I read two competing articles, both published on the same day, about the use of SDN in the WAN. The first is a solution brief by the ONF. It describes an architecture for cloud-backbone-providers to interconnect remote datacenters. While there this seems like a good solution with a lot of benefits, the requirement is that the provider owns it’s own WAN. When there is public WAN involved you have to fall back to tunneling your SDN, i.e. building an overlay network.
The [second])http://blogs.cisco.com/borderless/enterprise-networks-practical-differences-in-lan-and-wan-sdn-deployments/) post is by Cisco. The geographical nature of the WAN provides challenges for a centralized controller. Like I did, the author also points to such challenges in the past with ATM. Not with the amount of programmability, but with resiliency and failure in the network.
the distributed nature of WAN networks requires a distributed control plane
This does not only a apply for a WAN under single control, but especially for the public WAN, controlled by multiple providers.
I think this emphasizes the conclusion that SDN has the most advantages in the datacenter. When the LAN grows into a WAN your mileage may vary.