 17:22, 26 August 2016 (UTC)
 * It can be pulled out of OR territory if they are willing to make a public statement about it. Although I can understand if they'd rather not...
 * To prevent this discussion from happening in the future: is there some way we can gather all the information in this and the previous thread (February 2016) together, and somehow put it in the article? Either as a comment, or maybe as a separate (hidden?) page, and then link to it with comments? It would be silly to have this conversation over and over again every few months... --DanielPharos (talk) 17:26, 26 August 2016 (UTC)

Android is Linux
This article talks about Android as if it is different from Linux. Android is built on top of Linux. Sam Tomato (talk) 21:48, 25 June 2016 (UTC)


 * Linux is a kernel and Android is an OS. They should be not mixed in the same soup.
 * Plus Android is NOT similar to classicals linux distros. It would be confusing and distorting if wikipedia mix them under the same umbrella.
 * Most people dont know and dont care whether or not Android "is equal to" Linux (aka: Android have a little piece called Linux and just linux militants are remembering it). Wikipedia must not preach a gospel that no one is asking for.


 * Actually the article, sometimes yes sometimes no, talks about Android like a Linuxs limb. Of course, that's wrong.
 * The right is talks about OS because the wikipedia article is about OSes. Wikipedia should not be kernel centric sometimes yes, sometimes no, because it's wrong and not necessary. If it's Android then should say Android, and if itś other "Linux based system" then should say Linux Distro (or GNU/Linux). Wikipedia should not talk about "Linux based systems" because OSX is not named a "Match based system" e.g. — Preceding unsigned comment added by 186.92.106.193 (talk) 00:10, 1 September 2016 (UTC)

Linux ≠ Android
Right now the article sometimes mix Android, Chrome OS and others under the Linux brand. That is a non-technical "internet" culture but in this wiki article is unnecessary and wrong.

Linux is a kernel and Android is an OS. They should be not mixed in the same soup. Plus Android is NOT similar to classicals linux distros. It would be confusing and distorting if wikipedia mix them under the same umbrella. Most people dont know and dont care whether or not Android "is equal to" Linux (aka: Android have a little piece called Linux and just linux militants are remembering it). Wikipedia must not preach a gospel that no one is asking for.

Actually the article, sometimes yes sometimes no, talks about Android like a Linuxs limb. Of course, that's wrong. The right is talks about OS because the wikipedia article is about OSes. Wikipedia should not be kernel centric sometimes yes, because it's wrong and not necessary. If it's Android then should say Android, and if itś other "Linux based system" then should say Linux Distro (or GNU/Linux). Wikipedia should not talk about "Linux based systems" because OSX is not named a "Match based system" e.g. — Preceding unsigned comment added by 186.92.106.193 (talk) 00:17, 1 September 2016 (UTC)

Device shipment usually means eventual sales to customers
If the devices shipped did not sell, we would be getting MORE headlines similar to "4 million tablets worth US$ 900 million left unsold".

So, Device shipment means eventual sales to customers, or in some rare cases, prizes for event participation. --Ne0 (talk) 07:37, 23 September 2016 (UTC)
 * Yes of course devices are eventually sold (at least most), no business would buy much more than they can sell.. It could however be bought by the business in 2015 and sold to consumer in 2016. And sources only say about shipments in 2015. comp.arch (talk) 08:33, 23 September 2016 (UTC)

Best order of sections?
, made some edits, first one unexplained, that seems to just change order of section, undoing my change I did in September.

I intentionally moved "desktop" market share lower, as they are no longer a dominant computing platform (just as mainframes, are now at the bottom of the article). They are still important while under 50% market share (and the formerly dominant Windows, even lower).

