Slashdot Mirror
Ask Dan Kusnetzky About Linux Server Counts
How many people run Linux as a desktop OS? How many servers run Linux? Is the Linux server market share 8.6 percent or 24 percent or somewhere in between? Dan Kusnetzky is a heavily quoted analyst at IDC who wrestles with questions like these for a living. This is your chance to find out how analysts come up with all those numbers -- and why analysts seem to disagree with each other so often. One question per post, please. We'll send 10 of the highest-moderated ones to Dan, and post his answers as soon as we get them back.
Dialing numbers and asking what is running is what Gartner did when they came up with 8.?%. From their article, note they had installed on it :
1 ,2772060,00.html
http://www.zdnet.com/eweek/stories/general/0,1101
But Gartner's Hewitt was adamant that his methodology and findings were correct. Some 724 U.S.-based respondents had answered questions over the phone for the survey, he said, ranging from small organizations with fewer than 50 PCs to large companies with more than 500 computers, as well as educational institutions, Internet service providers and application service providers.
Respondents were screened to ensure they were knowledgeable about server purchases over the quarter, and they were asked what percentage of their server purchases were Linux servers, he said.
"We went to end users, rather than looking at just sales numbers, and asked them what servers they had bought over the past three months and what operating system they had installed on it over the same period," Hewitt said. "There was no question about whether Linux was preinstalled or not, we simply asked about new shipments and this is what we found."
Don't any of you find it all amazing how you blindly accept the statistics from IDC, but question the ones from Gartner?
All because IDC has larger numbers.
I would seriously like to see IDC justify these numbers as accurate. From what I've seen of their press releases they extrapolate off some pretty questionable data.
At least Gartner went to the trouble of performing a survey.
What you're really asking is if Gartner Group or whatever consulting group in question is willing to sell their name with a set of "results".
I'm sure they are, as long as there's some plausable truth in the results.
Bet their methodology is "Company Confedential".
FWIW: At a place where I worked (many years ago), that was a higher category than official US Govt. Secret.
Caution: Now approaching the (technological) singularity.
I think we've pushed this "anyone can grow up to be president" thing too far.
Personally, I call it a server if it has a static IP address. ftp, etc., can be easily added and removed as desired. Changing the IP address is much harder.
Caution: Now approaching the (technological) singularity.
I think we've pushed this "anyone can grow up to be president" thing too far.
I hope they aren't doing this on the basis of "# of copies sold", unless that's what they're trying to measure. It sure doesn't have much to do with the number installed. At home I have maybe 10 boxed sets of Linux, two CD's of Windows versions from MS, various different flavors and versions. I also have two copies of Linux installed, and one copy of Win95B. (Win98 got pulled, because I don't like my computer calling home without asking me. [Also because it crashed more than Win95, and was incompatible with the most important application we keep the computer for.])
But if I owned a hundred computers, I'd have the same number of copies of Linux, and over 100 copies of Windows (it comes automatically with just about all computers, even it you get them with only Linux installed).
So number of copies sold doesn't have much to do with number installed. It systematically overcounts Windows, and both over and under counts Linux.
Caution: Now approaching the (technological) singularity.
I think we've pushed this "anyone can grow up to be president" thing too far.
What are the breakdowns for age and linux use? Does older mean wiser or does younger mean more 31337.
Scott Ruttencutter
We Apprentice Developers and Designers
"One question per post, please"
Similarly, many Windows (yes, and *n?x) users run Napster and clones. Are these servers?
Jesus was all right but his disciples were thick and ordinary. -John Lennon
I'm wondering how these numbers help us in general. Right now, them seem to be nothing more than means to gloat and or demean competition. I've also read a couple articles like:
m l
http://www.theregister.co.uk/content/4/19661.ht
which show the meaninglessness of statistics like these up to a point. What are your thoughts?
NAS
How are copies counted?
How are throw-aways counted?
How are pre-load erases counted?
