Another edge computing facet of an IoT system architecture is known as the fog. The OpenFog Consortium released in 2017 a reference architecture (https://www.openfogconsortium.org/wp-content/uploads/OpenFog_Reference_Architecture_2_09_17-FINAL.pdf), in which they defined fog computing as:
A horizontal, system-level architecture that distributes computing, storage, control, and network functions closer to the users along a cloud-to-thing continuum.
Fog nodes are, as you would expect, placed nearer to edge devices in an IoT architecture. This allows for data analytics nearer the edge, and decision-making with minimized latency (minimized reach-back to the cloud).
Of course, the cloud still plays a role in these architectures, although cloud services are used for longer term, less cyclical data storage and more robust processing capabilities than are feasible at the edge or in the fog.
Fog computing allows data captured by IoT devices to be processed nearer to the edge. This supports use cases where actions must be calculated in real time, or where preprocessing of data is required.
Fog-nodes provide computation, networking, storage, and acceleration at the network edge. This allows for data analytics at the edge (closer to the data source), and command and control with minimized latency (for example, adjusting operations based on locally sourced data).
Fog node-to-fog-node interconnects allow for interoperability across the fog layer. Placing fog nodes nearer to the devices reduces communication bandwidth and costs, and provides lower latency:
Fog nodes communicate with IoT edge devices using various protocols including ZigBee, Bluetooth Low Energy (BLTE), ZWave, and Wi-Fi and cellular.
Fog nodes are likely to be integrated directly into existing network elements such as access points, routers, and gateways. At the highest layer is the cloud where global applications and services reside.
The fog coexists with the cloud in most use cases, and often relies upon cloud services for more robust analytics capabilities.