Spend time with anyone offering enterprise or industrial internet of things products and you’ll encounter the term edge: edge devices, the network edge or edge analytics. But the term by itself is almost useless because depending on who you are talking to the edge means different things.
For industrial automation, it typically means a cluster of 10-20 sensors on a single machine that feed into a small amount of computing. If you’re selling large connected building systems, it might mean a gateway device. If you’re a farmer it might mean an individual sensor deployed in the field. The edge is actually more nebulous than its counterpart … the cloud.
Figuring out how to define the edge matters if we ever want to get beyond the jargon that vendors use when they talk about their software and hardware products.
For example, Foghorn, an industrial IoT analytics company, released a new machine learning capability this week. It describes it here:
The new FogHorn Lightning ML software platform embeds machine learning at the edge, resulting in new process efficiencies, optimization, cost savings and improved product quality for end users in manufacturing, oil and gas, utilities, smart buildings and other industrial verticals.
At no point would you realize that by edge, Foghorn is talking about 32-bit ARM based computers and things like a Raspberry Pi. These aren’t sensors or possibly even a collection of sensors in an industrial setting. These are more like gateways.
I encounter this a lot and it’s more than just a barrier to understanding various products aimed at enterprise or industrial customers. It also hides the complexity of the infrastructure we are building, and what it will take to connect, manage and optimize it.
Whether we expect there to be 20 billion or closer to 50 billion connected devices in the next few years depends on how we define the edge elements of the network. If we think of such devices like our smartphones and home hubs as the edge, the lower number makes sense. But if we dig into the individual components managed by these hubs or located on phones, all of those numbers seem small.
It’s worthwhile to consider the individual elements, because in doing so you can understand what’s possible. Every new advance in sensor technology brings new capabilities to the overall internet of things. Adding 3-D imaging sensors to Microsoft’s Kinect and accelerometers to Nintendo’s original Wii enabled a new types of video games. That’s one reason I devoted my entire podcast guest segment to the rise of 3-D image sensors this week.
By recognizing individual sensors as the actual edge, you also segment out their needs and can fine tune them to optimize battery life, performance or whatever else you want. Apple created a special co-processor for computing motion from accelerometers in the iPhone that later made computing more efficient in the Apple watch. In that example, the optimizations led the way to an entirely new product category.
So from a design perspective, it’s worth considering the edge as the individual sensor, because in doing so you can make it more efficient. But you can also make it useful in surprising ways as well. A focus on the potentially billions of devices at the sensor-level also changes the way you build networks to connect things to the cloud and to each other.
Obviously not every sensor will get its own Sim or even a Wi-Fi connection. It’s more likely that manufacturers will build sensors with a variety of low-power connectivity options tacked on to work within whatever network already exists. This means that fragmentation may always exist at the sensor level, where sometimes wireless won’t even be an option. In industrial settings some sensors are so important for safety or the process that they will always be wired.
If you’re planning a real IoT system, using sensors at the edge means you’ll have to consider a variety of connection options. While many vendors don’t want to go that deep into the inner workings of their clients today, it’s a missed opportunity when they ignore a whole swath of capabilities by ignoring the real edge of the internet of things.
For more on the actual edge, check out Dean Bubley’s awesome post on sensors as seen from the point of view of a telecom analyst.