You are basically dumbing down the forwarding boxes (read Switches,routers and middle boxes)by retaining only the well-defined data plane and moving the control to a logically centralised controller.
The cost benefits are tremendous if you look at the way forwarding boxes get highly commoditized by using merchant silicon and the control plane benefiting from cheap and increasingly powerful x86 servers.
There are additional benefits by way of much simpler network management as SDN spells the end of complex proprietary and open standards protocols!
"The goal of OpenFlow is to let end users program network systems using servers as controllers. However, the switches and routers will still require some embedded processing to interpret the APIs and carry out their jobs."
Does that mean they want to move the processing power to the servers. I would expect finding ways to eliminate than transfer processing power to be the driver of innovation.
Can someone explain the cost benefits here?
Huge progress has been made at University of Stanford, OpenFlow, and its community since its inception few years ago. It is about time to implement OpenFlow FPGA, ASIC, and SoC at wider scale, starting with servers, storage, and networking in data center, LAN and enteprise networks. Implement OpenFlow in WAN will take time and can be develope in parallel. We must take a bold steps and working hard for future.