It's not clear to me that SDN is to replace GMPLS, or vice versa. They seem to serve different purposes. First, transport network has a pretty broad definition, including optical, packet etc. Second, transport networks are mostly operated in isolation with some use GMPLS, some use centralized NMS, etc. So it's far from saying multi-network/multi-vendor interop. Most of all, GMPLS (and many other TE-based protocols) is to optimized traffic transport within one particular network.

On the other hand, SDN is to enable services on top of the underlying networks, regardless how each individual network is configured and managed.

For instance, we can imagine an OTT video services that is to connect LTE users to the data center servers over multiple Ethernet, IP and optical networks. SDN would be used here to provision and control bandwidth "links" at application level, while each network may run GMPLS or whatever.

Hope this is making sense.

Here is the thing: this is all driven by the new services and applications. If the operators deploy a network from a single vendor (a common practice in many transport networks), multi-vendor standardization is nice but secondary. On the other hand, if the operator is to deploy a new service from data centers over multiple networks (and vendor equipment), it becomes essential to produce a common and abstract interface, at the same time, we need to hide the unnecessary underlying network complexity. So it’s to all vendors’ benefit to work together in this area.

Hahaha! Very cute! But seriously, IMHO, in years to come, the money is to be made from the applications (sitting north of SDN Controller); the network intelligence is from IP and routing protocols (not sure where it sits though :-)); the data transmission efficiency is from the transport gears that continue to innovate in pace with Moore's Law. There are many ways to program network links. OpenFlow is one of the standardized ways in doing so, and requires enhancement to support transport networks. No point to be religious and no point to resist the unavoidable. Happy holiday!

IMHO, the SDN architecture we are working on is suitable for many types of transport networks. First, as we all have experienced, one of the key requirements in backhaul and metro is in operation simplicity, where the operators can do point-click to provision the Ethernet and WDM links. Similarly, the key in long-haul backbone is to control the traffic aggregate into those well-provisioned and expensive trunks. In both scenarios, we can use SDN as the glue to bring routing, network planning and policy controls into the operation. IMHO, SDN is not about having the cheapest dumb boxes. Rather, the value is in reducing the operation cost.