I think a lot of the smaller details this article glosses over contribute to FB's ability to run on so few machines. For example, comparing my anecdotes of FB's internal network (they probably have the most organized/mature IPv6 network on the planet) with those of my AWS public cloud experience, I'd reckon that a lot of FB applications would require architectural redesigns to operate in that kind of environment. The…
A naive guess would be that providing services for users you don't control is harder than for users you do control.