infrastructure//edge computing

Edge computing is the practice of placing computation near the data source, user or physical process it serves, so that a decision does not have to travel to a distant machine and back. That may mean a regional point of presence, a factory gateway, a vehicle or the device holding the sensor itself.

Edge computing is the practice of placing computation near the data source, user or physical process it serves, so that a decision does not have to travel to a distant machine and back. That may mean a regional point of presence, a factory gateway, a vehicle or the device holding the sensor itself.

Proximity can reduce round-trip latency, bandwidth consumption and dependence on continuous connectivity. It also fragments deployment across smaller, heterogeneous and physically exposed machines. The word edge describes where computation sits relative to consequences, never one vendor runtime or ownership model (Workers implement one form of it, at the network edge, by running code in many distributed locations).

In a machine the placement follows the loops. On a drone the fast loop runs on the flight controller's MCU, vision on the companion computer and fleet analytics in the cloud; in a plant, features such as RMS, peak and band energy are computed next to the vibration sensor and sent every few minutes, because one raw channel at 25.6 kHz is about 4.4 GB a day. Edge deployment of learned models follows the same economy: a network is trained once on a data-centre GPU and run millions of times on small devices, where memory, energy and latency rule, usually after quantization and with the converted model validated again (training-serving skew). The deciding quantity is the deadline: a decision that must arrive within the round trip to a server has to be made where the data is (embedded system).

The word covers two places that are easy to confuse. The network edge is a provider's points of presence close to users, where a CDN caches pages and serverless code answers web requests in a few milliseconds (Workers). The industrial edge is next to the machine: a fanless industrial PC or edge gateway in the cabinet, rated for heat, vibration and dust, reading the PLC, running a vision model on the line and sending results upward. The first saves a user's round trip; the second keeps a plant running and deciding when its link to the cloud fails, while the PLC keeps the safety-critical control.

The cloud brings data to a large machine. The edge asks whether the decision should travel at all.