If you’ve been following the global rise of the Internet of Things (IoT), one term you’ve probably stumbled across is “edge.” Edge analytics, edge computing, edge processing…the edge seems to be all the rage in the IoT world, but what is it?
For that matter, what about all those other terms that seem to pop up around “edge”—words like “fog,” “cloud” and “core”? Is the IoT world just one big game of Mad Libs?
Fortunately, no. As we’ll soon see, edge, core, fog and cloud are all related concepts that come together to describe how and where IoT data is generated, processed and stored. And since the edge is the current focal point of many prominent IoT players, that’s where we’ll start.
Defining the Edge
The edge is a pretty straightforward concept: to borrow the old real estate adage, it’s all about location, location, location. The edge generically refers to the source of data—for example, an IoT sensor that collects temperature data is an edge. Connected assets in a digital factory are edges. Any device that’s physically collecting data is an edge. Edge computing, then, simply refers to computation performed right at the source of the data.
“Edge computing is a trend to push more compute proximal to the physical world or things that are collecting data from the physical world,” explained Jason Shepherd, director of IoT strategy and partnerships at Dell . “The processes, control systems, you name it.”
The concept of local data processing, however, is not a ground-breaking new idea.
“It's funny, if you start to look at these terms—edge computing, IoT, Industrial IoT—well all of this stuff has been around since the 1950s on the electric grid, when the electric utilities introduced SCADA,” said Stewart Kantor, CEO of Ondas Networks (formerly Full Spectrum). SCADA (Supervisory Control And Data Acquisition) is a control system architecture that, as Kantor explained, was designed for remote monitoring and control of industrial applications, and is still in use to this day.
“When the IT folks talk about edge computing, they’re talking about the edge of the network, which is somehow magically interconnected to the sources of these OT (Operations Technology) data,” said David King, CEO of FogHorn Systems . “When in fact the OT folks already have edge compute capabilities called PLCs (Programmable Logic Controllers) and DCSs (Distributed Control Systems), and they’ve got SCADA infrastructure already in place.”
Is there a storm brewing over whether users will choose “cloud” over “edge” computing?
To understand that, let’s first discuss what can be considered the edge’s counterpart: the cloud. If the edge refers to the source of data, the cloud simply refers to not the source of data. The cloud is a remote data center that can receive and process information from the edge, but does not generate that information. Its exact location is irrelevant, as the salient feature of the cloud is that it’s, well, somewhere else.
“The cloud is just a data center that nobody is supposed to know where it is,” explained Dr. Tom Bradicich, vice president and general manager of servers and IoT systems at HPE . “And a cloud, or data center, is a large building with lots and lots of servers and storage that run companies and run web properties.”
If we think about edge and cloud as two ends of a spectrum, we can start to understand why the edge is experiencing a surge of interest in the IoT world.
“If you look at the history of compute in general, in the IT world, the pendulum always swings,” Shepherd said. “It went from mainframes to PCs and then the Internet and the rise of the cloud and centralized computing. So kind of central, decentral, central.”
The IoT pendulum has reached its apex on the cloud end of the spectrum, and now its momentum is in the direction of the edge. But, as Shepherd and every other industry insider we talked to pointed out, what’s truly needed for IoT is a balance between edge and cloud, rather than an either/or approach.
“I don't think the cloud goes away,” Shepherd predicted. “It becomes a sort of symbiotic relationship… it’s not about one or the other, it’s about using different compute models to optimize your performance while also meeting other needs around security and uptime and all that.”
Seven Reasons to Use the Edge
Behold! The majesty of what an "edge" can deliver. Oh, and just in case things go wrong, there's a lighthouse nearby!
“There are seven reasons to compute at the edge,” offered Dr. Bradicich. Here’s what makes his list:
It takes time to send data from the edge to the cloud, process it and send it back from the cloud to the edge to enact some decision. For IoT applications that require low latency, this delay can be prohibitive.
“No matter how much people tell you that 5G is going to be so fast you can do real-time control in the cloud,” asked Shepherd, “do you really want your autonomous car asking the cloud what to do when you’re about to hit something?”
