Getting Granular On The Edge

Imagine a plane flying at 30,000 feet. Two things are visible—clouds and land. In the processing world, that land mass is the edge, and as the plane begins to descend the edge begins to take on a more distinct shape and different features begin to appear.

From the air, everyone can see just how large this market opportunity is. What they can’t make out are the winning models for success. But that’s beginning to change.

Coming into focus are three distinct demarcation points for the edge, each of which is defined by data. First, there is the endpoint device. This can be a car, a robot, or a 5G system. In most cases, though, companies are beginning to look at these endpoints as systems rather than sensors. So while sensors will be used to generate the data inside of these systems, edge systems take on a specific role of processing data for a specific use.

Rather than defining the device according to a physical connection, such as an IoT device, the edge defines them according to how data is utilized and partitioned. It may sound like hair-splitting, but the differentiation is significant. The real challenge at the edge is being able to figure out what gets processed where, and how to prioritize that data in a useful way. So IoT devices may contribute some of that data, and endpoint devices may process some or all of that data. But the real issue is about the time it takes to make a decision, which in the case of a fast-moving vehicle or a medical device can mean the difference between life and death.

The second demarcation point is the cloud, and how these two worlds ultimately divvy up processing remains to be seen. There always will be a need for massive amounts of processing for things like verification and simulation of complex chips or systems, or mapping global weather or genetic sequencing. Time sensitivity in those cases may be measured in hours rather than nanoseconds, but at the edge nanoseconds are critical. The challenge is in intelligently prioritizing what gets processed where, and even within those systems it’s about what gets moved to the front of the queue or what requires dedicated processing.

The economics of the edge versus the cloud are still to be determined, but it likely will be far less expensive than the cloud. The cloud uses a model similar to renting a car. While $50 a day may not sound like a huge amount of money for that car, used every day it amounts to $18,250 per year, which basically pays for the cost of that car in one year. If you rent a fleet of cars, the cost goes up accordingly. But if you only need the car or even a fleet of cars for a couple days, that model makes sense. The cloud provides virtually unlimited processing capability, but utilizing it effectively requires a sophisticated analysis of how much compute and storage capacity will be required for a given amount of time, and what it would cost to run that on-premise versus off-premise.

The third demarcation point involves IoT devices, and the line gets blurry here, as well. It’s not hard to think of a smart watch or a connected security system doing more local processing, particularly as new chip architectures and packaging options begin hitting the market. Some of that processing may even move closer to the sensors, creating hybrid types of devices that can act as a gatekeeper to determine what gets routed where.

Regardless of what roles these devices or systems take on, the edge will add another layer of data processing. And while this inevitably will turn into a competitive battleground as software, hardware and security architectures are developed to provide a range of compute options, it also will provide a whole new market opportunity for extremely fast processing, new software—and a whole lot more silicon, tooling, manufacturing and services.