I've always been curious about facebook scaled their "people you may know feature". Everything I've read suggests that they use contact list information uploaded by other users to introduce connections in the social graph. What I'm curious about is how they compute the intersection of the large sets that result, for a lot of people, in real time.