What is Edge Computing?

Edge computing means doing the work on the device that gathers the data, a phone, a camera, a sensor on a factory floor, rather than shipping it to a distant data centre. It is quicker, it survives a dropped connection, and the raw data never leaves that device. The trade is power, because such devices are far weaker than the servers they replace.

The default arrangement sends data to a data centre, has it processed there and sends the answer back, which is fine whenever nobody notices the round trip. It stops being fine in three situations. First, when the delay matters, for instance a robotic arm on a production line that must stop the moment a hand crosses the safety line. Second, when the connection is unreliable, because a sensor at a remote pumping station still has a job to do when the link drops. Third, when the data should not travel at all: a smart speaker that recognises its wake word on the device itself sends nothing to anyone until you have actually spoken to it. Doing the work locally answers all three, and the price is power, since the device is far weaker than a server rack. That is exactly why shrinking techniques such as quantization and distillation matter so much here. Most real systems end up dividing the work, keeping anything urgent or private on the device and sending the rest away whenever the connection allows.

A supermarket till that can still take payments when the network drops. Everything the till needs to price your basket and open the drawer lives in the machine, so the queue keeps moving during an outage. The head-office systems still get the sales figures, the stock counts and the reordering, just later and in a batch. The device handles what cannot wait, and head office handles what can.