To understand how Facebook's search works, you first have to understand what the "graph" is. The graph is a database that stores information about the users, pages, and other objects within the Facebook universe. It also includes the relationships between them. Each entity, or "node," within the Facebook graph—identified by a unique number called a fbid (Facebook ID)—has a set of attributes, or metadata, associated with it. The relationships between these nodes, called "edges," contain their own metadata to describe the type of relationship between them.

The graph database used by Facebook is quite similar to Google's Knowledge Graph and Microsoft's Satori graph-based repository. But in many ways, the structure of Facebook's graph is simpler than Google's and Microsoft's graph schemas, because Facebook has tuned the metadata for its nodes and edges specifically for social interaction—not to store product SKUs or how many times a particular actor has been to rehab. The Facebook Graph may not be able to answer those questions, but it contains data about other things that are useful. You can learn what entities are close to a certain location, liked by certain people, or otherwise tethered to a user through the social network's path of edges.

The problem is that there are hundreds of billions of entities in the Facebook Graph, with trillions of relationships connecting them and trillions of attributes. Sankar said that the attributes of Facebook's photos alone number in the trillions. So how do you index and search something that big?