I feel, having the first section showing all operating systems or platforms, is important. If you only look at desktops, then Microsoft was dominant with Windows, and which version of it most popular isn't he biggest news, with them all being the same platform (more or less). comp.arch (talk) 16:26, 5 November 2016 (UTC)
 * Even worse, the whole section was deleted. Haipowerfull replaced "Web clients" with a copy of "Desktop and laptop computers", and then deleted the original copy as a duplicate. I restored web clients, back to where it was. Sorry, I got the name wrong on the edit summary. --Ne0 (talk) 06:58, 8 November 2016 (UTC)

Linux is the name of a kernel & Chrome OS is a GNU/Linux system
There is a huge problem in this article.

GNU/Linux systems — equivalent to Windows 10 or Mac OS X — are named Linux, the name of a kernel. If Linux is supposed to refer to any operating system using that kernel, then Mac OS X should be named Mach.

In the same way, Chrome OS is a GNU/Linux system. Hence it makes no sense to compare the market shares of “Linux” and “Chrome OS” on the same graphic.

This results in a very confusing article. The reader won't understand which is which. — Preceding unsigned comment added by Excubia (talk • contribs) 18:45, 24 August 2016 (UTC)


 * Linux is ambiguous yes, some use it for GNU/Linux also. However, "ChromeOS is NOT a GNU/Linux distro [..] does not ship with the following". If it had a lot of GNU, what is usually expected, there would be no need for cublinux.com (note the authority on GNU, isn't happy with ChromeOS, while saying: "Now let's consider Google ChromeOS, a variant of GNU/Linux" but he doesn't control what name is used). "Some in the Linux community have criticised Chrome OS as not being a "normal" or "traditional" Linux distribution." comp.arch (talk) 16:11, 26 August 2016 (UTC)
 * I like your answer. There is still something: you must have a good reason to isolate CrOS from the linux group. Either it's because it is non-GNU (that would be perfectly acceptable) or because CrOS is such a huge part of Linux distros that it makes sense to single it out. This is not the case and the graphics are less understandable with percentages < 1%. My conclusion: either we merge everything into “stuff that runs Linux” or we distinguish “GNU/Linux distros” vs “non-GNU Linux distros”. Excubia (talk) 02:06, 5 September 2016 (UTC)
 * Feel free to edit this page (or any other) if you think you can make it clearer. Chrome OS (and even Android) is an edge case.. compared to GNU/Linux (or Windows if you will). Linux is the common name understood by techical people to mean GNU/Linux. I try to explain the big trends (and leave minor OSes such as ChromeOS mostly not explained, at least on this page until they get market share, yes, I know it's popular in US education, but no real installed base globally). I didn't "isolate" Chrome OS, StatCounter (the source) did, justifiably as it doesn't run the same "Linux" programs (only, with few exceptions, web apps). I just explained that it's not counted with [other] Linux. comp.arch (talk) 18:03, 5 September 2016 (UTC)
 * The “you” was meant to be a rhetorical “someone”. I think you, personnaly, do a fine job AFAICT :-) Excubia (talk) 20:01, 5 September 2016 (UTC)
 * Another distinct comment. As I read the links you cited, I understand that CrOS doesn't have a C compiler. Can we still say it's an “operating system”? Excubia (talk) 02:12, 5 September 2016 (UTC)
 * ChromeOS is an operating system, it doesn't need a C compiler (that is however unusual for GNU/Linux, but it isn't that), no more than Windows. comp.arch (talk) 18:03, 5 September 2016 (UTC)
 * See also from Stallman (the authority on GNU): "There are complete systems that contain Linux and not GNU; Android is an example." He really should look at Chrome OS the same (sometimes he's not up-to-speed, as he does not browse the web); until (Chrome OS (or Chromium OS), says it is "GNU/Linux" (when the maker does, which is never, as it has very little of GNU and it's not "exposed" to [web] apps), it shouldn't be discussed here (at best there, but there it will be protested). comp.arch (talk) 21:23, 5 September 2016 (UTC)

I suggest either we change Linux -> "GNU/Linux";

OR we change Linux -> "Linux Desktop" and add Chrome OS to it.

