(Featured image: Cyan's Blue Planet interface, photobombed by the Gigamon logo.)

Think of this as a follow-up to the "NFV: One Year Later" post. I'd like to take a few minutes to point out some of the network function virtualization (NFV) samples that were being shown at last week's SDN & OpenFlow World Congress.

Less than one year after ETSI convened its Industry Specifications Group (ISG) for NFV, the concept has become ingrained in networking's collective head. On the conference's exhibit floor, NFV was treated as a foregone conclusion, with the only question being: How do we get there?

Well, there was a second question, too: What unforeseen side effects are going to strangle us once we do get there? By definition, some of these can't be foreseen — but it was good to hear the occasional reminder. As analyst Ray Mota of ACG Research said over lunch (referring more to software-defined networking (SDN) in general), "How do we know we're not making things worse?"

The answer might take a while. Most carriers seem to be talking about doing something with NFV in late 2014 or early 2015 — although one service provider is challenging Cyan to get something deployed by the second quarter of 2014, said Joe Cumello, the company's chief marketing officer.

Consultant Tom Nolle of CIMI Corp. was at the Congress demonstrating what his CloudNFV group has come up with. CloudNFV's goal is to create proofs-of-concept for NFV and, more generally, to make sure NFV gets done the right way (by Nolle's definition, which is based on experience not only as a service-provider consultant but also as a coder).

CloudNFV uses XML to define network functions at a high level. Its first application will be an Internet Multimedia Subsystem (IMS) core, provided by Metaswitch through Project Clearwater, its open-source initiative — and Nolle says he's gotten the code to work at a scale of tens of thousands of simulated calls.

The software will get a formal launch and demo in December, at a venue yet decided. It depends partly on whether talks to productize CloudNFV bear fruit; Nolle notes that carriers like the idea but want to buy it rather than build it. "The one operator I was at lunch with said to the vendors as a group: 'Unless we can buy this from somebody, it's a science project.' Hint, hint," Nolle said.

Project Clearwater is also a Metaswitch product in its own right. Metaswitch intends to bring out a hardened, tested release of Project Clearwater — a production-ready VNF — available in the first quarter of 2014, with support available on a subscription basis.

Actual implementations of Clearwater might take a little longer. "The relative maturity of the capabilities of the NFV orchestrator products that are out there are going to be very crucial to our ability to succeed," and they're not quite there yet, said Martin Taylor, Metaswitch's CTO.

Metaswitch thinks its Project Clearwater reflects what carriers want from NFV, particularly when it comes to scale. Scaling is handled by dividing the work into a lot of small virtual machines, all stateless, that vanish when they stop being needed. (To supply the machines with state, Metaswitch uses open-source storage systems — "state stores," Taylor calls them — such as memcached and Facebook's Cassandra.)

NFV is going to require incremental steps, Taylor says, only because it's going to be so difficult to get to the ideal endgame of completely distributed functions. ("If you want to defer the moment you can get any kind of benefit out of NFV, go ahead. Boil the ocean," he said.) In that light, partial NFV instantiations are OK, Taylor thinks. Project Clearwater is one such, as it's not a complete substitute for a mobile carrier's IMS core.

The European Advanced Networking Test Center (EANTC) staged its first proof-of-concept demo at the Congress, showing off virtualized network functions from Huawei, Metaswitch, and Procera. The demo included a portability test, in which functions had to prove they could optimize their locations, and a test of elasticity, where the function had to scale itself up, without losing performance, as emulated subscribers were added.

The significance of being included in the test was that the functions passed a blanket NFV test plan and a vendor-specific test plan, both written by EANTC.

The three showcased technologies were:

• Huawei: An architecture for carrier-grade network address translation (specifically for the IPv4-to-IPv6 transition) and service chains. Both technologies, still in development, target carriers' residential services.
• Metaswitch: Perimeta, a virtualized session border controller.
• Procera: A virtualized version of PacketLogic, the company's deep packet inspection.

Cyan showed off its Blue Planet orchestration environment, manipulating applications supplied by various partners in the vendor's Blue Orbit group. Partly, the demo was meant to emphasize that Blue Planet is vendor-neutral; in fact, as Cyan has stated previously, there are 20 Blue Planet customers that don't use any Cyan equipment.

Vendors deeper in the ecosystem are likely to come up with "VNF helper" types of products. 6WIND showed one such idea, using its software to act as the data plane for a virtual switch. The claim here is that the virtual switch is a crucial bottleneck; it can slow down the performance of virtualized network functions (VNFs).

6WIND's demo showed its software essentially taking over the virtual switch's data plane by intercepting all packets heading that way. The control plane is unaffected, while all the data-plane work gets offloaded onto dedicated cores that aren't busy running things like the Linux kernel.

• NFV: One Year Later, the Real Work Begins
• Cisco Wants SDN to Liquefy the Radio Access Network
• SDN & OpenFlow World Congress: Carriers Plot Their Courses to SDN
• Telefónica’s Set-Top Virtualization Is Almost Ready for Prime Time
• ONF Will Tackle SDN’s Northbound Interface
• SDN & OpenFlow World Congress 2013 – Event Preview