What is edge computing?

When people talk about edge computing, you hear a lot about self-driving cars, autonomous robots, and automated retail. But my favorite example of edge computing is from a fast-food chain.

Every restaurant location runs analytics on smart kitchen equipment data to make decisions like exactly when to put the fries in the fryer for perfect crispiness. They use edge computing to hyper-personalize these kinds of actions for each store.

The company can create a forecast in the cloud to predict how many waffle fries should be cooked per minute over a day — easy when using transactional sales data.

But it’s at the edge where each store micro-adjusts the initial forecast with specific on-site, real-time data from their kitchen and point-of-sale systems. Using compute at the edge is how they can make sure everyone’s fries are crispy, whether it’s a slow afternoon or a crush of families after a little league game.

Delivering services quickly with a personal touch. That’s what edge computing can do.

Does edge mean the end of cloud computing? Definitely not! Not only is cloud computing a critical component in managing the edge, but edge computing is going to drive the next wave of cloud computing.

What is edge computing and how is it different from cloud computing?

Edge computing is a new capability that moves computing to the edge of the network, where it’s closest to users and devices — and most critically, as close as possible to data sources.

By contrast, in cloud computing, data is generated or collected in many locations and then moved to the cloud, where computing is centralized. Centralized cloud computing makes it easier and cheaper to process data together and at scale. But there are times when it doesn’t make sense to send data off to the cloud for processing, like in the following scenarios:

• There’s no internet, or the signal is limited, like on an oil rig using a satellite connection in the middle of the ocean.
• The data can’t be transferred off-site because of security concerns or privacy regulations.
• When a device needs to analyze data and make split-second decisions, like with robotics surgery. In that case, even a second or two of latency means sending data to the cloud and waiting for a decision isn’t an option.

Advantages of edge computing

Sometimes, clients ask me what makes edge different. The main benefit of edge computing is reducing the risk of network outages or cloud delays when highly interactive — and timely — experiences are critical. Edge enables these experiences by embedding intelligence and automation into the physical world. Think optimizing factory operations in a factory, controlling robotic surgery on a patient, or automating production in a mine.

And if super-speed and reliability are not convincing enough, I usually follow up with three more unique attributes of edge:

1. Unparalleled data control: Edge is the first point where compute taps into the data source and determines how much of the original fidelity is preserved when digitalizing the analog signal. Here’s where we implement what data is stored, obfuscated, summarized and routed. It’s also the point where we can add controls to address data reliability, privacy and regulations.

For example, when doing facial recognition to unlock a smartphone, it’s better to keep data at the edge. The AI models are trained for each user’s face without these images ever leaving the device. Since data is never transferred beyond our phones, it preserves our privacy and avoids security breaches in the cloud.

2. Favorable laws of physics: Edge is always on and has low latency thanks to reduced network uptime, round-trip times and bandwidth constraints.

For example, my team and I implemented a visual analytics algorithm in a factory production line to find defects in car seat manufacturing. As the seats moved down a production line, we deployed our low-latency deep learning inferencing models at the edge to automate defect detection in real-time. The solution keeps pace with the uptime and production line speed, which only edge computing could allow.

3. Lower costs: Processing at the edge makes cloud upload and storage cheaper. Why pay for full-fidelity data when a summarized view or key insights might be all you need?

I saw the cost-saving power of edge when I worked on my first edge implementation. It was an oilfield company whose oil wells were only accessible over-the-air — some via satellite and others only by helicopter.

Data storage was limited and immediate transmission of data was costly — if it was available at all. We had already been doing analytics on the oil well data, and our next step was to deploy some of these modules directly on the well.

We used edge computing to preserve data fidelity and optimize what was stored and transmitted. This way, we could still do rich analytics and keep the most important (and worth-the-cost) data.

Will edge computing replace cloud computing?

Not at all. Even with these amazing benefits, edge will not replace cloud computing.

For one thing, edge capacity is limited because edge reintroduces resource constraints on battery, bandwidth, storage and computing power. Not everything can run at the edge, I always say.

Instead, think of edge and cloud as part of a computing continuum. Cloud sits at the center and edge complements it, as it radiates out toward the “ends” of a network.

Here are three more reasons edge will not replace cloud computing:

1. Centralized, co-located cloud computing is still needed for performance and cost. Cloud’s data and enterprise app gravity is already big and is poised to grow. Accenture CTO Paul Daugherty predicts that “with most businesses currently at only about 20% in the cloud, moving to 80% or more rapidly and cost-effectively is a massive change that requires a bold new model.” Cloud will integrate with data and computed insights from the edge, and spur new apps that will be deployed at the edge.

2. Edge computing data is feeding into more AI, which in turn needs cloud more than ever. The inferencing that might happen on the edge starts with bringing together data for experimentation and model training. And that takes a lot of computing power. Cloud remains the best solution when we need to combine edge, enterprise and third-party data for discovery and AI model creation.

3. Edge is an extension of cloud and requires a common platform-based approach: Adding new technologies like edge to existing cloud platforms makes it much easier to manage and optimize applications.

The future is a new cloud continuum

Cloud and edge computing are distinct but complementary. Centrally, cloud brings data together to create new analytics and applications, which will be distributed on the edge — residing on-site or with the customer. That, in turn, generates more data that feeds back into the cloud to optimize the experience. I call that balance in a virtuous cycle.

New edge applications that create highly contextualized and personalized experiences are sure to come. It will be hard to top the crispy fries use case, though.