Bryan Larish, chief IT and networks architect at Verizon, took a different approach: “If I could go back in time, what I would have told myself?”
More coverage from future:net:
For Thomson Reuters, Cloud Scale Isn’t Enough
A Rogue Effort Brings SDDC to the Lowly Branch Office
He’s worked on cloud transitions before, both at Verizon and at a previous job. (His LinkedIn profile lists an undisclosed government agency as his previous employer). Having joined Verizon early this year, he’s been part of the team implementing the carrier’s software-defined networking (SDN) and network functions virtualization (NFV) plans.
Those plans have involved the use of the cloud and the implementation of leaf/spine networks, as is usually the case. What Larish wanted to share, though, were the implications this has on the company’s operations. Knowing what he knows now, he had many recommendations for his past self. Here are a few highlights.
Larish referred to this as “preparing the runway.” NFV and the cloud demand some changes in the way a carrier operates, but the new stuff isn’t that far removed from what they were doing before, he contended.
The message to convey to operations is: “We’re trying to make things better. It’s going to look a little different, but it’s not that bad,” he said.
Sometimes, people are the biggest obstacle to something new — as Columbia Sportswear learned in its software-defined adventure, described earlier in the conference. But in a big company, processes aren’t exactly easy to change, either.
Take the concept of servers. A business like Verizon has an “entire machinery” of processes for buying servers. But that’s not the right kind of thinking when you’re moving to a virtualized world.
“When you’re talking about combining everything onto the same physical set of hardware, at that scale you have to start thinking about racks,” Larish said.
The concept of reliability — crucial to any telecom carrier — needs revisiting as well. Verizon didn’t realize this at first.
“As you talk through this reliability issue, you need to get the focus off of hardware and onto the software,” Larish said.
Having two or three of every box isn’t as important as it was in the old telco days, he noted. Neither is capacity — in fact, one benefit of moving to an OpenStack model is that it’s OK to over-provision the hardware, he said. This is new thinking for a telco.
And then there’s everybody’s favorite bureaucratic game: Who owns this? A virtual switch is a networking function, but in real life, it’s a server. Eventually, something is going to go wrong, and “you have some interesting conversations about who actually owns that,” Larish said. “These kinds of things have to be planned out in advance.”
The term “layers of abstraction” comes up a lot in networking these days. Often, it involves hiding a technology’s complexity from users — so they can use APIs to “program” a device, rather than actually knowing how to program it, for instance.
But “the abstractions, as they exist today, don’t necessarily talk to each other well when it comes to troubleshooting,” Larish said.
This is a case where Verizon learned the lesson early and took action. Specifically, its SDN choices involved using the same vendor for physical and virtual networks, to avoid having the technologies separated into “information silos,” as Larish put it.
Another lesson: Use metering technologies, because “everybody asks for more than they need,” Larish said. It’s not always disingenuous, either; sometimes users don’t have a good guess as to how much of a resource they really need. Usage data can help make their requests more realistic.