With a 116-station WiFi network being built for India's railway system, Google says it's building evidence that carrier services can indeed be delivered in a cloud-native fashion, based on microservices built from containers.
"I think the cloud-native model is here for NFV, and I believe re-imagining carrier network services as cloud-native is possible," said Geng Lin, the company's CTO of enterprise network and infrastructure services, speaking at this week's NFV World Congress.
Of course, Google (Nasdaq: GOOG) has a few advantages over your everyday telco. The applications in question are greenfield and don't carry a decades-long legacy. Google isn't messing with tectonic layers of OSS amassed via acquisitions, nor is it being held to five-nines reliability -- and the company has ready access to a large public cloud (Google Cloud Platform) for "free."
Still, the Railwire project provides a look at the possibilities for the microservices approach in an NFV world. It's working at scale. Google believes its deployment will rank among the largest over-the-top broadband services ever created, having grown to about 6 million users per month; Google expects that figure to be 30 million when the project is done.
Lin's talk was also a sequel to last year's NFV World Congress, when Google first presented the Railtel project. The first station connected to the WiFi network launched in January 2016; Google is now up to 116 stations and aims to reach 400 in 2018.
The network's control plane is virtual, hosted in Google Cloud Platform and split up into microservices.
"As the network software-ization becomes reality, [and with] the separation of the control plane functions from the data plane, we see most of those control-plane functions are software, and you can easily decompose them and reconstruct them into the cloud," Lin said.
The microservices Google has developed include a vast amount of monitoring, and that's where analytics come in. In place of traditional OSS, Google is using a cloud-based analytics platform, a key step toward creating a network that can respond to changing conditions automatically.
"Instead of just paying attention to the so-called abnormal events, we actually collect network health data at a constant rate, regardless of whether anything goes good or bad," Lin said.
That's been a trend in analytics and security. Vendors are experimenting with recording literally everything a network does -- something that's become feasible only in recent years -- and then using machine learning or some similar technology to winnow the information.
This makes it possible to monitor the network's health comprehensively. But the real goal is to cut down on human interaction with the network. According to Lin, Google uses its analytics not only to alert humans if something has gone wrong, but also to indicate what the impact is -- which services have been affected, for instance. That reduces the time humans spend on analysis and triage.
The analytics platform can also trigger the network to fix itself, in the case of problems that are simple enough -- again cutting down on human intervention.
As a result, two engineers are enough to monitor this network that's serving 6 million people per month, Lin said.
— Craig Matsumoto, Editor-in-Chief, Light Reading