Fog Computing

A smart city has 100,000 sensors: traffic cameras, air quality monitors, water pressure gauges, street lights, parking meters. Each generates continuous data. Sending all of it to the cloud would consume enormous bandwidth and cost a fortune. A single 1080p traffic camera produces 2-5 Mbps of continuous video. Multiply by 10,000 cameras and you need 20-50 Gbps of sustained upstream bandwidth, just for cameras alone. At AWS egress prices, that is hundreds of thousands of dollars per month.

Processing it all on-device is equally impractical. These are small, cheap sensors with minimal compute power. A $15 air quality sensor cannot run a machine learning model. A traffic camera can maybe run a basic motion detector, but not cross-intersection traffic flow analysis.

Fog computing fills the gap. It is a layer of compute nodes between the edge devices and the cloud. A fog node might be a gateway device at a street intersection that aggregates data from 50 nearby sensors, filters out noise, runs local analytics ("is traffic flowing normally at this intersection?"), and sends only meaningful events to the cloud ("traffic jam detected at coordinates X,Y, estimated duration 20 minutes").

Cisco coined the term "fog" in 2012. Fog is cloud that is closer to the ground. The concept is specifically relevant in IoT, industrial automation, and smart infrastructure deployments where thousands or millions of simple devices generate high-volume data that needs local aggregation and real-time processing before (or instead of) cloud transmission.