Chrome OS is not GNU, but it does use Linux (kernel).

The proof of that is Crouton: Chrome OS + XFCE = Xubuntu.

So Chrome OS is just a browser on top of the Linux (Kernel) --Ne0 (talk) 10:39, 26 September 2016 (UTC)


 * I think it's best to not go with "GNU/Linux", ot e.g. not exclude Chrome OS. Neither use Linux Desktop, in case there may be non-Android tablets. Conversely there would also be Desktop Android. Just Linux to catch the most OSes would be ok, as numbers are small anyway. Where it matters say Linux excl. Android, to not double-count. We can deal when say Chrome OS gets very popular. You can of course whange as reflected by sources, but I will be looking at the bigger picture, mostly. I know about Crouton, doesn't matter what you can add, if you would change to Xubuntu, it will get counted as such. Sources count Chrome OS as an OS; yes a browser + kernel can be that. comp.arch (talk) 10:01, 6 December 2016 (UTC)

100% Linux Country
South Georgia and the South Sandwich Islands is the only country with Linux user majority (100%), with Linux counter registering 25 Linux users, and the total estimated population is at 30 people  :) Should we mention something about this in the article ? --Ne0 (talk) 14:37, 6 December 2016 (UTC)


 * No, first, it's not a country. It's a British overseas territory, and 30 people.. Fun to know such trivia here on the Talk-page, but I leave such out mostly (small population [mostly islands] tend to also have very erratic web traffic statistics; see also Antartica). An exception can be stating "Unknown" high, that I feel/felt should be in.. It's going down so I do not make much stink about it. comp.arch (talk) 16:27, 7 December 2016 (UTC)

Analysis and absolute numbers "Platform wars: the final score"
from a venture capitalist (at least I find this [whole] article and absolute numbers interesting): "Globally, something around 5bn people (give or take perhaps 250m each way) have a mobile phone, out of around 5.5bn people over the age of 16. [..] This implies something around 175m iPhones in China across all three operators. This is more iPhones than there are in the USA. [..] Plotting this as a chart (showing only Google's rounded numbers at scheduled events), a March 2016 number of 1.5bn total Google Android and hence 1.35bn phones looks plausible, and might be high.

[showing s-curve of a saturated market] [..]

Annual smartphone sales will rise to close to 2bn units and PC sales fall to close to 200m, while the smartphone install base will rise from 2.5bn to close to 5bn and the PC install base fall from today's 1.5bn to closer to 1bn (if that). So mobile has 10x the unit volume and 5x the install base - 'a billion is the new million'. This is why all the industry investment is shifting to the ARM/iOS/Android ecosystem from the WinTel ecosystem. [..]

That is, the war is over. Yes, we'll go from 2.5bn smartphones to 5bn, but the dynamics of the two ecosystems will not change much with that growth." comp.arch (talk) 17:47, 5 September 2016 (UTC)
 * Interesting stats...
 * Developing countries are more likely to use Android, while developed countries are more likely to use iOS or Windows
 * Also, developing countries are more likely to pay for Android apps (atleast in Asia)
 * ...Should we add this to the article ? --Ne0 (talk) 19:28, 9 November 2016 (UTC)
 * @User:Comp.arch, there seems to be a lot of correlation between these two maps: File:StatCounter OS Worldwide December 2016 map.png & File:IMF_Developing_Countries_Map_2014.png --Ne0 (talk) 13:02, 2 January 2017 (UTC)
 * few links --Ne0 (talk) 12:27, 4 January 2017 (UTC)


 * [at some point I wrote a longer answer and my browser crashed and deleted it..] "the World Bank made a decision to no longer distinguish between “developed” and “developing” countries in the presentation of its data. Nobody has ever agreed on a definition for these terms in the first place.[9]" comp.arch (talk) 16:15, 9 January 2017 (UTC)

