How Graph databases grow to operate at web-scale

The main issue that i am trying to understand that is how a connected graph can be partitioned and sharded to scale horizontally. I have been looking at facebooks TAO db architecture. TAO — Facebook’s Distributed database for Social Graph | by Ameya | Coinmonks | Medium .,https://www.usenix.org/system/files/conference/atc13/atc13-bronson.pdf . It has master-slave rdbms backed data storage with layers of in-memory caches.

stack-ofthis seems to be a good resource to answer the question, if someone has specific insights , please share