SDN aims to make networks easy to configure in software rather than hardware. The goal is to provide data centers with flexibility on the level associated with cloud computing.
9 Bandwidth Hogs: Reality Vs. Myth
(click image for larger view and for slideshow)
Software-defined networking (SDN) has been in the "coming soon" category for many years, but an Interop keynote panel discussion on the topic showed room for debate over what it ought to look like when it finally gets here.
SDN is too often spoken of as a single event that will wipe away all current networking technologies, when in fact "the underpinnings are already in place," said moderator Eric Hanselman, chief analyst at 451 Research.
The point of SDN is to make networks easy to configure and reconfigure in software rather than hardware, with many more networking functions migrating from being embedded capabilities of a network appliance to being defined in software. Network systems are migrating incrementally in that direction as networks follow the same path toward virtualization as servers and storage, he said. Ultimately, the goal is to provide every data center with the flexibility associated with cloud computing.
Huge data centers operated by the likes of Google and Facebook "already look like what we attribute to SDN," agreed Martin Casado, chief architect for networking at VMware. "The big guys did it first because they're big," he said, not because they're the only ones who would value the flexibility promised by SDN.
Microsoft has been learning the same lessons through cloud services such as Azure, said Rajeev Nagar, group program manager for Windows core networking. "Scale changes everything," he said, adding that "when you're provisioning and de-provisioning thousands of networks a day," it's impossible to do that all manually, making automation essential.
The third member of the panel, Rajiv Ramaswami, executive VP and general manager of infrastructure and networking at Broadcom, sees a need for "faster and flatter networks" to support the fast-growing demand for machine-to-machine communications. Networking chip designers like his firm are at "the bottom of the food chain" in an ecosystem that will be increasingly stratified, with more hardware capabilities made accessible for manipulation through software, he said.
Just how much access software applications should have to basic network functions proved to be one area for debate, with Casado arguing, "the less the application has to know about the network, the better it is for everyone." Although system software for networking can make networks more flexible, application design is simpler when it is protected from the complexities of the network, he said.
Microsoft's Nagar said that might be true in general but applications such as Lync can get real performance benefits from the ability to dynamically reconfigure the network. Big-data analytics applications also could benefit, given their requirements for large-scale data movement, he said. For that reason, he argued in favor of providing "deep visibility" into the network infrastructure where needed.
Of the three, Casado was most willing to admit he is uncertain how SDNs will develop and how they will affect the IT organizations that must support them. He believes the decoupling of logical switching functions from hardware is inevitable, following a pervasive pattern in IT of systems architectures becoming less monolithic so that individual elements can evolve independently. "Anything after that is anybody's guess," he said.