Graphics API
Operating Systems usage of Graphics API... ...should we add it to article, under Market share ? --Ne0 (talk) 05:12, 17 January 2017 (UTC)


 * I don't think so. Maybe elsewhere. Comparison of..? This is a part of an operating system, if you stretch the definition. This is an implementation detail (that most will not know or care(?) about). I do take the position that a UI is part of an OS. Then should we have a table of X vs. Wayland vs whatever Windows and macOS use? [I predict that Vulkan wins the battle, yes it's not native on Windows and macOS, but available on and an iOS.] comp.arch (talk) 19:43, 30 January 2017 (UTC)
 * Ok, moved it to Graphics library --Ne0 (talk) 07:21, 1 February 2017 (UTC)

I'll say at end of month that the word is desktop-minority
I'll qualify; e.g. minority on Fridays, Saturdays and Sundays (and even one Thursday; 20. Oct.), but not [other] weekdays. Also depends on regions. Add to the lead? comp.arch (talk) 22:15, 30 October 2016 (UTC)
 * Added as a pie graph --Ne0 (talk) 05:41, 28 January 2017 (UTC)

Anyone notice the fall of Firefox along with Internet Explorer ? ...and the rise of Safari along with Google Chrome ? 

Note that, Firefox & IE are mostly used on desktop, while Safari & Google Chrome are also used on smartphones ! Fall of the desktop OS, along with it's exclusive browsers, loud & clear ! --Ne0 (talk) 10:12, 3 February 2017 (UTC)

W3Counter 2015-02 data do not match those within table
Most (if not all) of the numbers on https://www.w3counter.com/globalstats.php?year=2015&month=02 differ from those in the table (web clients). Like 7.36% instead of 7.46% for XP, 2.44% instead of 2.56% for Linux,, 39.28% instead of 39.85% for Win7 etc etc.

Is this a transcription error or has the data on w3counter changed? --2.246.134.14 (talk) 19:36, 9 January 2017 (UTC)


 * EDIT: Yes it changed in the source see its archive. Former theories: I'm not sure why, this is not info from the month before, or after (or year later); it was pout in in this edit so it's not in the middle of the month. Another of my guess, beware of doing that, as numbers change until month is over (at least at StatCounter; some may not show incomplete). comp.arch (talk) 16:02, 10 January 2017 (UTC)
 * On StatCounter, data for the last day of the month is updated around 36 hours into the next month. So updating monthly stats after the 2nd of following month should be safe. Also, data mismatch can occur when User Agents are re-identified as belonging to another Operating System, and existing graphs are updated to reflect this. I believe this is what happened in this W3Counter case. --Ne0 (talk) 08:03, 16 January 2017 (UTC)
 * Updated W3Counter line with December 2016 stats --Ne0 (talk) 05:52, 24 January 2017 (UTC)
 * Note that the median (if it is a median?) on the top of the table has not been updated. -- 2001:8A0:6D12:6900:84A2:D66B:81AB:C32B (talk) 18:30, 14 March 2017 (UTC)
 * Stats on top of the Web clients table is the total web usage by Operating System kernels(NT, XNU, Linux), as determined by StatCounter. I last updated it, when I added StatCounter stats for Dec 2016. --Ne0 (talk) 07:12, 15 March 2017 (UTC)

Nintendo Switch
The Switch ran on a BSD core - http://i.imgur.com/By9ojVt.jpg so should it not be moved to the "Unix-Like" column? — Preceding unsigned comment added by Donatj (talk • contribs) 17:46, 23 April 2018 (UTC)

Table sorting
There seem to be some conflicting ideas about how the tables in this article should be sorted. These tables really need to be made sortable, though that may require adding quite a lot of data-sort-value markup. -- FeRDNYC (talk) 08:28, 3 September 2018 (UTC)
 * Single-source tables like the Gartner and Steam stats seem to be sorted chronologically, newest-first.
 * The Web Clients multi-source table is sorted alphabetically by source, but then it falls apart after that:
 * The StatsCounter group is sorted newest-first.
 * The W3Counter group is sorted oldest-first.
 * The first table in § Mobile Devices.Smartphones isn't sorted by anything. It's just a jumbled collection of stats lines, all mixed together hodge-podge.

