Sorry not to be replying to any particular post, but the sheer volume makes that a little difficult to manage.
It was good to see that, after a relatively brief spate of misdirected criticisms of our survey as being tainted by pro-Microsoft 'bias,' many contributors here saw that the data itself is pretty uncontroversial (and in fact easily reproducible), and instead began to address themselves to the questions that the survey was intended to raise -- namely, questions about what is an appropriate sampling methodology when attempting to measure HTTP server 'market share.'
Those are the sorts of conversations we were hoping to start, and it's good to see them under way here with such vigor.
Just to be clear: We have no real objection to the Netcraft results per se -- only to their being marketed as an unambiguously accurate picture of something called 'Web server market share.' We simply think that sampling this market is a more complicated affair than the endless recitation of the most commonly-sited Netcraft numbers would suggest.
A number of the contributors here who grant the legitimacy of our criticisms of Netcraft's methodology have raised the point that a sample based on Fortune 1000 sites isn't necessarily a good proxy for Web server market share either. (Since some of these sites are nothing more than glorified brochureware, and so on.) I think that's entirely correct.
In a sense, our survey simply sets one type of partial snapshot, with its own kind of built-in sampling bias, alongside another. But then our aim wasn't to be definitive. It was simply to remove the halo of definitiveness from the Netcraft survey -- and to get people thinking about what it would take to be definitive in this context.
And as I say, some of that thinking is on display here. Folks like ChaosDiscord are almost certainly right to suggest that it would be more accurate (or interesting) to sample the server choices of high-traffic sites. We hope to cover some of this territory in future surveys.
Thanks to all those who looked past the fact that we happen to make commercial software for IIS, and actually engaged with our survey's findings and implications. And happy Thanksgiving to one and all.
Slashdot Mirror
Sorry not to be replying to any particular post, but the sheer volume makes that a little difficult to manage.
It was good to see that, after a relatively brief spate of misdirected criticisms of our survey as being tainted by pro-Microsoft 'bias,' many contributors here saw that the data itself is pretty uncontroversial (and in fact easily reproducible), and instead began to address themselves to the questions that the survey was intended to raise -- namely, questions about what is an appropriate sampling methodology when attempting to measure HTTP server 'market share.'
Those are the sorts of conversations we were hoping to start, and it's good to see them under way here with such vigor.
Just to be clear: We have no real objection to the Netcraft results per se -- only to their being marketed as an unambiguously accurate picture of something called 'Web server market share.' We simply think that sampling this market is a more complicated affair than the endless recitation of the most commonly-sited Netcraft numbers would suggest.
A number of the contributors here who grant the legitimacy of our criticisms of Netcraft's methodology have raised the point that a sample based on Fortune 1000 sites isn't necessarily a good proxy for Web server market share either. (Since some of these sites are nothing more than glorified brochureware, and so on.) I think that's entirely correct.
In a sense, our survey simply sets one type of partial snapshot, with its own kind of built-in sampling bias, alongside another. But then our aim wasn't to be definitive. It was simply to remove the halo of definitiveness from the Netcraft survey -- and to get people thinking about what it would take to be definitive in this context.
And as I say, some of that thinking is on display here. Folks like ChaosDiscord are almost certainly right to suggest that it would be more accurate (or interesting) to sample the server choices of high-traffic sites. We hope to cover some of this territory in future surveys.
Thanks to all those who looked past the fact that we happen to make commercial software for IIS, and actually engaged with our survey's findings and implications. And happy Thanksgiving to one and all.
Joe
Port80 Software