SAN JOSE, Calif. — NFV was introduced as a concept more than six years ago, but the wireless telecommunications industry is caught in a “paradox” that is precluding it from reaching a critical mass, said Scott Shenker, professor of computer science at U.C. Berkeley, in a keynote at this week’s Open Networking Summit.
“We have to admit that we have a very serious problem. The rate of adoption and innovation is much slower than anybody thought six years ago,” he said.
The problem boils down to integration, or rather a lack thereof, Shenker explained. Because the three components of NFV — virtualized network functions (VNFs), NFV management and orchestration, and NFV infrastructure (NFVI) — are so tightly combined, it has created a paradox that makes it unnecessarily complex and difficult for operators to implement the concept at scale.
“Individual components by themselves are not that hard,” Shenker said, adding that the decision to combine various components and push NFV as a single solution makes deployment hard because operators have to fundamentally change their infrastructure. “These three components are now so tightly woven together that it’s hard to innovate,” he explained.
Integration Begets Innovation
Integration challenges have to be solved before innovation occurs, Shenker noted, adding that “if you have solved the integration problem then your components can come from different vendors.”
What the industry needs is a more nimble approach to NFV, hence the creation of Lean NFV, a concept Shenker and others introduced on stage and in a new white paper. “Current NFV efforts are drowning in a quicksand of complexity that arises in several crucial aspects of NFV operations,” the paper touts.
“Current solutions are complex to deploy because they are too closely coupled to how the rest of the infrastructure is managed; they are complex to automate because so many components must be coordinated; and the process of onboarding new VNFs is extraordinarily time-consuming because VNF developers were never given clear and practical guidelines for designing VNFs that can gracefully co-exist with general NFV management solutions,” co-signers of the Lean NFV movement wrote.
Lean NFV is calling for multiple changes, including a universal integration mechanism, a decoupling of NFV’s core components, and removing the need for specialized virtualized infrastructure managers. Doing so will leave more of the design of NFV open to innovation and empower it as an open, extensible architecture, Shenker explained.
Sylvia Ratnasamy, associate professor of computer science at U.C. Berkeley and a co-signer of the Lean NFV concept, said NFV needs a key value store that stores keys with associated values and allows different components to discover and exchange state.
Key value-based integration will make VNF lean and fulfill the promise of putting an end to vendor lock-in, she said, adding that avoiding lock-in on network management is equally as important as removing lock-in from network equipment vendors.
“The essence of integration is allowing different components to discover and exchange state,” Ratnasamy said. “A key value store allows vendors to evolve in a way that’s incremental and independent of each other.”
NFV should rely on common infrastructure management features and operate like any other workload running in a data center, she proposed. “They should all look the same from the point of view of the infrastructure.”