How is piracy (of proprietary SW) counted?
Aide: Grant drinks too much to command an army. Lincoln: Find out what he drinks and give it to my other generals!
Given that it's rather ironic that a Microsoft-funded study would actually slant in favor of the competition (even if there's a pretty clear bias against it in the first place), what do you, as someone who is part of this industry, think of the ethics involved in doing this kind of research for hire, and perhaps more importantly, should anyone outside the company commissioning the study really be paying attention to any such research in the first place?
/Brian
If you really want to check all of his quotes, try the same Google search with his name spelled correctly.
Have fun: Join D.N.A. (National Dyslexics Association)
The things that it clearly does not take into account are:
- Servers that come with Windows (or some other OS) that people wipe and install Linux on. Keep in mind that many vendors do not provide Linux as a pre-installed option, and (thanks to Microsoft's anti-piracy efforts) often do not offer to sell unconfigured systems. The result is that many companies who want to run Linux on their servers must buy Windows licenses anyway.
- Many companies and organizations build their servers from scratch. If you assume that all servers are purchased from vendors and OEMs, these servers will seem to materialize out of thin air. Also, people who are savvy enough to build a server from scratch are also more likely to make an informed decision about what type of server to use (and thus, be more likely to go with Linux or BSD).
- "Recycled" servers. In many companies, it is common practice to move aging desktop systems (especially the more powerful ones) into low-end server roles. For instance, intranet servers in small to medium sized companies. If the IT people know what they are doing, these former desktops will get formatted and rebuilt, and there's a good chance they'll get rebuilt with Linux, since one of the strong points of Linux is its ability to make good use of older hardware.
- Beleive it or not, some people dual-boot servers. This is especially convenient when you put the bootable image on a hot-swap drive sled. This may sound odd, but it is partiularly convenient for production enviornments, where you might only need a server up for a night or two after a commit cycle. It's also conveinent with large database enviornmetns, where you need to do periodic aggrigations. I really, really doubt that you'd be able to find this option on a vendor's configuration page.
- "Junker" servers, often built from aincent hardware. People save themselves piles of money by using 386's and such for routers and firewalls instead of paying for custom hardware. This is a good solution for many people, especially since Linux does a good job of utilizing older harware. This is different from "recycled" servers because routers and firewalls tend to be transparant peices of network infastructure. However, no one would despute that they are important.
There are probably a few situations that I'm leaving out. The point is, how important are vendor statistics for measuring the Linux market share?It seems to me that the only way to make informed statements about server market share is to gather statistics at the source. Put together a statistically significant sample of companies, individuals and organizations, and ask them for a list of the servers they use, a description of of each (including OS), and an explanation of their choices. This seems like the only way to actually answer the bigger question of server OS market share. The answer might not make anyone happy, but at least it would be an answer.
--
In spite of the suggestions and all the tests that I have made, I have not cavato a spider from the hole.
In your estimation do you think MS' campaigns for "Shared Source" combined with their opinions of Linux as a server platform been effective in relegating Linux's ?
By extension, what affect do you forsee on the numbers of server sales as a result of these campaigns?
BOSTON SUCKS!
The Microsoft funded study is obviously skewed. First let's look at how Microsoft computes its own "Market Share." In its computation, it adds in an accounting for "pirated copies." So clearly, for basis of comparison in most cases, Microsoft is not concerned only with "SALES" figures. So by limiting the study to only machines that are sold with Linux installed, they are skewing the truth immeasurably for two reasons: (1) People don't often BUY Linux (2) Microsoft has a standard practice of including "unpurchased copies" in use when calculating market share.
Now I cannot know if Microsoft's number in this study include pirated software, but whether they did or not for this particular study is almost irrelevant! That fact is, you cannot count Linux installations by the number of machines sold with Linux on them. That's insane because that's not how Linux is most frequently distributed. It's also inconsistant with Microsoft's standard position where they claim an accounting for unpurchased copies in use. At the very least they should make efforts to remain consistant. But if they did, the numbers wouldn't be quite as favorable. They KNOW this.
