Major advances in technology tend to follow a predictable pattern. The initial hype gives way to cold realization, which leads to further development and, ultimately, production-level deployment in which the results rarely live up to the initial hype.
Software-defined networking (SDN) is entering the second stage of this process. The closer enterprise executives get to the technology, the more skeptical they become of claims of insanely dynamic networks and everything/everywhere/immediately application environments. This is not to say that such a world may not be possible someday, but it won’t be quite as easy to get there as many initial boosters have implied.
For one thing, argues Acrux Consulting CEO Glenn Evans, the SDN market is too fractured to even begin thinking about a universal virtual network. The OpenFlow standard has garnered a lot of support, but there are still proprietary interests at work as many vendors seek to provide base-level compatibility and then add unique capabilities of their own. In the business, it's known as "embrace and extend." And too many developers are looking at SDN as a means to solve current networking problems rather than to build an entirely new data infrastructure.
There is also little understanding of what sort of environments are ripe for SDN. Pica8's Steve Garrison hit all the high points of SDN in a recent presentation to the Emerging Technology Summit in San Francisco – network decoupling, dynamic configuration, reduced OpEx and CapEx – but he also pointed out that this will primarily benefit large data organizations. Smaller organizations probably won't see much benefit from SDN, although those that transition to the cloud may – repeat, may – see lower service costs once their provider deploys the technology.
And while we can acknowledge all of the wonderful things SDN could bring to the networking table, let's pause for a moment to consider what it cannot do. LineRate Systems founder John Giacomoni notes that while SDN fabrics are effective at connecting hosts on Layers 2-4 by decoupling the control and data planes, they do nothing for the Application Services Layer and higher (L7+), which frankly is where most of the performance issues lie. What's needed is an SDN Services solution that allows netops and devops to program the network so that all elements of the networking stack are operating under the same set of rules.
As data center architect Joe Onisick points out, too many enterprise executives focus on implementing SDN without first understanding that there are many different flavors of the technology that will perform to varying degrees under different deployment models. And few seem to seriously question what they hope to accomplish under an SDN framework. SDN, after all, encompasses everything from packet delivery and physical/virtual network integration to flow management, scalability and overall network management. Each of these elements will require a deep-dive analysis of current capabilities and future goals if the enterprise is to make a successful transition. And with much of the technology barely out of the lab, it could be a while before complete working architectures are available.
By no means is this intended to imply that SDN is a fraud. It's just that there seems to be a lot of hot air circulating around the SDN market at the moment. And sometimes the best thing for hot air is a little cold water.