The latency advantage need not be so dramatic. According to King, many IoT applications simply don’t work if the delay between data and decision is too drawn out. “Ninety-nine percent of data is useless after a couple seconds,” he said. “If you haven’t spotted a pattern or an error or an anomaly in that data within frankly a few milliseconds or a few seconds, it’s kind of useless to the operations people running the factory.”
If you process your data at the edge instead of sending it to the cloud, you remove this delay and can make immediate decisions for a truly real-time response.
Sending data from the edge to the cloud takes up spectral resources—there’s a limit to how much information can be sent at one time. According to Dr. Bradicich, this limit will become more of a problem as more IoT data is generated.
“There’s just not enough bandwidth to go to the cloud with all the information coming from all the things,” he explained. “When we get to the point where every car is producing more data per hour than a full length high-definition movie, gigabytes and terabytes of data, times all the cars in a city, times all the cars in a state, going to clouds, there’s not enough bandwidth. There’s just not enough bandwidth for all the big data that you want to get from the edge.”
Just as there’s a bandwidth cost in sending data from the edge to the cloud, there’s also a much more relatable monetary cost. “Even if you sent all your data to the cloud and you didn’t care about how fast it was, there are network costs,” said Dr. Bradicich. “So it costs money to send it to the cloud, and it costs you to pay the cloud to compute.”
If your IoT application is mission critical, you can’t afford problems with network reliability. By processing data at the edge, you eliminate this potential pain point.
“There are certain elements where you just need to have that compute local because of the bandwidth savings, the latency, and it always works no matter whether I lose my backend connectivity,” said Shepherd.
Computing at the edge provides more security than computing at the cloud, and for a fairly simple reason.Dr. Bradicich illustrates this reason with an analogy:
“I like to use the metaphor of Ronald Reagan, President of the United States in the late 1980s. He had the best security in the world—he was President of the United States. When he left the White House, he got shot. What happened? It wasn’t a matter of security being the best, it was a matter of vulnerability. When you leave your house, you’re more vulnerable. If I never left my house I would never get into a car accident. So it’s a matter of vulnerability. When you send the data all over the state, all over the campus, all over the world, it’s vulnerable to hostile attacks.”
As regular readers will recall, security is one of the biggest issues in the IoT world, and one that has yet to be resolved . Edge computing is one more safeguard against potentially devastating security flaws, as Shepherd explained.
“The notion of hooking up a process control plant with chemicals where things go boom if something goes wrong… you don't want to connect that directly up to the cloud. You want layers of protection in the network of computing happening locally.”
According to Dr. Bradicich, edge computing can help overcome some data redundancies. “If you collect all the data at the edge, you have to have that amount of storage. If you send it all to the cloud, you have to have that amount of storage. So there’s not a 100percent duplication, but there’s some duplication in resources, duplication in upgrading, duplication maintenance.”
7. Data Sovereignty and Policy
The last of Dr. Bradicich’s seven reasons for edge computing simply comes down to external restricitons. If your IoT data is sensitive or there are other reasons it can’t be sent to the cloud, you have no choice but to process it at the edge.
“I’ve worked with customers who have told me, ‘None of those other six reasons matter. But I have a policy where my data is not allowed to leave my edge, so I can’t send it to the cloud.’ These data sovereignty policies exist because sometimes you’re not allowed to send data across country borders. So you have data sovereignty, sometimes called geofencing. So you might just have a geofencing data sovereignty policy issue.”
Edge and Cloud: Better Together
As immersive streams of data begin filtering into homes and workplaces, whether they be at an office or out in the field, a blend of edge and cloud computing will be the most effective choice for crunching the loads of data that are being generated.
One advantage of cloud computing is that is doesn’t take up any room, at least not anywhere you care about. If you’re constrained for space, it might make sense to send data to the cloud instead of processing at the edge. “It might be in a very very tight space in a submarine,” offered Dr. Bradicich. “And since none of the seven reasons matter, you can transmit to a cloud and do your computing there because of space.”