My humble input:
Thank you for your time.
CrazyLegs
"Pork!!" said the Fish, and we all laughed.
As a sales and marketing manager, I often use analyst reports from IDC or Gartner when working with the CEO or CFO of a prospective customer because they trust the source. It amazes me how quickly a CEO will abandon the advice of an IT staffer based on a report from IDC or Gartner.
If you were the CEO of a company faced with a long term technology decision, how much would you rely on analysts? What sources other than analysts would you look to to aid you in making your decision?
-- $G
I agree that this is a good way to poll about the server OS.
But, as you said, if they didn't have a cross group survey, then it's pretty useless.
ISPs tend to use *nix deriatives, and office networks (file & print sharing, mainly) tend to be NT, frex.(Just the two most distinct groups that I could think of, don't flame me for this)
You could get a totally true, but meaningless statistic if you poll your target correctly.
--
Two witches watch two watches.
--
Two witches watched two watches.
Which witch watched which watch?
How do you (try) to determain Linux market share?
Many Linux installations have not been bought, after all.
And while we are at the subject, how do you diffrenciate between Linux used as a desktop platform and as a server platform?
--
Two witches watch two watches.
--
Two witches watched two watches.
Which witch watched which watch?
That's easy: get it from the FBI. They've got Carnivore systems planted all over the Internet that could snoop out this info. Since it's aggregate data, no warrant necessary. :-)
One thing I thought is if server x can do on one machine what server y does on four machines. Doesn't server x hurt itself?
Emile Snyder
www.talentcodeworks.com
How did you go about researching your figures? Who was involved? How many people companies or vendors did you ask? What statistics did you track?
In short how did you come up with your figures. We could ask about the other figures too but since you didn't publish those someone else may have to answer that.
--= Isn't it surprising how badly I spell ?
SuperID
Free Database Hosting For Developers
During Linux servers market share study (or similar studies involving Linux), are you concerned about the fact that there's no big Linux company who will ever buy you such work, meanwhile Microsoft or other giants sure will ask again for it in the future ?
Does it affects you or your team work ?
I've read your report, and I didn't find any description of the mechanism by which you came up with your estimate of the average number of times that a given purchased copy of Linux is installed. That's probably my own error, but since you're answering questions, would you be willing to enlighten me?
You use the number "15". Frankly, I'm surprised that it is so big. Upon what data did you base that estimate? Who did you interview to get it? I realize that any such estimate would have to account both for the very large number of installs at large colocation and/or service provider shops, as well as the number of untrackable network installs that take place. But it would also need to account for the number of times where a machine had Linux installed upon it, was used as an experimental development platform, and then was wiped, not to mention the number of cases where somebody bought a distribution, and then never installed it at all.
I would like to see a breakdown of server O/S weighted by usage. For example, a server that receives 1000hits/hour would have 1000 times more weighting than a server which receives 1 hit/hour. Such a weighting would give a much more honest view of usage. In other words, which server O/S is getting the most work done. I suspect this would lead to significantly different results from what we have seen so far.
"To those who are overly cautious, everything is impossible. "
Many Linux desktops have ftp, telnet & http ports open, so do they count as severs too?
no, they count as targets for l33t haX0rs, especially since the people who just installed mandrake from that wal mart cd don't even know they're open or use them...
--saint----
When most desktop and server Linux users are building their own boxes, how can sales of server and pc systems be relevant? Most people I know that use Linux (myself excluded) first bought a PC with wDOS on it, removed that OS and installed something else. What kind of formula do you use to make up for this disparity between systems sold and systems built ?
