What Is Edge Computing?
The Lakeside Technology Center on Chicago’s near South Side is a behemoth. Encompassing 1.1 million square feet, extending eight stories tall and housing thousands of servers, the onetime headquarters of RR Donnelley printing is now among the largest data centers on the planet.
But the potential future of data processing is located some nine miles west, just beyond (aptly enough) the city’s edge in the shadow of a big-box retailer. It’s a single-room, windowless, blink-and-you’ll-miss-it facility — the antithesis of Lakeside.
The module connects to two others deposited locally by edge-computing startup Vapor IO. The company plans to implement its brand of edge infrastructure in 20 markets by the end of 2020, and Chicago was the first stop. It’s also a physical testament to the growing desire to move processing capability beyond leviathan data centers and closer to the action (so to speak) in order to preserve bandwidth and reduce latency — or the surface area data must travel to be processed.
“The whole point of edge computing is to get closer to devices, to reduce the amount of data that needs to be moved around for latency reasons, to get closer so that responses are faster,” said Matt Trifirio, chief marketing office at Vapor IO.
How fast? The southwest Chicago site, which offers proof-of-concept trials to clients, recently processed a trial run of a facial recognition app. Run along a traditional centralized path, the job took 240 milliseconds to complete. Via the edge center, a mere 13 milliseconds sufficed.
Edge Computing Basics
Compared to head-spinning emergent technologies like quantum computing, the concept of edge computing is pretty simple to grasp despite its technological complexity. Here’s the basic idea: A job that requires little or no travel typically can be completed more quickly and easily than one that requires a long trek — especially if, at the end of said journey, lots of other workers need to use your machine.
These days, there’s a good chance that everything from the light bulb in your kitchen to the car in your garage is “connected.” That never-ending — and always-increasing — stream of processing jobs means heavy strain on data centers. Edge computing eases that burden by moving some of the processing closer to its point of origin — as close as possible to where the action occurs.
And so, rather than traveling to the cloud, the job is done “on the edge.” Sometimes that means the processing occurs where it’s launched — in the device itself. Think sensor-equipped wind turbines and assembly machines. For bigger jobs, it also sometimes means processing in “cloudlets,” which are essentially decentralized mini-data centers that can handle certain commands of certain users. A cloudlet could theoretically be even smaller. “A Dell box with an Nvidia GPU chip inside sitting in the corner, running applications that are edge-native, maybe that’s my cloudlet,” edge trailblazer Mahadev “Satya” Satyanarayanan said. A professor of computer science at Carnegie Mellon University, Satyanarayanan authored the 2008 paper “The Case for VM-Based Cloudlets in Mobile Computing.
What is edge computing?
While Vapor IO’s edge data centers aren’t quite that small, they’re closets compared to a traditional data center. The ones that stand independent of larger structures closely resemble shipping containers that you’d find at a port of call. But while they’re not much to look at, but they portend big things.
Some industry watchers believe the cloud (which is dominated by Amazon, Microsoft and Google) will one day be used mostly for storage and massive computations. And due to the massive amounts of data they gobble up, many of tech’s future marquee breakthroughs — things like autonomous cars that rely on LIDAR to avoid crashes, large-scale drone delivery and holographic performances that will make today’s smoke-and-mirrors versions seem like Pong— are almost certain to incorporate edge data centers. Major innovations that are now taking off, like high-tech farming and IoT, also lean on the edge.
Laying the Groundwork for an Edge Future
But if those bright-and-shiny applications represent the glamorous endpoint of edge computing, right now the client base is more akin to the foundation layer that goes on before makeup is applied. Of the four industries that comprise the bulk of Vapor IO’s client base — telcos, cloud providers, content delivery networks (CDNs) and streaming gaming — only the latter could be characterized (charitably) as glamorous. Still, they represent important steps forward.
As telecommunications companies upgrade to 5G and advance into cloud platforms, there’s been a push toward network virtualization.
“In order to do that, the telcos need data centers that have low-latency access to their towers, and I think they've mostly proven themselves to be not the best at running data centers,” Trifiro said.
The rollout of new infrastructure, with an eye toward cost reduction, isn’t particularly grabby, “but over time,” Trifirio said, “that’s what will enable the platform for these sexy new revenue-generating applications and experiences that everybody talks about when they anticipate the exciting stuff.”
At the same time, cloud providers are building edge computing into their IoT tool chains (Trifirio: “it's kind of this transparent edge-to-core capability”) or offering edge products that developers and enterprises can purchase. (Think EC2 instance purchasing.) And while CDNs have basically always lived on the edge, as it were, they existed to cache content for stuff like rich websites and video in order to prevent ridiculous buffering times. Now they’re getting into compute and, more specifically, security.
