“We’re strategically placing these [data centers] through all our networks — wireless and wired,” he said. “The idea is to build the cloud to serve any app in any location and have that connectivity pre-built.”
Verizon began working on the project in March 2015 by evaluating technology such as hypervisors, virtualization software, network interface cards, and other hardware. In the summer of 2015, Verizon conducted the design and testing of the data centers, dealing with such things as cabling, cooling, power, and selecting the hardware.
By November, it had gone live with five sites in the United States: two with 100-plus nodes and three with 400-plus nodes.
The service provider is running Red Hat’s OpenStack Platform 7.0 based on OpenStack Kilo, but Emmons said, “We’ll move to Liberty [the most recent OpenStack release] very soon. We’re running Big Switch for software-defined networking (SDN) controllers and running white-box switches with Big Switch’s OS on them.
“It’s pretty exciting on the tip of the sphere. Taking the latest releases from our vendors of the OpenStack code of our SDN controller and making that stuff work is a lot of long days and nights and weekends.”
Verizon is using Big Switch for the core-and-pod design as well. Core-and-pod uses individually designed pods that hang off a routed core layer. Verizon chose the modular approach because it anticipates its data center environments will evolve.
“By doing a core-and-pod, we’re able to plug more pods in without disturbing the rest of the data center,” said Emmons. “We envision having versions of pods iterate fairly quickly.”
Verizon is using commercial off-the-shelf hardware but plans to look at Open Compute Project (OCP) hardware in the future. COTS was cheaper than OCP at the time Verizon was shopping.
Emmons said that Verizon’s roadmap for virtualizing its networks begins with data centers and then will “move to aggregation and then the edge.”