The heat from below can burn your eyes out
Hi Dan,
According to IDC figures, the Linux desktop market share was 5% and now it's less then 2%. Gartner numbers are of course differently (way lower if I'm not mistaken)
Yet, when I look at developments of projects like XFree, KDE, Gnome, Linux kernel - or when you get the daily list from freshmeat, or even talking to the the ISP who host mirrors of ISO images of Redhat, Mandrake etc - then you see that linux get FAR more then 2%. Hell - if it was 2% and you account the developments of Linux - then each developer works 25 hours per day on a porject!
So, as you can see - the numbers here are definately wrong here - and those numbers are actually hurting the Linux community. If an ABCD company wants to make a software for the Linux desktop and they see those IDC figures - then they will say something like "oh, 2%? no thanks - we'll make it for Mac - they are %5+", and we'll loose..
Comments?
Hetz (Heunique)
Is there a relationship between company size and linux use?
Do we have evidence of any companies with multi-million dollar revenues that rely on linux solutions for their servers or do most companies that use linux servers do it because they lack the money? (choice/only option)
-S
Scott Ruttencutter
We Apprentice Developers and Designers
Dan,
How does IDC determine if the answers they are getting are valid? What I mean by this is how does your organization insure that the people they are surveying being truthful in their responses? I can see scenarios where a well co-ordinated group of people could be swayed by a vendor to influence the results of a survey, especially in a field as specialized as the IT industry.
Also, if you suspect that the results of a survey are tainted, do you publish it anyway (with caveats included, of course) or redo it (at your time and expense)?
If "disco" means "I learn" in Latin, does "discothèque" mean "I learn technology"?
In studies of market share(or studies in general), we often hear quotes about who funded them. This seems to somehow imply that those who funded the study had some influence in how the data was gathered/interpreted.
Even if the funder does not infulence the questions asked they may well control if the "results" see the light of day.
I have often wondered how biased polls are based on the questions asked, the demographics of the people polled, etc. When results about polls are made public, is it also possible to obtain information about how the poll was conducted in a simple, by request method?
It's quite trivial for methodology to affect the results. That is why in any kind of scientific publication the method is considered at least as important as any conclusion.
Not only is no methodology available "conclusions" are being called "results".
Hi Dan,
This is probably going to be a tough question for you (if you ever get it, that is - given the bias on Slashdot, this probably won't be modded up).
A big problem I see with your methodology is that you probably overcount Linux server shipments. From what I understand (I may be incorrect here) you count each sale of Linux as a server shipment. However, many of these copies may be tested but never used, a great many are used in home computers or development workstations. The problem is, unlike with Windows or OS/2, there is no ``client version'' vs. a ``server version''. There is no easy way to tell if a copy of Linux is used in a server environment or a client environment by sales figures alone. So my question is: given total Linux sales figures, how do you estimate the number used in server environments?
Secondly, is it possible that the descrepancy between your numbers and Gartners' is due to a problem in the way you estimate Linux server sales?
---
/bin/fortune | slashdotsig.sh
The technical approach of talking to publicly accessible servers and attempting to fingerprint the OS is fraught with methodological problems.
The naive approach of asking hardware vendors how many units of hardware ship with what OS is obviously flawed.
So why not just do the dumb brute-force thing? Dial phone numbers at random and ask people what they use? Too expensive?
How much is accurate data worth, anyway?
Do you have any insight about trends that can be gleaned from these surveys - that is, regardless of what number one uses for the percentage of users that adopt Linux, is it static, taking off, steadily increasing, or what?
Toronto-area transit rider? Rate your ride.
At what point does a computer become a server? Many Linux desktops have ftp, telnet & http ports open, so do they count as severs too?
I just remembered an Idea that a friend of mine suggested:
What if IDC could work with the Linux distributions (RedHat, Mandrake, SuSE, TurboLinux, Debian) to add a small program which will run after the first internet connection has been succsessfull..
When this program runs - it will ask the user to "register" his copy of the distribution. If it has been purchased from one of the distributors - then the user can add his serial number. Some other questions like will this distribution be used as a server, a workstation, or combination of the 2, or a development workstation. The survey SHOULD be annonymous (unless the user wants to give some details about himself)
By that way - the distributors can give the numbers back to IDC - and IDC can publish a report which will tell that the number of Linux installations - and that number is X. X is combined of Y free download version and Z purchased copies of Linux.
