Web Servers: Don't Count Apache Out
There's been discussion in the open source world about a decline in the popularity of the dominant Apache web server. These concerned are fueled largely by the Netcraft survey of the internet, which shows a 20% decline in Apache's market share over the last three years, from a high near 70% to the 50% or so it enjoys today. But bearing in mind the old saw about lies and statistics, it's worth digging a bit more to see what these numbers mean.
Looking at Netcraft's figures more closely, it becomes apparent that this is more than an "Apache vs. IIS" battle. One significant change for the Apache numbers happened in mid-2007, when the 8 million or so Google-served sites started reporting themselves as "GFE" rather than Apache - even though GFE may well be a private fork of the Apache codebase. That's a significan chunk of the 158 million sites that Netcraft surveyed right there.
Although most of the rest of the Apache decline appears to correspond exactly to a rise in IIS, it's not clear how significant this is for the average personal or corporate site. For example, Microsoft picked up 5% of the market in one fell swoop in early 2006 when Go Daddy moved all of their millions of parked (i.e., not active by any sensible definition) domains from Apache servers to IIS. A win for Microsoft, to be sure, but not an indicator of any particular trend.
Web monitoring company Pingdom has recently looked at this issue from another angle: they asked which web servers run the most important sites, rather than just counting noses. By checking headers from the 100 most popular web sites (according to Alexa), they came up with a different set of numbers: 49% Apache, 20% IIS, 4% lighttpd, and 27% other/unknown. By that measure, Apache is still over twice as popular as IIS, despite Microsoft's relationships with the sort of enterprises that generally own the most popular sites.
There's another thing evident from the Pingdom statistics as well: the choice of web server software carries with it a good deal of inertia. Out of the 49 Apache sites, for example, there are still more announcing themselves as Apache 1.3 than Apache 2.0. On the IIS side of the house, IIS 6.0, and even 5.0, far outnumber the recently-released IIS 7.0.
I'd love to see Pingdom's numbers extended in two ways: covering a few more orders of magnitude (say, the top 10,000 sites) and resampled over time (the way that Netcraft works). I suspect that would tell us more about server use that really matters. Meanwhile, it doesn't look like Apache is in any danger of losing its dominance among the top dog sites any time soon.
Do you think Apache's dominance holds up across the top 10,000 sites?