Big Trim
Had a Big Trim of the intro - which is now a more reasonable size. Please don't quibble about the numbers, or add wikilinks and refs all over the place. An intro, as per policy, should be clean and readable - there's plenty of work needed in the body of the article (and there details and referencing are appropriate). - Snori (talk) 07:28, 30 November 2018 (UTC)

Desktop at least temporarily back to more popular than mobile
I noticed desktop has had a few good weeks [in November]. In all of them it dips below 50% but most of the time is way above. It seems to be an exceptional month with mobile now more popular (but not for the whole month). Should we make a big deal of this here? Or wait for a press release as StatCounter released when desktop got less popular?

EDIT: the reason probably is "officially" Android share is dropping below Windows BUT note "Unknown" is going way up, about the same as the Android drop so I expect we need to investigate that first. I may do that and/or ask StatCounter directly. I have a feeling much or all of this "Unknown" would be Android. Such has been the case before (some fork of Android now?); at least it shouldn't be Windows. comp.arch (talk) 18:41, 30 November 2018 (UTC)

Flawed statistics
Majority of statistics are from Statcounter.com this site only counts OS by web usage. There are probably (tens/hundreds?) millions of computer OS's that never connect to the internet that are excluded by this analysis. So any statement of global OS use based on this sites statistics is flawed and unreliable. No source that doesn't account for the worlds entire installed computer base (as opposed to those that are being used to access the internet) should be used as a reference for this article.

As it stands the title of the article should be ammended to read "Usage share of operating systems used to browse the internet" — Preceding unsigned comment added by 2620:104:4001:71:8567:26E3:CCEC:2596 (talk) 08:20, 6 April 2018 (UTC)
 * Oppose proposal; all methods of measurements have limitations and you've identified one of the them. Rather than change the title, I suggest just describing the limitations of the measurement methods on the page. There is a policy of keeping titles concise. Klbrain (talk) 09:12, 22 July 2018 (UTC)


 * Oppose proposal, because having the full picture (from elsewhere) seems wouldn't change the ranking. I looked into this extensively when Android was getting popular. Of course, many (most?) Android users also browse the web, but a lot do not and only use apps (yes, gotten through the internet, but not counted as web use then). What operating systems do you propose as more popular than it or e.g. Windows? I looked into the possibilities of embedded real-time operating systems, but most embedded microcontrollers use no OS. Other much used computers not accessing the web (while possibly operating as web servers, not what is counted) are mainframes. There are just too few of them, but each one is huge. comp.arch (talk) 18:50, 30 November 2018 (UTC)

Android and iPad battle/switching places and tablet share statistics
I noticed a seemingly a contradiction in Statistica's predictions, but more likely an improvement is predicted sales of tablets. They predictably, by now, predict a decline in installed base of desktop PCs (in the US) every year from now on, and commulative decline of 31% by 2022 (from 2017). In the same graph they show "Mobile PCs" (assumed to be laptops, the term from the title) and tablets to increase their share. However, mobile PCs/laptops will not, but almost, make up for the decline in desktop PCs. I.e. the total decline is predected to be 0.7% decline.

They predict a 5.5% increase in the installed base of tablets (so and thus the installed base of those and the PCs above are predicted to increase for at least a few years).

Now the "contradiction", from their other prediction on tablets (for a similar period, but I guess older prediction, since starts one year earlier in 2016):

"The statistic shows the installed base of tablets in the United States from 2016 to 2022. The installed base of tablets in the U.S. is forecast to fall to at 168.92 million units in 2019."

I.e. having the installed base already peaked and declining year by year from now on.

