"The social graph is tightly interconnected; it is not possible to group users so that cross-partition requests are rare. This means that each TAO follower must be local to a tier of databases holding a complete multi-petabyte copy of the social graph. It would be prohibitively expensive to provide full replicas in every data center. Our solution to this problem is to choose data center locations that are clustered into only a few regions, where the intra-region latency is small (typically less than 1 millisecond). It is then sufficient to store one complete copy of the social graph per region. Figure 2 shows the overall architecture of the master/slave TAO system."
https://www.usenix.org/system/files/conference/atc13/atc13-b...
edit: Oh hah, didn't realize parent was former DB person at FB. I was too, just a few years before :-)