Behind the Scenes of Google Scalability

holy fucking shit. The fridge rack we installed in the Herndon datacenter is there, large as life, in one of the slides, with the entire (circa 2003) hwops team's signatures on it. I have trouble believing that it was in fact the optimal rack design (given the cooling and power that GOOG favors) but I guess if it ain't broke, and has several patents on the genius features that make it easy to service, why 'fix' it...

The 'never seen a petabyte of data, never used a thousand machines' slide was funny. I'm not a particularly strong CS guy, but I happened to work on molecular dynamics right after getting my undergraduate degree in chemistry, and can note for the record that I got allocated 1000 CPUs during a test deployment... which we promptly soaked. Whether the exercise was a useful one is hard to say.

But the genius of Google (and Amazon) is that now it doesn't matter. You just have to find the right model for your code so that it harnesses an arbitrary number of unreliable nodes, and the libraries (and infrastructure) do the rest. It's really amazing how far things have come in 10 years.