PALO ALTO, Calif. -- Carrier Network Virtualization -- With network functions virtualization (NFV) moving out of proof-of-concept trials and labs, carriers need to look beyond the enabling technologies to the new services.
During a panel at this week's Carrier Network Virtualization conference in Palo Alto, Calif., moderator Michael Howard, senior research director of carrier networks at IHS Inc. , summed up the next NFV phase as thinking about virtual network functions (VNFs) as virtual network services.
"I really like how Michael introduced the theme," said Amit Tank, senior principal cloud architect at AT&T Inc. (NYSE: T), during the panel. "When you think about something from a service point of view, you have to ask some key questions."
Based on AT&T's Domain 2.0 effort, Tank provided a list of NFV, SDN and cloud-based challenges for the industry during his panel presentation. While capex and opex are often cited by vendors as the key reasons for tackling virtualization, Tank said the focus should be elsewhere.
"Yes, capex and opex hold big promise but it's not going to start raining money as soon as you start doing this," he said. "A technology provider may try to sell you this promise, but the thing is you have this one box and now you are trading it for a complex ecosystem of players. I don't focus too much on the savings. What gets me really excited is this vendor-agnostic architecture that this vision unlocks."
Tank advised service providers that are embarking on their SDN, NFV and cloud journeys to have their open source and software stacks in place to enable speed of innovation.
"Speed of innovation is the ability to leverage something that has already been created in open source," he said. "That is simply one of the greatest goals that you can accomplish. Do you want to be in the position where you can leverage that right away, or do you want to be left out?"
Tank also said that a cloud strategy was vital for a service provider's NFV success, and that carriers needed to take a pragmatic view of what can be virtualized and where. He said Mesos and Kubernetes were two communities that were making great strides.
"There are some prominent things playing out right now, like vCPE," Tank said. "When Docker and rkt become mature enough to be orchestrated to fully, securely run VNFs, that's going to be an interesting development."
Other challenges outlined by Tank included East/West and North/South packet throughput, service placement and policy management and DevOps. Service providers' DNA doesn't include software, which makes DevOps harder than it looks, according to Tank.
"Things like DevOps are extremely important to your strategy but they are also part of the challenges," he said. "DevOps is easier said than done. DevOps for a simple deployment software tool, it takes a lot of work to get there, let alone for VNFs."
For Tank, the biggest challenges service providers face in moving to a virtualized, cloud-based software world wasn't a service or technology; it's the people service providers need to employ while instituting new cultures.
"You're going to have to hire new people, new talent and train them so they are actually able to do this, or sometimes pay off technology providers to help you achieve certain solutions," he said. "It's a lot about culture as well."
During the same panel, Stephan Massalt, vice president, Swisscom Cloud Labs, advocated pushing beyond the current thinking for NFV. After attending the conference for the first two days, Massalt said "that a lot of this stuff is about doing what we've been doing for the last 20 years and putting that into software."
"Are we preparing for the previous war?" he asked. "We need to think of cloud-native IT instead of telco."
For example, service providers need to "design to fail" instead of using the infrastructure for "five nines" reliability. Massalt cited Netflix's "chaos monkey," which was also referenced a day earlier at the conference. Netflix created its chaos monkey programs to disable certain pieces of equipment on Netflix's networks so its developers could learn how to write programs that will work in those environments. (See NFV: Security Issue or Security Opportunity? )
"Chaos monkey is designed to fail," Massalt said. "If something goes down, the system recovers itself. There's nothing wrong with that. You might see a glitch in the movie you're watching and that's it. Instead of design for worse case, design to fail."
Instituting "reliability by scale" means having a lot of services that control small functions instead of having a big box that does everything. "So it's about having horizontal scaling instead of vertical scaling," Massalt said.
Other suggestions by Massalt included focusing on Continuous Integration /Continuous Deployment (CI-CD) instead of certification programs, deploying IT clouds with open source, standardizing open APIs and using microservices instead of virtual appliances.
"We're still talking about virtual appliances," he said. "With microservices, I can distribute my application where it needs to be by decoupling control planes from data processing planes."
— Mike Robuck, Editor, Telco Transformation