But perhaps the biggest advantage of cloud computing is the fact that data can be aggregated from a variety of remote sites. This enables your edge to incorporate information from other edges via a centralized command center.
“For example, let’s say I'm working at the edge in a manufacturing site,” said Dr. Bradicich.“But I'm going to control my robot arms in my production based on information from another edge, another manufacturing site in Europe. So I have one in Mexico and one in Europe and I want them to communicate. Two edges. The way I communicate is through a cloud. So they both send their data to the cloud, and then the cloud sends the data from the other guys’ site to their site and they can make decisions on optimizing production. That’s probably one of the biggest reasons to use a cloud. When you need to enjoy the benefit of aggregated data from other locations that are not at your edge.”
Since edge and cloud each have different advantages, a combination of the two will often be more powerful than simply choosing one or the other. Edge processing, for example, could be used as a sort of filter for what should actually be sent to the cloud.
“At the edge, you need to do some processing, some filtering, to decide what decisions can be made at the edge and what should be sent to the cloud,” explained Olivier Pauzet, vice president of market strategy at Sierra Wireless .
A model like this achieves the advantages of edge computing without sacrificing the centralized control of the cloud.
“This is not a cloud replacement technology,” said King. “Frankly, it’s a way to deliver many more real-time, high-value applications that are managed or coordinated or orchestrated from the cloud. But really, the vast amount of the actual processing of this OT data is happening at the source, before sending it anywhere.”
Fog and Core: Filling in the Spectrum
To clarify what’s meant by the terms “fog” and “core,” let’s return to the idea of edge and cloud being two ends of a spectrum. On the edge end of the spectrum is local data processing—wherever the data is generated, that’s where it’s processed as well. On the cloud end of the spectrum is remote data processing—the data is sent from the edge to a distant location to be processed.
With this picture, start on the edge end of the spectrum and shift a little bit over towards the cloud. This will give you something quite close to the source of the data, but not quite the edge. This is where you would label the core.
“To us, the core is on-premise compute spanning highly local micromodular through full on-premise data centers,” said Shepherd, whose company Dell recently revealed its three-tier IoT compute model consisting of edge, core and cloud. “It bleeds into the edge with server-class compute close to physical things and to the cloud over wide-area connections.Some people include what we call the core in their definition of edge, but we feel it’s important to differentiate between the two. The core has benefits of both extremes – more powerful compute than the edge will ever have due to cost, power and space constraints, but still close to things for responsiveness, improved security/governance and reduced backhaul costs.”
In short, the core combines the physical proximity of edge computing with the more dedicated processing power of the cloud.
But what about the rest of the spectrum? What’s between core and cloud? In fact, we can cover the entirety of the spectrum, with the exception of the cloud end, with a bit of a catch-all term: fog.
“The easiest way to put it is that the fog is everything that’s not the cloud,” explained Shepherd.“And that includes not only the edge, but all of the field devices, and control systems, plus gateways and on-premise servers, and just all of these distributed compute nodes all over the place. The fog also includes all of the networks. Wide-area networks, local-area networks. The edge and the core tend to be a bit more proximal in context, proximity to where you’re running the compute. The fog tends to be more network-based, a little bit more abstract. It’s about all of the relationships between all of these edges and how data is exchanged in real time, and how you offload and load balance and the backup and all that.”
Living on the Edge
Edge computing, like all other IoT design options, must be considered in the context of your specific application. It’s a great choice for many IoT applications, such as those that require low latency, extra data privacy measures, mission critical reliability or any of the other items on Dr. Bradicich’s seven-fold list. With that said, the edge isn’t a replacement for the cloud, and the degree to which you utilize edge and cloud processing will differ from design to design.But hey, balancing trade-offs is what engineers do.
Now that you’re educated on the edge, clear on the cloud, confident in the core and familiar with the fog, your IoT applications are primed for a whole new model of compute. And maybe—just maybe—you’ll get a chance to see what life’s like on the edge.