SDN has lots of promise but I don't think the chosen analogy with the switch network topology recovering on link failure being a real benefit. In this simple example of centralised control plane (brain), if the switches lose connectivity to the central control plane then traffic between switches will never re-converge. Even when the switches maintain connectivity to the centralised the control plane, depending on the physical separation between the switches and the centralised servers, network re-convergence time can be significantly increased.
When each switch runs its own control plane, it is not sharing fate with anyone. Each switch will build its own map. Re-convergence time can take longer than having a centralised control plane in close proximity but at least it will recover.
The centralised control plane looks really great in simple tests but the real world is never so simple. Take for example the case where an A380 engine exploded on a Qantas flight spraying the wing area with shrapnel. The central site in France, monitoring the performance data from the aircraft in real time, was inundated with hundreds of persistent and reoccurring alarms making it extremely difficult to determine the coarse of action required to return the aircraft to safely. A true credit to the skill of the pilots (local brains) working in conjunction with the human operations team in France (centralised brain).
What is really needed is distributed (local) command and control with centralised oversight.