Alcatel-Lucent CTO States the Case for NFV

News Analysis Craig Matsumoto, Editor-in-Chief, Light Reading 4/3/2013

SANTA CLARA, Calif. -- The industry needs hardware virtualization, and while network functions virtualization (NFV) might not be the answer, it's a good start, Alcatel-Lucent's CTO said here yesterday.

Marcus Weldon was speaking as part of the launch of Nuage Networks, the AlcaLu spin-in focusing on software-defined networking (SDN). (See Alcatel-Lucent Spins Up Its SDN.)

The answer is certainly not ATCA, which is "the last attempt of the industry to standardize and has largely failed," Weldon said.

What's wrong with ATCA is that it's a box mentality, Weldon said. A particular function is scaled up by adding the exact card for that function, and when you run out of room in the box, you've got to install and configure another box.

What NFV promises is freedom from boxes, in a sense. You'd still need equipment, but a function could grow by being activated in as many instances as necessary. "That's the problem. We're running at box scaling rules and we have to get to Web scale," Weldon said.

NFV isn't necessarily going to save the world. In fact, Weldon took a couple of jabs at the group: "They're not going to standardize anything; they're just going to point to standards," he said. "They're united, but they're not sure what they're united around'

Fair enough. The NFV has only met a couple of times, just long enough to establish that they exist and determine how well the members get along. The group's next meeting is in three weeks in Santa Clara, at Huawei Technologies Co. Ltd.'s site, which added a touch of timeliness to Weldon's talk.

Virtualization is important because it's happening on the data-center side, and telecom, for now, can't possibly keep up, Weldon said.

He gave the example of WebRTC, open-source software that allows for real-time communications on the Web. Apps using WebRTC could scale up rapidly at peak times -- if everybody ran to grab a particular Web video promoted in a Super Bowl ad, for instance -- but on the telecom side, IMS wouldn't be able to respond in kind. IMS is still bound by telecom systems that can't be expanded so quickly.

NFV sometimes gets described as an attempt to bring cheaper hardware into telecom. But the real benefits are in scaling, not penny-pinching, Weldon said. By escaping that box paradigm and working on a virtualized basis, operators would be able to deploy functions more quickly, and scale them up and down at will, he added.

Of course, cost matters, too. NFV could result in some telecom equipment moving to virtual form running on general-purpose processors. But it wouldn't happen to everything, Weldon argued.

Anything coming in contact with the physical layer is probably not do-able on a general-purpose processor -- that would include coherent optics, the radio access network and DSL, Weldon said. He also argued that core routers won't virtualize, as the number of processors needed to replace a core-router ASIC would be just ridiculous.

Virtualization comes with risk, as Weldon mentioned repeatedly during his talk. Server virtualization is based on the assumption that server failures are no big deal; that's fingernails on a chalkboard to the telco five-nines mentality. Changes to OSSs might be required as well.

Then there's the telecom culture, Weldon said. Network operators would have to work hand-in-hand with IT, "and that's not a comfortable conversation or an easy one."

For more

— Craig Matsumoto, Managing Editor, Light Reading