Google leans heavily on SDN for reliability, velocity and availability of its network

With weekly upgrades to its networks, Google relies heavily on SDN to increase its velocity while also keeping its availability the same for users. (FierceTelecom)(Mike Robuck/FierceTelecom)

SANTA CLARA, Calif.—Google would not be Google if not for software-defined networking, according to a keynote address at ONF Connect by Google Fellow Amin Vahdat.

Google is an operator member of ONF, and earlier this year Google put its Stratum code into open source with ONF. Stratum is part of ONF's Unified, Programmable and Automated Network (UPAN) Exemplar Platform, which is working on the next generation of SDN for ONF's membership.

"Open source software-defined networking: Why did it win? Actually when we started, and when I started in this field, maybe the motivation was, 'Wow, these boxes that we buy from vendors are expensive and we can replace them with cheaper boxes,'" Vahdat said. "That was certainly an easy motivation to say we're going to move from these big expensive boxes to white boxes, and the network will become cheaper.

"Actually, in retrospect, that's not why SDN won. I think that the number one reason we have this SDN is it allows us to balance availability with velocity."

RELATED: ONF operators push reference designs out for membership review

Vahdat said that Google is constantly increasing the capacity of its network, probably along the same scale as the entire internet when it was 2 or 3 years old.

"It is the pace at which we have to move and then making sure that we can maintain that pace while maintaining the availability for infrastructure," he said. "It's very challenging. It's probably what we spend most of our time on.

"I would argue that without SDN, it would not be possible to do that. It's not about can we go buy cheaper boxes for our network? We couldn't buy the network that delivers this capability of going at this rate while maintaining our availability for enterprise. As you might imagine, it has a huge amount of value for us. The availability to be able to maintain that balance is probably the number one reason why we would have to do SDN, and why it's entrenched."

Vahdat said that Google has hundreds of thousands of switching elements across its network, and it's introducing new network elements "24/7, every day of the year." Any one of the new elements could cause Google's network to crash, but Google makes sure they're safe through emulation and verification by putting them into a copy of its network that's running on the side.

For Google, it's all about the velocity of improving the network on the fly while balancing that with availability so that end users don't feel any hiccups with the upgrades or new network elements.

"How do we have essentially a copy of our network deployments running in software on the side, so that we can test whatever we're planning on making for the network actually does what we expect? Without SDN, I would once again argue that that's just something that's not possible," he said. "We need to be able to iterate to Google a software architecture essentially every week. We have to be able to do this every week, every single week across our network, and be able to deploy it globally. Without SDN, this would not be possible.

"What I want to emphasize here is it's not about the fact that the boxes are cheaper. We're glad about that, but it's all the other benefits that are really why SDN is so entrenched for us."