SAN JOSE, Calif. — Open Networking Summit North America — Instead of being a panacea for the telecom industry, network functions virtualization (NFV) has led to a deepening maelstrom of complexities, according to three of the principles behind a new concept called "Lean NFV."
Lean NFV came out of cloaking mode to take the center stage Wednesday at the Open Networking Summit in San Jose. Scott Shenker, professor of computer science at University of California, Berkeley, and one of the founders of Lean NFV, said the integration of certain components of NFV have led to a morass for vendors and service providers as they strive to make NFV work.
Lean NFV breaks traditional NFV, which originated in a ETSI white paper six years, down into three components. The first component is the NFV manager, which handles common lifecycle management chores for virtual network functions (VNFs) and end-to-end NFV service chains.
The second component is the computational infrastructure, which includes the compute resource that can be either bare metal or a virtual fabric. Bare metal can be managed by a compute controller such as OpenStack, while virtual fabrics can be managed by an SDN controller from the likes of OpenDaylight.
The third component is VNFs, which can include both data plane and control plane components. The VNFs also can have an elemental management system (EMS) to handle VNF-specific configurations, according to a Lean NFV white paper.
"So, why has this (NFV) been so hard for the community? We realized that the problem was that they were focusing on the components, which weren't all that difficult, but it connected them in very complicated ways," Shenker said in an interview with FierceTelecom. "They made two mistakes in particular. One is that they embedded the NFV logic into the infrastructure manager. The second one is they defined all these pairwise APIs between the components. If you want introduce a new feature, you had to go and muck around with those pairwise APIs, and that made it hard to deploy because you had to change the infrastructure. It made it very hard to onboard VNFs because how do you build a VNF that can integrate with this?"
He added, "And, they made it almost impossible to innovate on because these components were now very tightly tied together. It was very hard to pull one out and stick something else in."
There have been other industry initiatives to simplify or improve NFV, including LF Neworking's announcement yesterday of its OPNFV Verification Program (OVP), which is the first iteration of a VNF compliance program that will grow to include NFVi.
To solve some if its orchestration and automation issues, AT&T developed ECOMP, which included 8 million lines of code, and later put most of ECOMP in the Linux Foundation to help create ONAP. In parallel, ETSI has developed its own Open Source MANO architecture and code.
All of which, according to Lean NFV, made NFV more complicated. Service providers and vendors were caught up in almost endless developmental cycles to make NFV work, but instead NFV made things even more complex.
Lean NFV proposes adding a fourth element to NFV, which is a key-value (KV) store that serves as a universal focal point for integration. Shenker credited Sylvia Ratnasamy, associate professor of computer science at U.C. Berkeley and a co-signer of the Lean NFV concept, as the originator of the KV store.
For Lean NFV, it's all about integration, integration, integration.
"The components themselves are pretty simple, but you think what's the minimum we need to do so that we can integrate these pieces together? There are three steps," said Shenker, who was one of the co-founders of Nicira before it was sold to VMware. "One is use universal integration and then compose a key value store, so then the way you communicate is through talking to a common key value store. Two, you do not allow people or at least you don't start by requiring the infrastructure, to know anything about NFV, so you don't have to change that, and the third is having the wisdom to know you can stop.
"That's really all you need to do to get things to be interoperable, to achieve integration. If you do that, then you solve the deployment problems, and you leave the rest of the system open to innovation. The key insight is that you make it easy to deploy now, which means we can solve the problems today. Because you've defined so little, you're open to innovation. We figured out that the key part is to get the integration right, and leave everything to innovation. That's sort of the kernel of the story."
Ratnasamy said in an interview with FierceTelecom after the keynote address that she and Shenker had been working on the NFV management integration element for about six years.
"We were actually working on what I would say was NFV before it was called NFV back in the lab in our research groups at the university," Ratnasamy said. "Then as we saw some of the first (NFV) solutions emerging we were like 'Why is that so complicated? Why is ours not?' And I think that helped us crystallize what it (Lean NFV) is."
In addition to solving service provider headaches related to NVF, Ratnasamy said vendors liked the concept as well.
"They look at this and they say 'Okay, this we can get behind,'" she said. "Because it starts out potentially as a small change, and then you can code the changes you're making. We're not asking vendors to go in and rewrite the code that they've written for 20, 25 years. So, that has certainly resonated very strongly with them."
Shenker and Ratnasamy brought Constantine Polychronopoulos, vice president and CTO of VMware's NFV/Telco unit, into the Lean NFV fold. Polychronopoulos said Lean NFV could be a savior for mobile edge computing (MEC) by using the KV store in a hierarchical manner to lower latencies for things like gaming in 5G.
Not everyone is a fan of Lean NFV, including Tom Nolle, president of CIMI Corp. In his blogs, Nolle long has argued that NFV was flawed pretty much since its inception.
"If you look at their white paper, as I have, what you see is an attempt to resolve the serious issues with NFV by framing the wrong concept in a different way," Nolle said in an email to FierceTelecom. "Why are we still inventing things like 'managers' instead of saying that VNFs are like application components and should be deployed using cloud tools like Kubernetes? This is my problem. You can't solve a problem by redefining your mistakes. You have to correct them."
Polychronopoulos, Shenker and Ratnasamy said the next stage for Lean NFV is to get it into labs, proof-of-concepts and live trials to prove that Lean NFV is the answer to the current spate of problems related to NFV.
Aside from those three, additional signatories on the Lean NFV concept and white paper included: Ravi Srivatsav, formerly with NTT; James Feger, currently with F5 Networks and formerly with CenturyLink; Nagesh Nandiraju Comcast; Krish Prabhu, formerly with AT&T; Uri Elzur, Intel; Mirko Voltolini, Colt Technology Services, and Christian Martin, Arista.
"I think we were a bit surprised that most of the people we talked to were quite interested in this approach," Shenker said. "Some people didn't sign the paper because their corporate policy didn't let them sign on as an individual.
"I think that it has to go into production. It has to go into widespread use, but the people who've had experience with this, this seemed to sort of cross the boundaries with them. We've all done system design in many different areas for a long time and one of the keys if you want a system to be long-lasting is to get the modularity right, and this is getting the modularity right."