Ritual sacrifices, mostly. Just kidding ;) For the “faster” and “less disk space” parts, we use a lot of standard techniques like bitpacking, compactionless b-trees and other things that have been designed in the past 10 years — we’ll blog more about that stuff very soon. The “understandable code” part can probably be explained by the fact we focused on that extensively during implementation: we did lot of API revisions, gathered lots of feedback in private beta — and we still have a long way to go!