Fasten your harnesses, because the era of cloud computing’s giant data centers is about to be rear-ended by the age of self-driving cars. Here’s the problem: When a self-driving car has to make snap decisions, it needs answers fast. Even slight delays in updating road and weather conditions could mean longer travel times or dangerous errors. But those smart vehicles of the near-future don’t quite have the huge computing power to process the data necessary to avoid collisions, chat with nearby vehicles about optimizing traffic flow, and find the best routes that avoid gridlocked or washed-out roads. The logical source of that power lies in the massive server farms where hundreds of thousands of processors can churn out solutions. But that won’t work if the vehicles have to wait the 100 milliseconds or so it usually takes for information to travel each way to and from distant data centers. Cars, after all, move fast.
That problem from the frontier of technology is why many tech leaders foresee the need for a new “edge computing” network—one that turns the logic of today’s cloud inside out. Today the $247 billion cloud computing industry funnels everything through massive centralized data centers operated by giants like Amazon, Microsoft, and Google. That’s been a smart model for scaling up web search and social networks, as well as streaming media to billions of users. But it’s not so smart for latency-intolerant applications like autonomous cars or mobile mixed reality.
“It’s a foregone conclusion that giant, centralized server farms that take up 19 city blocks of power are just not going to work everywhere,” says Zachary Smith, a double-bass player and Juilliard School graduate who is the CEO and cofounder of a New York City startup called Packet. Smith is among those who believe that the solution lies in seeding the landscape with smaller server outposts—those edge networks—that would widely distribute processing power in order to speed its results to client devices, like those cars, that can’t tolerate delay.
Packet’s scattered micro datacenters are nothing like the sprawling facilities operated by Amazon and Google, which can contain tens of thousands of servers and squat outside major cities in suburbs, small towns, or rural areas, thanks to their huge physical footprints and energy appetites. Packet’s centers often contain just a few server racks—but the company promises customers in major cities speedy access to raw computing power, with average delays of just 10 to 15 milliseconds (an improvement of roughly a factor of ten). That kind of speed is on the “must have” lists of companies and developers hoping to stream virtual reality and augmented reality experiences to smartphones, for example. Such experiences rely upon a neurological process—the vestibulo-ocular reflex—that coordinates eye and head movements. It occurs within seven milliseconds, and if your device takes 10 times that long to hear back from a server, forget about suspension of disbelief.
Immersive experiences are just the start of this new kind of need for speed. Everywhere you look, our autonomously driving, drone-clogged, robot-operated future needs to shave more milliseconds off its network-roundtrip clock. For smart vehicles alone, Toyota noted that the amount of data flowing between vehicles and cloud computing services is estimated to reach 10 exabytes per month by 2025.
Cloud computing giants haven’t ignored the lag problem. In May, Microsoft announced the testing of its new Azure IoT Edge service, intended to push some cloud computing functions onto developers’ own devices. Barely a month later, Amazon Web Services opened up general access to AWS Greengrass software that similarly extends some cloud-style services to devices running on local networks. Still, these services require customers to operate hardware on their own. Customers who are used to handing that whole business off to a cloud provider may view that as a backwards step.
US telecom companies are also seeing their build-out of new 5G networks—which should eventually support faster mobile data speeds—as a chance to cut down on lag time. As the service providers expand their networks of cell towers and base stations, they could seize the opportunity to add server power to the new locations. In July, AT&T announced plans to build a mobile edge computing network based on 5G, with the goal of reaching “single-digit millisecond latency.” Theoretically, data would only need to travel a few miles between customers and the nearest cell tower or central office, instead of hundreds of miles to reach a cloud data center.
“Our network consists of over 5,000 central offices, over 65,000 cell towers, and even several hundred thousand distribution points beyond that, reaching into all the neighborhoods we serve,” says Andre Fuetsch, CTO at AT&T. “All of a sudden, all those physical locations become candidates for compute.”