E-Handbook: What the union of IoT and edge computing means for IT pros Article 3 of 4

Gunnar Assmy - Fotolia

IoT edge computing: What's the practical potential?

As IoT devices proliferate, cloud just isn't fast enough. But that's exactly where edge computing comes in, enabling analysis at speeds the future demands.

Enterprise IoT systems cannot survive on the cloud alone. As the internet of things grows more pervasive and use cases proliferate, certain applications will require real-time or near-real-time analysis. Waiting for analytics back from the cloud -- and paying for it -- simply won't be feasible.

These are two of the many reasons it's hard to have an IoT conversation today without hearing the term edge computing. As the name implies, it involves processing raw data at the boundaries of the network, as close to its points of generation as possible -- often on the devices themselves or on edge gateways.

Despite the seemingly straightforwardness of the topic, there is a lot of confusion around IoT edge computing in terms of what it is, how to implement it and what its killer apps are.

Defining the IoT edge

When you talk edge computing, you will inevitably also hear about fog computing. Some say the terms are interchangeable, while others contend that edge is a segment of fog.

Robert Schmid, IoT chief technologist at Deloitte Digital, defined the edge as compute power right next to the machine and not managed somewhere by someone else. He places fog away from the machines themselves -- either at a gateway or other processing area.

"It gets a little, pun intended, foggy what's edge and what's fog," he said. "So, I think we still have a definition to play with here."

Jason Shepherd, IoT and edge computing CTO at Dell Technologies, maintained fog is too abstract, adding that edge, more often than not, is tied with a physical location.

"The way I describe it to people is edge computing is moving compute as close as both necessary and feasible to subscribers that need it," he said. For a telco, this could be at the bottom of a cell tower. On a plant floor, it would be on the production line. For a vehicle manufacturer, it's in the car itself. "And if people want to talk about fog, it's all the edges plus the network and everything between that's not the cloud."

How to perform IoT edge computing

"The current state is a bit of a wild west," said Chris Gardner, an analyst at Forrester Research. "There are plenty of great use cases, but there isn't really a consistent set of standards for architecture and management."

But companies out there are ready to solve the IoT plus edge equation. Many edge strategies available from the big players in the IoT market today involve deploying containers operating microservices within an IoT system -- be it on the sensor itself, the edge gateway connecting the sensors, at a micro modular data center or at a telco's cell tower.

What everyone really wants is to drive entirely new revenue streams and customer experiences.
Jason ShepherdIoT and edge computing CTO, Dell Technologies

For Cisco, these microservices use the proprietary IOx, the company's application environment that Vikas Butaney, vice president of IoT product management, said has two levels. First, it allows organizations to host microservices at IoT edge gateways and traditional networking environments. Second, using Cisco Kinetic, it helps create rules that handle IoT data and create logic at the edge.

Dell, along with a slew of other companies, including big-time players like Intel, Samsung and VMware as well as industry players such as FogHorn Systems, Striim and Canonical, has opted for the vendor-neutral, open source EdgeX Foundry, hosted by the Linux Foundation.

"Think of EdgeX as a framework," Dell's Shepherd said. "It has APIs that bind stuff together, with open, vendor-neutral governance. Each microservice that plugs into it can reference these APIs regardless of how proprietary or open that microservice is. It's the rug that ties the room together."

Other open source options are also available. Where EdgeX is on the application layer, for example, the open source Akraino Edge Stack involves APIs at the infrastructure layer.

Shepherd equated these frameworks to the "plumbing" of IoT.

"It's not to compete with where you make your money," he said. "You make your money on the '-ities' -- manageability, security, usability, connectivity. You don't make money on the plumbing in the middle, that's why we're trying to standardize this through de facto open standards."

However, choosing between proprietary and open source, Gardner said, is less important than matching the edge platform with the use case.

"For example, IOx has proven to work in some manufacturing scenarios," he said. "If I was a manufacturing company, I'd want to explore this. Chances are some of the architectural challenges may have been solved or, at the very least, debated in prior implementations. Referencing prior art is never a bad thing."

The true state of edge computing today

IDC predicted that at least 45% of data created by IoT devices will be stored, processed, analyzed and acted upon at the edge of the network by 2019.

It's not come one, come all and everyone's ready to go deploy edge right now; that's not where things are in the marketplace today.
Vikas Butaneyvice president of IoT product management, Cisco

But are we even close to an IoT edge computing reality?

"Edge is pervasive now," Gardner said, citing the installation of servers closer to users at branch offices and the presentation of IoT-collected data from airplanes to pilots. "What is new is using local resources to augment IoT and mobile technologies in new ways." For example, he added, a soccer club uses beacons in its stadium to determine which jerseys to market to fans based on where they sit -- a use case that would have been impossible just a decade ago.

At Deloitte, Schmid has found that almost 80% of its customer projects include an edge component, often because sending data to the cloud isn't feasible.

Schmid offered the example of fire prevention at a galvanizing plant. Because there's fire every time an object is dipped into a galvanizing pot, fire prevention and detection isn't as simple as installing a smoke alarm from a local hardware store. "Some fires could be mis-detected," he said. "And there's just not enough time to wait for what happens between detection on the floor all the way to the cloud. It's a simple latency example."

In another illustration, Schmid described a manufacturing plant where edge computing was the solution to a processing power problem. The sheer amount of IoT data gathered from the plant's machines and sensors required extra compute power to transfer to the cloud for analysis, where a machine learning model was generated that could then be pushed back down on premises.

Cisco's Butaney discussed a midsize city that deployed IOx and Kinetic to manage its intersections and fleets. "Imagine the city has 1,500 intersections and an IT staff of 12 people," he said. "They have 1,500 police cars and firetrucks they have to manage. It kind of brings to light: How do you make it easy to deploy and manage?" Viewing at scale, he added, was critical.

However, these examples don't necessarily mean IoT edge computing is ready for prime time.

"It's not come one, come all and everyone's ready to go deploy edge right now; that's not where things are in the marketplace today," Butaney said. "I think the needs of our customers are pretty basic and pretty important to start with. Then we can build things like machine learning and more sophisticated edge compute models into those devices."

The killer app of IoT edge computing

There's certainly a business case for edge computing, even if it hasn't been fully fleshed out yet. It's at the edge where the action takes place, so it's easy to understand why certain IoT applications require processing to take place there. From latency to bandwidth to security, IoT edge computing makes sense.

But what's the killer app?

In Gardner's opinion, there isn't one yet. But he has seen interesting IoT edge computing use cases in transportation, for example to monitor sensor data on train cars and buses, as well as to help monitor freight and luggage and in utilities with energy companies monitoring temperature and vibration on rigs to ensure worker safety. He also sees various customer experience applications -- for example, in retail with augmented and virtual reality -- and in entertainment venues to enhance fan viewing experiences.

Shepherd said he believes computer vision to be the overall killer app, as processing image-based data close to its source and backhauling only meaningful events is extremely valuable. He has also seen an increase in manufacturing quality control and autonomous vehicle computer vision applications, as well as its use in energy and transportation. He also thinks IoT edge computing will see an uptick in retail and healthcare -- where it is able to affect its prime target: the people.

"What everyone really wants is to drive entirely new revenue streams and customer experiences," he added.

Dig Deeper on Enterprise internet of things

Data Center
Data Management