SAN JOSE, Calif. — XO found that network functions virtualization (NFV) meant, not only changing the network, but also changing the way the service provider works with vendors. And along the way, XO found a bit of DevOps religion.
What’s interesting is that XO itself is still a rather siloed organization, said Matt Bateman, who told his story during a session at this week’s NFV World Congress.
As senior network architect, he decided the only way to make NFV work was to involve every facet of the company, including networking, IT, and marketing. The team he assembled was small, but it connected departments that don’t always talk to each other.
“We wanted to have a cross-functional team, because we didn’t want to have something that we threw over the wall and have it rejected,” Bateman told me after his session.
NFV: Extremely Green
As Bateman explained during his talk, XO started its NFV planning 18 months ago by doing the usual thing: issuing a request for proposal (RFP) for end-to-end “NFV” and reading through the submissions. That led to the first dose of reality: No vendor could provide everything.
“They were good in some areas and had deficiencies in others, or they went for a service-integrator type of model,” he said. In the latter case, the vendor would rely on partners to fill certain gaps — partners that were tightly knit into the vendor’s ecosystem and didn’t necessarily suit XO’s needs.
But there was a bigger issue. “It’s obvious now, but it wasn’t at the time: The technology was extremely green.” It was too early to expect a production-ready architecture — and those vendors that claimed to have it had catered to specific use cases, he said.
Bateman drew two conclusions. First, the technology around NFV was changing “almost daily,” as he put it. Second, XO was going to have to be its own systems integrator, choosing disparate pieces that got tested in its own lab for its own use cases.
Combined, these factors suggested working in an iterative fashion with vendors inside the lab, ironing out NFV’s wrinkles along the way. It was going to be a lot like software development. XO’s NFV selection process was about to go DevOps.
XO whittled its vendor list down to seven. Bateman set the ground rules: They each got as much time as they needed to prepare for development that would take place entirely in XO’s lab. Then each vendor would get one week in the lab, the vendor’s staff interacting with XO’s, to hammer out software that would specifically work in XO’s environment.
The idea was to complete as much development as possible in one week, rather than undergoing the usual months-long process of email and document swapping. Any roadblocks were pushed aside, to be worked on later.
It was “a very open, very DevOps, agile model,” Bateman said. “I was very surprised how much we got accomplished in that one week.”
This was new for the vendors, too, especially considering they would ultimately have to collaborate with competitors. Questions about intellectual property rights sprang up early in the process. Bateman didn’t detail how that issue was solved, but he noted that all the vendors were willing to cooperate. “That was not a showstopper, and I fully expected it to be.”
XO came out of the process with an NFV software stack and then developed an orchestration layer, using the same one-week fire drill of a process. “Within probably the next couple of months, we’re going to be able to advertise to the world the stack that we settled on.”
The story doesn’t end there, though. XO has no way of knowing whether this process will work a second time. Vendors might not be so agreeable running through this routine again — even though they might have gotten some valuable education out of it, as Bateman believes they did.
Moreover, all XO has at the moment is a software stack. When it comes to operating this new NFV environment, will the company be able to stick to its newfound non-siloed processes?
“We’re starting to talk about what that would look like, but we don’t really know,” Bateman told me.
Adding to this uncertainly is Verizon’s $1.8 billion bid for XO that was announced in February. The deal is expected to close in the first half of 2017.