I'm working on a project with about 5,000 sensors reporting temperature and vibration data every 10 seconds. My manager wants to use Edge Computing to process data locally before sending it to the cloud. I feel like it's overkill and we should just stream everything to Kinesis. Am I underestimating the bandwidth costs, or is "Edge" just a buzzword for this scale of data?
3 answers
At 5,000 sensors every 10 seconds, you're looking at millions of messages a day. While a single temperature reading is tiny, the "overhead" of the HTTP/MQTT headers and the ingestion costs in the cloud (like Kinesis Shard hours or IoT Core per-message fees) add up fast. The real value of Edge here isn't just cost; it's "data reduction." If 99% of your data is "Normal," why pay to send it? An Edge gateway can run a simple anomaly detection script and only alert the cloud when the vibration exceeds a threshold. This drastically reduces your storage and processing costs in the long run.
What happens if your internet connection goes down? Does the system need to trigger a local shutdown if the vibration gets dangerous, or is this just for long-term historical analytics?
Edge isn't just about latency; it's about privacy too. If any of that sensor data is sensitive, processing it locally means it never has to leave your facility.
Exactly, Susan. Compliance is a huge driver for Edge. By the time the data hits the public cloud, it's already been filtered or masked, which simplifies your security audits
Thomas, that's the clincher. If there is a "safety" component, you can't rely on the round-trip to the cloud. We had a similar setup in a factory where we needed sub-millisecond response times to stop a motor. We used AWS IoT Greengrass on a local industrial PC. It handles the local logic and then syncs the "boring" telemetry data to the cloud whenever the bandwidth is cheap or available.