Living on the Edge: Amazon, AT&T, Packet Pursue “Cloudlet” Computing

Xconomy San Francisco — 

The word “cloud”—as in cloud computing, and cloud storage—has served as a handy shorthand term, but it has always been inherently vague and a little misleading. When businesses and consumers use a cloud service provider like Amazon Web Services or Apple’s iCloud, their data, photos, and music don’t get processed in the misty skies above us, as a poet friend of mine once assumed. Instead, of course, they travel via the Internet to someone else’s data center—an unquestionably physical array of servers set on solid floors that can bear the weight.

But cloud gets it right on one score—it suggests distance. Cloud computing has made it possible to outsource our data storage and processing to big commercial facilities that are often far away from us, where land and electricity are cheaper. For more than a decade, this model has offered advantages for startups and large companies alike—allowing them to scale up their computing power as needed, pay only for what they use, and subscribe to software that’s constantly updated for them.

For the most part, distance didn’t pose much of a hindrance—but now that’s changing. The masses of data being generated by cars, connected devices, and factory machines are creating a centrifugal force, pulling the cloud to the periphery. In response, the centralized-cloud sector is spawning local “cloudlets,” or “edge clouds,” in a movement that could help reshape the structure of the Internet and speed the pace of technological change.

Even as businesses are still transitioning from their in-house data centers to giant centralized cloud-computing facilities, companies from Amazon to telecom giant AT&T to startup cloud service provider Packet (more on them below) are pushing their networks of data centers out to the edge of the Internet, closer to their users. Here’s why:

1. New technologies such as self-driving cars and virtual reality games rely on the lightning-fast movement of data across the wires and nodes of the Internet. Lag times—what IT folks call “latency”—can prevent a car from reacting fast enough to avoid a crash, or make a mobile VR gamer queasy when the image in front of her eyes doesn’t match her head movements.

2. Streaming video gave rise to the earliest manifestations of “edge computing,” when companies built simple data storage clouds closer to their customers than centralized cloud computing data centers. Those content delivery networks, where video libraries were stored, made the trip toward viewers’ TV monitors shorter.

3. Next-generation chip designers are making devices smarter. A car or industrial machine equipped with AI chips can take over some of the computational burdens that are often offloaded to cloud-computing sites. This is the device side of edge computing. The consequences of this will take years to play out, but cloud service providers and telecom companies are trying to figure out how they can interact with intelligent devices—which are often in motion—so they can amplify the impact of fast-moving technological advances.

Case in point: Early this year, the AT&T Foundry opened an edge computing test zone in Palo Alto, CA, and began evaluating its potential in a partnership with GridRaster, which powers high-end virtual reality and augmented reality experiences on mobile devices such as smartphones. Performance did improve when AT&T’s data processing resources were located close by, the Foundry team reported this week. The experiment also yielded other lessons—such as the need for techniques to better split up computing tasks between mobile devices and cloud computing resources.

(Software companies are already popping up to solve such problems. San Jose-based Swim.ai is developing software to empower computation in devices, and also to offload tasks to outside data centers when the device can’t handle them.)

The AT&T Foundry is now collaborating with Nvidia to combine edge computing with 5G to enhance video gaming. The AT&T unit also announced that it is enlarging its edge computing test zone to cover the entire Bay Area, as it works with outside companies on other experiments. As the telecom giant envisions it, edge data centers will eventually be distributed at cell towers and other locations across its network, so that no customer will be more than a few miles from one of these cloudlets.

Will every cloud service provider have to build a granular edge-computing infrastructure of its own, like AT&T? Not necessarily. Smaller providers, such as SoftBank-backed Packet, may join together to create shared edge computing networks.

Packet, based in New York, announced a Series B funding round Tuesday, and plans to use part of the $25 million it raised to build as many as 50 new edge data centers over an unspecified period.

“We are working with multiple cloud service providers to interconnect Packet with their platforms and partner on edge ecosystem development,” CTO Ihab Tarazi told Xconomy.

Before joining Packet in June, Tarazi, the former CTO of Redwood City, CA-based Internet exchange point operator Equinix, took a six-month break as entrepreneur in residence at Sutter Hill Ventures to ponder the potential of edge computing and mine the rich contacts of Silicon Valley as he planned his next move.

Tarazi came to see edge computing as the next big wave of cloud development.

“The edge is so much bigger than the cloud,” he says.

Tarazi is now heading Packet’s 10-employee San Francisco office, which will move to Palo Alto in October. He says he signed on with Packet, whose total staff numbers 49, because he sees it as well-positioned to succeed as cloud computing services transform into distributed networks of data centers close to the sites where data is created.

Packet was founded in 2014 by CEO Zachary Smith and his twin brother, CMO Jacob Smith. In 2015, the startup began life as a cloud service provider by building a centralized data center 30 miles outside New York, in Parsippany, NJ. In 2016, it added three more core data centers in Sunnyvale, CA, Amsterdam, and Tokyo. Last year, it began expanding to the edge, by setting up computer processing resources at multi-tenant data centers in 11 other cities including Seattle, Dallas, Los Angeles, Atlanta, Chicago, Toronto, Hong Kong, and Sydney, Australia.

Packet’s next wave of edge expansion, announced this week, will take it “much closer to the network edge,” Tarazi says. The 50 new edge data centers will sit at the base of cell towers in metal buildings the size of an 18-wheeler truck trailer, where they’ll be connected to the Internet through both fiber-optic cable and wireless at the tower itself, he says. Packet is working with all the big cell tower companies to explore sites, Tarazi says.

Six edge data centers are already in progress; of the first three with announced locations, one will be in the Boston area and two in Chicago. The primary benefit for users is expected to be reduced delays (latency) in the movement of data, Tarazi says. Packet allows customers to choose the location where their data will be processed—a policy consistent with its overall business model.

The company is known as a “bare metal” cloud service provider, which means that its customers don’t need to process their data under pre-set conditions, such as the operating system preferred by a cloud provider, Tarazi says. Packet users can specify the software, server type, semiconductors, operating system, and networks they want to work with, creating a customized data center in a public cloud, the company says. As a further inducement, Packet just announced that it can now automate deployment of a customer’s application in 60 seconds. Developers can use the system to test their software under various server configurations.

Tarazi says Packet customers include mid-sized enterprises and developers within companies, while telecom companies and big enterprises are starting to show interest. The startup is working with wireless companies to make 4G more effective for them, and to implement the faster, next-generation cellular technology 5G, he says.

With its new capital, Packet’s fundraising total now amounts to $36.2 million. Its Series B funding round was led by Third Point Ventures, and joined by Battery Ventures, JA Mitsui Leasing, Samsung NEXT, and earlier investors SoftBank and Dell Technologies Capital.

In June, Packet released a “State of the Edge” report  created with industry partners Arm, the giant U.K. chipmaker owned SoftBank; Vapor IO, an edge cloud builder based in Austin, TX; edge delivery network builder Ericsson UDN; and Rafay Systems, which focuses on the deployment of applications to take advantage of edge computing.

The report describes layers of edge cloud-computing centers that could inch even closer to end user devices than the 50 edge data centers Packet now plans to build. These sublayers could include an “access layer,” which would collect data and analysis produced by nearby smart devices, and then selectively send some of that data to a higher “aggregation layer,” which might coordinate with other edge clouds on that level.

How can cloud service providers decide how close to the device edge of the Internet infrastructure they need to be?

“That’s a question for a lot of people; devices are getting more powerful,” Tarazi says. “The device and edge computing infrastructure will meet somewhere in the middle.”

Photo credit: Depositphotos