On: Analysis of Google Caffeine.

Wouldn't the cluster have a certain minimum size to hold the index data? So it's not going to start at one server and scale up, it's going to start at hundreds.

But of course, I could be completely wrong, not knowing anything about how Google works.