In software-defined networking (SDN) circles, AT&T has prided itself on taking a lead among carriers when it comes to adopting new ideas such as DevOps and open source participation.
So it’s fitting that the company is also giving the P4 language a spin. The work got mentioned as part of an NFV World Congress keynote by Ken Duell, AT&T’s assistant vice president of development and engineering for the packet-optical network.
“Right now, it’s experimental,” he said. “I’m a manager now, and the code was clear enough that I could understand what it was trying to accomplish.”
Duell had discussed some of his team’s work during a P4 workshop held at Stanford University in November. His slides summarize a couple of proofs-of-concept. One used P4 inside a white-box switch to forward packets to the application layer.
The other one, developed with Juniper, used an Ethernet VPN (EVPN) to connect a physical edge switch with a virtual router, both at the provider edge.
That use case — connecting new, virtualized networks with old, physical equipment — was the one Duell emphasized during his NFV World Congress talk.
“We had a team that was able to program that in a P4 construct in 78 lines of code, in just a matter of weeks,” he said.
The work relates to the AT&T Integrated Cloud (AIC) architecture, which attempts to bring the installed physical network under the same roof as the virtual infrastructure that’s building up for SDN and network functions virtualization (NFV).
At NFV World Congress, Duell described using EVPNs to connect the two worlds. EVPNs would provide resilience, because if a virtualized network function (VNF) moves, the connection can still find the new location. And EVPNs can help with scale. NFV means turning individual pieces of equipment into multiple VNFs; an EVPN’s ability to aggregate customer flows into virtual tunnels would reduce the number of network entities an operator has to manage, Duell said.
AT&T has started seeing clients set up service chaining on this architecture, said Duell. Popular use cases that are emerging include changing the network topology at the end of the day in preparation for data center sync-ups, and altering classes of service dynamically, depending on network conditions.