Does Google’s DIY Server strategy Mess With Market Stats?

The fact that Sun overtook Dell in the server market recently threw my mind into complete confusion given that I’d recently read that Google is the world’s 4th largest maker of computer servers. The story on Zoli’s Blog actually dates back to October 2004, so it could be that Google is now actually third (still a little ahead of Sun, and Dell has now stumbled).

Google has lots of hardware. In October 2004 it had 450,000 servers—not so much a server farm as a server empire—spread over 25 or so locations around the world. The neat thing is that most of them run “one application” in a parallel kind of way. No point in Google buying expensive anything. Better for it to buy commodity servers for the sake of cheap CPUs and create a grid. Better still, buy the boards direct from AMD and hire a few guys with screwdrivers.

Does Google’s DIY approach mess up the server OS figures?

If Google is 3 or 4 or so in server manufacture and all its servers run Linux (and I believe they do), then that’s probably 10–15% of the server market running Linux just by counting Google. I wonder whether IDC counts these figures in? I think not. It never lists Google as a server vendor, for obvious reasons (it isn’t one) and it’s the server vendors that the stats are compiled from.

If this is the case then the Linux server market share figures that are often quoted are wrong by at least 10 percent—and all other figures for UNIX, Windows, OS390, etc are also overstated.