Recently I've stumbled upon this paper, where FB presents Robotron, a system for managing a massive production network in a top-down way. Design goals of Robotron has been/is to reduce effort and errors on management tasks by minimizing direct human intervention.
FB infrastructure is a "network of networks" containing multiple domains: edge point-of-presence (POP) clusters, a global back-bone, and several large Data Centers (DC). Interestingly the paper describes the network-wide abstraction layer that models and stores various network device attributes as well as network-level attributes and topology descriptions, e.g., routers, switches, optical devices, protocol parameters, topologies, etc. Physical and logical components are modeled as typed objects, value elds, and relationship fields. And there are APIs to provide operations to retrieve objects and their attributes.
Robotron looks like a very promising approach that the research should look at in order to improve/extend the management practice for SDN-NFV convergent infrastructures.