The outer rings of networks are particularly attractive from a security perspective.
“You want security products to be as far out in the network as possible, close to the perimeter,” Trifirio said. “That lets you detect anomalies faster, before they penetrate the network further. It also lets you lock down a smaller portion of the network, and sequester it.”
Security’s close cousin is privacy, which the edge could similarly imporove. Carnegie Mellon’s Satya sees a market for software that could hold back certain aspects of data prior to moving upstream to the cloud. Here’s a hypothetical scenario he offered concerning water companies: Imagine only redacted water-meter readings hitting the company’s cloud service, but if that data were to suggest a problem — a leak, for instance — edge data could provide fine-grain analysis.
“With the data on my cloudlet, I have the opportunity to decide whether or not to release it,” said Satya, who envisions the strategy put forth by outfits like Vapor IO as emblematic of the edge’s path forward.
A “Sexy” Application
Though the edge holds great promise, it’s also difficult to kickstart — particularly in terms of supply chain.
“You need network operators, cloud providers, equipment providers, middleware orchestration applications, software providers,” Trifirio said. “You need that whole value chain.”
Vapor IO is one of dozens of companies that make up the Kinetic Edge Alliance, a partnership of edge-deployment and technical-support outfits working to make that build-out of edge infrastructure as efficient as possible. The alliance also underscores edge computing’s wide-ranging ecosystem — one that’s beginning to facilitate some of those “sexy” applications.
In late September, more than 4,000 miles from Chicago in the German city of Wolfsburg, a small group of fans and journalists watched a top-level soccer match with their smartphones — with being the operative word. Watching live from inside the stadium, they hovered smartphones over the on-field action as a plethora of game-related information splashed across their screens. Tapping on any given player summoned standard sports statistics as well as more granular data — like a player’s running speed, or how much distance he had logged since the game’s start. All in real time.
The augmented-reality demo was one of two held in recent months by Paris-based startup Immersiv. Its application, dubbed Arise, uses computer vision and machine learning to process in real time millions of data points that are supplied by a sophisticated player-tracking system. Here’s the catch: data must keep pace with action or the concept crumbles.
“It was very important for us to be able to have low latency and to transmit the data as fast as possible to the user’s smartphone, Stephane Guerin, co-founder of Immersiv, told Built In.
Arise feels like a natural step in the trend toward second-screen augmentation. Seventy percent of respondents in a recent Statista survey said they used a digital device to look up information about the content they were viewing. The more seamlessly information integrates with content, it seems, the better the experience. To an even greater degree, that’s likely the case with the stadium experience, which lags behind the televised one in terms of at-the-ready information.
That’s certainly how Immersiv sees it. Prior to creating the company, Guerin and his co-founder worked in the sports industry developing pretty standard mobile applications. Immersiv was launched, Guerin said, because they were “convinced that AR would play a big part in the next generation of the fan experience.”
To achieve liftoff for Arise in Wolfsburg, the company turned to edge computing company Vodafone. And for its first demo, at a German basketball league game, it worked with KEA partner and Deutsche Telekom-backed MobiledgeX. Those were the best options for clearing the all-important latency hurdle, which all developers of similarly advanced AR apps face. In the Bundesliga demo, for which edge servers were placed inside the stadium, processing times clocked in at single-digit milliseconds, Vodafone director of innovation Michael Jakob Reinartz told SportsPro. MobiledgeX aims for similar hyper-proximity.
“For us the biggest benefit, it’s a question of reducing the latency of course, that's very important for us,” Guerin said. “But also the idea of MobiledgeX is to have servers close to the users. If we want to deploy, for example, in many stadiums in Germany or in other countries, it would be difficult to find data centers near the [all the] stadiums. MobiledgeX basically wants to have servers in all the major cities and to be able to deploy very easily, close to each user.”
Immersiv will fine-tune Arise, adding even more stats and data, ahead of a planned launch in early 2020.
Waiting for the Edge's “Spreadsheet Moment”
Despite an ever-increasing number of computing devices and key early-stage gains by companies like those in the Kinetic Edge Alliance, Satya said edge computing is still in a holding pattern. Developers of edge-dependant applications would ramp things up if they were confident that infrastructure builders were investing big in the edge, he explained, but edge computing creators likewise want to see that same demonstrable commitment from developers.
One potential way to resolve the standoff and ignite the necessary partnership? A successful bridge application, one that — to borrow a product phrase — crosses the chasm, moving from early adopters to the broader public. Just like what the spreadsheet did for the IBM PC back in the 1980s.
“If you go back and look at the sales data, the thing that transformed personal computing was the invention of the spreadsheet,” Satya said.