What do you think, Dan? what the slashdot readers think about it?
Hetz (Heunique)
Do you base your data mostly on marketing analysis or do you actually go a pay a consultor to scan machines on the net? If there are scans involved, how do you pick the IP blocks to be scanned and what's the uncertainty associated with such a method (and how is this uncertainty guessed)? If there are no scans involved, why not? If this is "maket analysis", can you defined that for me? Which factors are involved? And a different question: who's the target market for this kind of study? How much does such a thing cost?
- Ask Gartner their methodology,
- Document it so that it can be reproduced;
- Make sure and get Gartner to say "yep that's how we did it".
- Dan&Co reproduces the methodology and compares the numbers
Might be overly scientific. Maybe consider it an "Open Source" version of market research because of the peer review and verification. Some may think this approach has no place in the market research area.But, I (for one) think it would be interesting to call Gartner's Bluff (if indeed that's what this is). I personally place them (and ZD, etc) into the Shill category.
Thoughts anyone?
Don't sweat the petty things. But do pet the sweaty things.
Might he have the stats on the BSD's. People from the BSD community are curious. :)
TIA
If it is possible or already happened, do analysts in general (and you in particular) find it a worrisom possibility, and if so, are there any attempts/ideas to deal with the issue?
Thanks,
DVK
"The right to figure things out for yourself is the only true freedom everyone shares. Go use it"-R.A.Heinlein
In studies of market share(or studies in general), we often hear quotes about who funded them. This seems to somehow imply that those who funded the study had some influence in how the data was gathered/interpreted.
My question is: Do those who fund a study influence how the study turns out (ie. Microsoft studies show higher MS market share). Or is it that these corperations only decide to fund groups who they know will most likely return results in thier favor.
---Lane
What's the point of moderating?!
I have a question on the area of predictions in general. For example, fellow IDC analyst Jill House has been severaly negative on Palm over the years, with regard to the Win CE operating system and devices. A sample quote from her in Feb. 2000 read, "If I was Palm, I would be beside myself with panic."
The issue is, that over the last 3 - 4 years she's been predicting the demise of Palm and the rise of Win CE, a claim that has never materialized. Who verifies the reliability of these predictions and keeps the analysts accountable. With the frequent sound bytes and one-liners that they give to the press, these analysts have significant influence over public perception of the issues. But how is policing done when the analysts don't analyze very well?
Sincerely,
Daniel McCarty
Palm OS Developer
Have fun: Join D.N.A. (National Dyslexics Association)
Do these analyses factor in the so-called third world? Most of these analyses are US-centric or some times do include the continent across the pond but what about Africa and Asia? Do these analyses *really* take inputs from these continents?
Thanks for your time,
- Brad Heintz
--
http://www.bradheintz.com/
- updated
I have often wondered how biased polls are based on the questions asked, the demographics of the people polled, etc. When results about polls are made public, is it also possible to obtain information about how the poll was conducted in a simple, by request method? Now if the answer to that question is, no, how much can we rely on polls, since we have no way of verifying if the questions asked and the people interviewed were heavily biased to favor one outcome over another? (Such as in the recent large discrepancies of the 8% vs. 24% use of Linux as a server results that we've seen on Slashdot recently).
When People magazine does an issue devoted to "what's hot" in fashion, do they interview Jane Doe from Des Moines, Iowa? No.
So why are OS numbers reported with equal rating? Not all users are equally suited to *choose* an OS, therefore not all users *choices* are equally interesting. I'd really like to see a breakdown of OS by user-type (levels of education, field of degree if applicable, occupation, etc). Keep in mind this applies just as much to business. A technology company presumably put more informed thought into their choice of server than an art supply house or whatever.
--
324006