Has one in every 100 people in Japan downloaded MySQL's database? Figures published by the company this week imply that's the case, but common sense suggests otherwise.
Software companies love to brag about download figures but in reality they can be misleading, analysts and some vendors say. Companies routinely include incomplete downloads in their total, even though the same user is likely to have tried again after an initial failure, and hardly any vendors estimate how many people actually install and use their software.
"Obviously they want to make the numbers as high as possible, so I suspect most people choose the highest number without doing any filtering," said Paul Mutton, an internet services developer at Netcraft, which measures software use on the web.
Free and open-souce software
Developers of free and open-source software have a particular penchant for the claims, in part because they don't have sales figures to measure use of their software. Several vendors at Solutions Linux in Paris this week said they had passed the hallowed milestone.
"Our software has been downloaded more than a million times," said Jorg Janke, founder and president of ERP (enterprise resource planning) company ComPiere, echoing a common refrain.
Vendors don't promote the idea that downloads equals number of users, but they don't dispel it either. Mozilla reported in October: "Adoption numbers have exceeded expectations with more than 100 million downloads since Firefox's introduction in 2004."
The prize for gratuitous marketing goes to Opera Software, whose chief executive promised to swim from Norway to the US last year if Opera's browser was downloaded a million times in four days. Predictably, Opera reached its target, and even more predictably its top executive did not swim the Atlantic.
MySQL's announcement this week was tied to the opening of a Japanese subsidiary. "In 2005, there were more than one million downloads of the MySQL database server by Japanese users," the company said. (It also reported a million downloads for MySQL 5.0 in November.)
Japan has a population of 127 million, including 18 million under age 15. Allowing for infants – not to mention people without computers – that means almost 1 in every 100 people downloaded the software, if each download corresponded to an individual user.
It didn't, of course. The number of successful downloads is typically far lower, and the number of unique users who install and use the software is lower still, Mutton said.
"I've downloaded Firefox 10 or 20 times myself, to get different versions and to put it on different machines, so that's reduced the number of users by a factor of 20 right there. Plus there are people who use it for five minutes and give up," he noted.
A rough guide to downloads
Vendors don't pretend the figures are more than a rough guide to the popularity of their software, and they insist they take care to produce accurate figures. But what the numbers actually represent is unclear.
"There obviously is not a standard for this," said Opera spokesman Tor Odland.
To keep its numbers accurate, Opera only reports downloads from its own server network, not from sites such as Download.com, where it has less control over the numbers, Odland said.
But Opera doesn't know if the same person downloads a program to different machines at home and at work, and it doesn't distinguish between partial and completed downloads. "We estimate that the number of failed/partial downloads are low because we have a tiny application, which is easy to install," Odland said.
MySQL's database is considerably bigger, but it too doesn't distinguish between completed and partial downloads, said MySQL spokesman Steve Curry, in an email response to questions.
"But we take this in to consideration when reporting download numbers," he wrote. "For example, if there were 13 million download requests one year, we would conservatively report this as 'more than 10 million downloads'."
Getting the numbers
Totting up the figures is relatively easy. To get the Japanese figure, MySQL checked the IP addresses of the users hitting its server and counted up the .jp domains. Companies often write scripting programs to add up the totals.
The figures for ComPiere are supplied by SourceForge, which hosts its application downloads.
"As far as we know it is counted per IP address, per day, to prevent hiking up numbers," Janke wrote in an email. "If they download the same file twice daily for a week, it would count seven times (as far as I know)."
Software makers could provide more helpful numbers if they tracked installations and actual usage. A reward scheme run by Google's AdSense division suggests this is possible, at least for some programs. AdSense pays $1 (57p) to referral web sites every time a user clicks through a link that leads to Firefox being downloaded and used.
"They have a way of identifying when users installed it, and also when they actually used it," Mutton said.
In general, the proportion of downloads that lead to installation and usage is probably very small, he said. "It could be as little as ten percent of downloads are actually used," Mutton said, "although that's just my estimate."