After the emergence of the spreadsheet, and Lotus 1-2-3 in particular, “many mom-and-pop stores saw how valuable this software could be to their business … And PCs became the platform needed to run this incredibly valuable application. So PC sales went up, and clones and competitors started emerging.”
The virtuous cycle was complete.
So what might be edge computing’s spreadsheet? One possibility, Satya said, is computer-vision-enhanced, hands-free wearables — essentially Google Glass on steroids. That might come as a surprise to those who remember the Great Smart Glasses push, circa 2015, as more punchline than breakthrough. But even though Glass flopped commercially, its repositioning as an enterprise device — to aid those in logistics, manufacturing, surgery and other industries — has gone well enough to necessitate a recent upgrade. This second edition even supports computer vision and “advanced machine learning capabilities,” according to Google.
Still, any wearable with hopes of becoming a so-called killer app — a feature that’s so desirable that people happily sign onto its required hardware too — will almost certainly need an extremely wide focus. Something along the lines of smart glasses that allow surgeons to perform guided procedures and Ikea customers to make sense of furniture assembly instructions, Satya said by way of example.
“It’s always possible to narrow the problem definition enough that you can squeeze it into a mobile device. But the moment you generalize the problem, the moment it has to work in a broader range of scenarios, things fall apart very quickly. So having the edge nearby to offload computation makes a huge difference.”
But some builders are hardly idle while we await the advent of edge computing’s Lotus 1-2-3. A recent MobiledgeX report, in which the company polled seven edge-reliant startups, basically doubles as a cri de coeur to the global telecommunications industry to get building.
“The key message from developers … was to stop waiting for a myriad of edge computing use cases to be proved before rolling out the infrastructure,” the report reads. “Application developers need more sites in order to test their solutions and commercialize them for their users, wherever they are in the world.”
In addition to what some view as insufficient cooperation between hardware builders and software providers, the fact remains that building out an edge computing network is difficult work. Vapor IO had to figure out how to make its modules operate in some remote hinterlands while also supporting multiple tenants, giving clients the ability to remotely observe processing speeds, and keeping the whole operation humming without need for constant onsite maintenance.
And as advanced as digital tech gets, hardware stubbornly remains beholden to physical conditions. Examples include the fan in your laptop or complex chilled water or oil systems in large-scale data centers. To that end, Vapor designed its server racks as cylinders rather than rectangles in order to optimize airflow. It also added a host of sensors, monitors and air re-directs to maintain ideal temperatures.
Edge data centers must also be able to withstand external elements. As the authors of a recent report by edge membership organization State of the Edge wrote, “Unlike centralized hyper-scale data centers, which can be carefully and methodically located, edge data centers often need to go near the data, no matter how harsh the location.” Flood and seismic risks, air pollution and temperature fluctuation all must be taken into consideration.
Access to land and fiber can be hurdles, too. Multi-billion-dollar real estate investment trust and Vapor IO partner Crown Capital — the largest owner of wireless infrastructure in the U.S. — has a lot of both, including in Chicago, where its expansive fiber routes connect Vapor’s edge modules. The city has long been America’s railway hub. After railroad companies used their land-grant rights to have telco partners run fiber-optic lines along rail lines, it also became a major fiber hub. Network-wise, not all cities are so fortunate.
Access to power sometimes poses a challenge, as well. Take the aforementioned southwest-Chicago structure where Vapor IO set up a rack chamber: it already existed as a Crown Capital site, which meant plenty of juice was already flowing through it. Similar sites, attractive though they may be, aren’t always so lucky. Vapor sometimes has to wait months for an electric company or utility to extend the necessary power lines.
All of which is to say, the edge industry needs copious capital to flourish. According to Trifirio, who declined to offer specific figures, the deployment of data centers requires “multiple millions of dollars per city.”
As a network is pushed further from the fortress-like cloud, issues arise regarding the physical security of outposts — even as the edge makes data transmission more secure. But that might be less liability than opportunity. Development of tamper-resistant hardware and tamper-prompted data erasure stalled out a bit as cloud computing took off in the early 2000s, but the rise of edge computing will likely resuscitate both research and business opportunities in the field, Satya said.
“As edge computing becomes prominent and important, many of these [security] ideas that have kind of been dormant will become significant again. You might see lines of edge-deployable hardware that have some additional hardening.”
While the advancement of edge computing is rife with challenges, none appears to be anything resembling an existential threat — especially considering the imminent tsunami of forthcoming technology.
“Going to the cloud, doing your work there and coming back is going to take far too long for the kinds of applications you really want in the future,” Satya said. “Five years ago people were asking, ‘Do we need the edge? Why can’t the cloud do it?’
“Nobody questions that now.”
Images via Vapor IO except as noted.