Edge Computing for IoT

When to use

• Designing edge gateway architecture and selecting hardware (Raspberry Pi to industrial)
• Deciding what to process at the edge vs offload to the cloud
• Deploying ML inference models on constrained edge hardware
• Containerizing edge workloads with K3s or Azure IoT Edge modules
• Implementing edge-cloud data sync and conflict resolution strategies
• Building systems that must function during internet outages

Core principles

1. Intermittent connectivity is the default — design store-and-forward from day one; connectivity is the exception, not the guarantee
2. Edge runs inference, cloud runs training — push optimized models down, pull raw data and drift metrics up
3. Protocol translation lives at the gateway boundary — one cloud-side identity, many device-side protocols; canonical data model applied at the seam
4. Pin container versions on edge devices — latest in a factory is a production incident waiting to happen
5. 100ms is the latency ceiling for control loops — anything safety-critical or real-time stays local regardless of network quality

Reference Files

• references/gateway-and-processing.md — gateway responsibilities, hardware selection guide, edge vs cloud processing decision criteria, hybrid pattern
• references/edge-ml-and-containers.md — TensorFlow Lite / ONNX / TensorRT frameworks, ML deployment pipeline, K3s, Azure IoT Edge, container best practices
• references/sync-and-offline.md — data synchronization strategies, conflict resolution, offline operation with local storage and RTC drift handling