Edge computing

Edge computing means running computing where the data is produced, far from the big data centres. The "edge" is a place such as a factory floor, a shop, a vehicle, a mobile tower or a camera. The computer there is small, but it can process data on the spot and send only a short result to the cloud.

Here is how it works in general. A sensor or camera produces a large amount of data. Sending all of it to a data centre would take time and cost bandwidth, and the connection may fail. So a small computer at the same place does the first processing. It decides, for example, that nothing unusual happened, and sends nothing. Or it finds a fault and sends only a short alert. The cloud still handles work that needs more power, such as training models on many sites' data.

Edge computing has a cost. Many small computers at many sites must be installed, protected and updated, which is harder than looking after one data centre. A computer at the edge also has limited power, so models are made smaller to fit. There are five usual reasons to use it. Delay: the answer is needed in milliseconds. Bandwidth: a camera stream is several megabits a second. Cost: sending all data to the cloud is expensive. Connection: a mine or a farm may have none. Rules: some data must not leave the site. Often the work is split: the edge makes the quick decision, and the cloud stores the data and trains the model.

Take, for example, a camera at a factory gate that must recognise a vehicle's number plate within a fraction of a second. A computer next to the camera does this and sends only the number; see ANPR. Consider a mobile tower that runs video services close to users, so the video does not travel across the country. Say a driverless vehicle must see an obstacle and brake at once. It cannot wait for a reply from a distant data centre; see self-driving stack.

Related products are edge AI, industrial edge platforms and edge functions. The idea is a close companion of industrial IoT.

How is it different from the cloud? The cloud runs in large, central data centres that can do heavy work for many users. Edge computing runs in many small places near the data, to respond faster and to avoid sending everything away.