Android on tablets has overtaken (up to 61.34%) iPad in 2018, then iPad recovered, then in April Android just barely got majority (50.50%) then iPad recovered, now at 51.39% comp.arch (talk) 00:06, 1 December 2018 (UTC)

Does this page need to go into detail about the growth and decline of different segments of the computing market?
In this edit, the article had a bunch of stuff copied from mainframe computer added to it, in a new "Decline" subsection of the "Mainframes" section, discussing the decline of mainframes.

It also has a large section "Crossover to smartphones having majority share".

Do detailed discussions of the growth and decline of various segments of the computing market belong in this article, or do they work better in articles about the segments themselves, with this article mentioning the growth or decline without all the details and just pointing to those articles, or sections thereof, with "See also:" links? That growth and decline is less an issue of "the OSes used in that segment became {more, less} popular" than "the platforms themselves became {more, less} popular", with the platforms being the hardware as much as the software (e.g., computers you can hold in your hand became more popular than computers you can put on your desk or on your lap, for reasons that may have had more to do with the form factor than to do with Windows vs. {Linux,Darwin}). Guy Harris (talk) 20:11, 25 April 2020 (UTC)

The HPC system information is wrong
The current (2018/10/26) page text claims: "In June 2017, two AIX computers held rank 493 and 494,[230] the last non-Linux systems before they dropped off the list." The "last non-Linux" part of this claim is wrong.

As of 2018/10, the #8 machine on the TOP500 list is Sequoia. It is an IBM BG/Q system, and the operating system it runs on the compute nodes is not Linux. It does implement some parts of Linux syscalls (e.g., common read and write parameters), but it implements only partial syscall support, and is far from "Linux". Partial syscall support reduces the memory/compute footprint on each node -- a good trade-off for HPC -- but supporting only a few Linux features is not "Linux".

Even among "Linux" vendors, support may be 99% (or 98% or 90%, or whatever) rather than 100%. For example, some vendors support HugeTLBfs for only a subset of the hardware's page sizes. This may be a good engineering choice, but may still violate Linux compatibility -- e.g., some Linux compatibility tests may fail.

One way to explain this may be to say HPC systems use Linux, or some HPC-friendly subset of the Linux API, and that since the AIX systems fell off the list, all systems either use Linux or a subset of the Linux API. — Preceding unsigned comment added by 2600:1700:daa0:e2e0:b961:705e:436f:b84f (talk) 03:48, 29 October 2018 (UTC)


 * Oppose as while this may be technically true, is it really helpful for most readers (and vendors' and TOP500 info claim "Linux" and we should go with what sources say)? The Linux kernel (and the full "Linux" OS) is a moving target, just as any non-dead OS is, and can we say that e.g. some old Linux API is fully supported (regarding e.g. huge pages, a thing not always supported in Linux). One other thing is that the host nodes (as opposed to compute nodes) support Linux, I believe have full support. Yes, those do not supply the full compute power, just a fraction of, but you can say the same for the GPUs, that provide a large part (often majority) of. In a sense the compute nodes are just as much "accelerators" to the host nodes as the GPUs. comp.arch (talk) 19:01, 30 November 2018 (UTC)
 * My laptop's OS has the same parameters to read and write that Linux has, but it's most definitely NOT running Linux; heck, an operating system for the PDP-11 that came out in 1979 had the same parameters to read and write that Linux has. (And, yes, the latter statement is the similar to "PL/I had the same comment characters as C89. :-))
 * I.e., "has the same parameters to system calls dating back to V7 that Linux has" is rather far from being sufficient to say the OS is close enough to Linux to call it Linux. Now, if the Compute Node Kernel has some calls that implement more than the POSIX versions of those calls *and* extend them in a Linux-compatible fashion, perhaps saying they run a "Linux subset" would work. Guy Harris (talk) 05:27, 26 April 2020 (UTC)