How many ways canst thou measure Linux shipments?
Let me weigh the counts. IDC tells it
IDC VP of server research Dan Kusnetzky has given us a detailed explanation of how IDC arrives at its OS figures, and a glimpse into how the new figures - which should be published Real Soon Now - will look.
One thing's for sure: Dan's not getting drawn into a mud throwing competition with any other market research on the subject, despite provocation. He wouldn't say which market research company, but as he spoke, a swarm of midges rose out of the back of the monitor, and hovering briefly spelled out the word 'GARTNRE GROUP'. These are Register midges, and we quickly removed the typo to spell the name of … well, you can guess the rest.
Gartner pegged Linux shipments at eight per cent a couple of weeks ago, considerably lower than IDC's figures. Which you can recall, prompted Microsoft's PR agency Waggener Edstrom to get in touch with us, agog with excitement. The resulting controversy was enflamed by people "trying to protect their business decision," Kusnetzky felt, and he prefers to explain IDC's methodology rather than diss Gartner's.
"Two research companies asking different people different questions and coming up with similar but related answers," doesn't make for a scrap, he insists.
IDC has made a few interesting tweaks to its own model for the forthcoming figures which directly address the issue of in-house replication of distros, and for the first time too will include downloads from ftp servers in the official tally. These couldn't be included in the final total before, says Kusnetzky, because he couldn't corroborate the website download figures.
On the other hand, IDC has been rigorous about disregarding unusued copies, such as magazine cover mounts, CDs that get bundled with network cards, and the like. IDC is also not counting some full shipments that aren't really being used as full shipments: for example, when it's being used to update a specific library. An new shipment over an existing Linux installation hasn't expanded net usage, of course. Real world demand is the criteria here.
Overall, these include some losses and some gains for Linux enthusiasts. But the metric you're probably waiting for is that roughly 15 copies are replicated and used from each paid-for distro, according to Kusnetzky. And in addition, for every paid OS shipment, too, there's slightly less than one additional free download from the company's ftp server.
Purely on the basis of the replicants and downloads, then, Linux usage could be "significantly higher" than has previously been estimated. We'll have to wait a little while before finding out what the final tally is. IDC subscribers get the figures first.
Out for the count
"Market research should be seen with some skepticism," says Dan disarmingly. IDC's method goes like this. Although the exact methodology is proprietary to IDC, it hasn't changed drastically for twenty years, he says. IDC starts on the supply-side, talking to the vendors. Then, it's out into the field. If usage figures show up a discrepancy, then either the vendor is telling outright lies (not unheard of, says Dan, but major porkies are spotted fairly quickly) or selling at a price significantly different to what they say it is. IDC carries out 300,000 demand surveys to balance the vendor, supply side of the picture.
There's a few other parts of the process which should interest Linux watchers. Kusnetzky says the research tries to separate OS shipments from hardware shipments. Particularly relevant in the case of Linux and the free BSDs, as old machines that have served their purpose are often redeployed as web or file and print servers, or routers.
And Linux too has the unique proposition of wanting to stay hidden for political reasons. Where a Linux or a BSD does the job, but doesn't fit in with corporate IT policy:-
"It's better to get the congratulations for a job well done than the criticism for using the tools, and never tell anyone that it was running on Linux and running PERL that you picked up in that summer class at the University."
In all some 140 distros have been tallied, with several being unknown outside their native country, and more, such as those assembled at a University and sold in the college bookstore, unknown even beyond that campus.
IDC is also to introduce a new category of 'severstation' in its figures, where a server is also being used as a workstation in a small workgroup. Linux usage is split roughly three ways between server, workstation and serverstations.
Fascinating stuff, and for more, check out the Slashdot Q and A.
Sponsored: Beyond the Data Frontier