Network functions virtualization (NFV) is often associated with general-purpose chips from the likes of ARM Ltd. or Intel Corp., but network processor vendors believe they'll have a place in the NFV revolution as well.
They were preceded by Fred Feisullin, who's part of the CTO group at Sprint Nextel Corp. but who was talking mainly for the European Telecommunications Standards Institute (ETSI), the organization that's helping carriers develop NFV.
ETSI really could use chip-level help with NFV, Feisullin said. A library of tools to help build the virtualized network functions (VNFs) would be useful. And some help in using the multiple cores of a processor would be handy too, he said; multicore processors have a reputation for being difficult to program.
Part of the impetus for NFV comes from carriers coveting the agility of players such as Google and Facebook, Feisullin said. Carriers would love to be able to junk equipment as it gets older, or more easily upgrade it to handle new functions.
"Google doesn't have standards-based equipment, and they don't have a lot of vendor-specific things in their hardware," Feisullin said.
It's a contrast to the traditional carrier model, where equipment was supposed to stay in place for a decade or more.
Capex tends to be a focus when people talk about NFV, since the idea is to replace that carrier equipment with general-purpose hardware that's more expendable. But it's the opex side where NFV's savings are most promising, Feisullin said.
"Carriers are willing to spend the capex if they can save the opex," he said. "The next two to four years will see a dramatic increase in the efficiency of carrier networks due to network functions virtualization."
As for the types of equipment affected, it's not in routers and switches but in the Layer 4 through 7 equipment that NFV is going to make its mark, said Nabil Damouny, a keynote speaker from Netronome. "NFV's not going to virtualize a Layer 2 switch," he said.
Netronome's point was that its chips can act as an offload for the general-purpose processors that will drive NFV gear. The Netronome Network Flow Processor (NFP) would be an acceleration engine for certain functions. Netronome tried this with Open vSwitch, an open-source data-plane switch; moving some Open vSwitch functions to an NFP boosted performance by a factor of 10, said Nick Tausanovitch, Netronome's director of solutions architecture.
EZchip, by contrast, envisions its chips doing Layer 2 through 7 work, taking over swaths of the data plane. Sandeep Shah, director of systems architecture, ran through several examples including security and traffic management.
— Craig Matsumoto, Managing Editor, Light Reading