Applying Edge Computing in IoT Systems
As device fleets grow, the "send all data to the cloud, run all logic in the cloud" architecture starts to show its problems: high bandwidth costs, loss of control during network outages, and failing real-time requirements. Edge computing was born to solve exactly these problems.
Scenario 1: Protocol Conversion
Industrial sites use a wide variety of device protocols (Modbus, OPC-UA, BACnet...), and no cloud platform can natively support every one of them. The edge gateway performs protocol parsing and normalization locally, then uploads data to the cloud in a unified MQTT + Thing Model format, so the cloud only has to deal with a single "language".
Scenario 2: Local Linkage
Linkages like "flame sensor triggered → shut off the gas valve immediately" are extremely latency-sensitive. A round trip to the cloud can take hundreds of milliseconds or even seconds, and it depends on network availability. Pushing the linkage rules down to the edge gateway for local execution delivers millisecond-level response and keeps working even when disconnected — state is synced back to the cloud once the network recovers.
Scenario 3: Data Cleaning and Downsampling
A single vibration sensor can produce thousands of samples per second; uploading them raw wastes both bandwidth and storage. Aggregating on the edge first (averages, peaks, feature extraction) before reporting can reduce uplink bandwidth by an order of magnitude or more, and eases the load on the time-series database as well.
Principles for Dividing Work Between Edge and Cloud
- The edge handles: work that is latency-critical, protocol-related, or involves large volumes of raw data;
- The cloud handles: the global view, historical analysis, cross-site coordination, visualization, and management;
- Offline autonomy: the edge must be able to keep core business running while offline — this is the bottom line of the architecture design.
FastBee's device-side firmware and gateway access solutions follow the same division of labor. If you are interested, see the Device documentation.