There is definitely mixed feelings about where OpenFlow is today in terms of real world use cases and what could be done in the future to make it really useful.
Current Use Cases:
1. Hybird port - This seem to have become the most practical use case for enterprises without too much disruption. In this application, the switch data plane primarily operates independently for packet forwarding (traditional protocols). Openflow is then applied to edge ports to make important policy decisions especially around security or services chaining. Several vendors have advanced this approach including HP, Meru Networks and Brocade.
2. SDN WAN - This application uses Openflow to manage the overall WAN connectivity and apply policies to the overall movement of packets. This is very practical as the migration overhead is very low but the benefits are significant. Sonus Networks does have a good solution in this area.
3. Open Source - This has become a very good vehicle to advance the overall SDN agenda. Projects such as ONOS and Opendaylight are creating controllers and applications that could really help this approach longer term. On the switch side Open Networking Linux does have some traction while some vendors are implementing Open vSwitch on physical switches.
4. Overlay - Since there have been limitations with hardware implementations of SDN, several vendors have focused on vSwitch style solutions with the use of encapsulation protocols such as VxLAN and NVGRE to inter-operate with the physical switch network. A good example these solutions would be VMware NSX and Midokura Midonet.
How can we bring some of these solutions to a fully implementable physical network SDN solution?
Areas to Improve:
While the Open networking movement has created new models to deploy SDN and Openflow, there is still some work needed in terms of commercial deployments.
One key feature that is needed in both commercial switches and Controllers is the full implementation of Multiple Table pipelines. On the switch side, the Broadcom OF-DPA SDK is one means to implement things but there is still a lack of support in shipping commercial SDN controllers. Getting this done would solve the major scaling challenges with traditional switches.
The next challenge; how to ensure migration in the most challenging of situations? Examples would be Campus networks and Enterprise Data Centers. The larger vendors will not embrace all the attributes of Open SDN, so in that case existing networks will not have a clear path except for the case of software based overlay solutions. An important note is that Switches could also initiate and terminate overlay tunnels which means the network edge could be fully SDN while the network core remains as legacy.
Openflow and SDN has actually struggled to take hold based on many factors within the industry but there seems to be some good opportunities to move things forward in the next year or two.