The choice of architecture affects latency, data costs, and resilience in the event of a network outage. That is why manufacturers are migrating to the edge.
Four specific scenarios in which processing the data at the source is the only reasonable option.
An abnormal current spike in a motor or a temperature deviation must trigger an immediate shutdown. With the cloud, the entire cycle (acquisition → transmission → processing → feedback control) takes more than 500 ms. The edge gateway runs a local Python script that continuously monitors the Modbus registers and trips the protective relay in less than 50 ms—without waiting for the cloud and without relying on the network.
An IEPE accelerometer mounted on a bearing generates 25,600 samples per second. Sending the raw data to the cloud at a continuous rate of 200 KB/s would fill up a 4G plan in just a few hours. The edge gateway calculates the FFT on 1,024 points in 12 ms, extracts the frequencies characteristic of the fault (rotational order harmonics, BPFO, BPFI) and sends only the reduced spectrum to the cloud—64 values every 10 seconds. Data reduction: 99.7% of the original data volume.
In factories with unstable 4G coverage, dead zones, or during production shutdowns, connectivity is not guaranteed. The Eziwan gateway stores all data acquisitions on the internal eMMC memory (4 GB) or an optional micro-SD card (32 GB). Local alarms, control sequences, and safety rules continue to function normally. Upon reconnection, the missing 2 hours, 8 hours, or 30 days of data are automatically uploaded in chronological order.
An industrial 4G plan with 1 GB/month becomes insufficient as soon as you have more than 20 sensors collecting data every second. Edge filtering solves the problem at its root: only anomalies, hourly aggregates, and state-change events are sent to the cloud. 80% of the data is processed and discarded locally. The result: a SIM bill reduced by a factor of 5, with data usage of 200 MB/month instead of 1 GB for the same level of information.
A true embedded Linux computer, not just a router—with all the tools an R&D engineer would expect.
Run your analysis scripts directly on the gateway. Python 3.11 with pip, numpy, scipy, pandas, scikit-learn, and paho-mqtt pre-installed. Direct access to Modbus serial ports, GPIOs, and TCP sockets. Automatic startup via systemd.
Node-RED is natively integrated into the gateway. Visually build data pipelines: Modbus reading → filtering → local alarm → MQTT publishing. Interface accessible via a browser on the local network (port 1880). Pre-loaded industrial nodes.
Docker Engine with support for Alpine-based images. Deployment of edge microservices: local InfluxDB database, embedded Grafana, Eclipse Mosquitto MQTT broker, or your business application. Volume persistence on eMMC or micro-SD.
Define alarm rules in YAML or via the Eziwan interface. If temperature > 80 degrees C → turn off output 1 relay. If pressure < 0.5 bar → send an SMS via the built-in GSM modem. If vibration_rms > 12 mm/s → create a local event and log it. Everything operates with zero cloud dependency.
Embedded time-series database (SQLite or local InfluxDB). Writes up to 5,000 data points per second to eMMC. Configurable circular buffer ranging from 1 hour to 90 days, depending on capacity. In the event of reconnection, cloud backfill resumes exactly where the interruption occurred, without duplicates.
Local JSON REST API exposed on the industrial network at port 8080. Your PLCs, HMIs, or SCADA software can read the latest values and write setpoints without going through the Internet. Token-based authentication. Endpoints: GET /api/variables, POST /api/control, GET /api/alarms.
OT, edge, and cloud work together seamlessly. You configure it once, and the entire pipeline is up and running.
Compare the leading solutions on the market based on the criteria that matter for an industrial deployment in France.
| Critere | AWS Greengrass | Azure IoT Edge | Moxa ThingsPro Edge | Eziwan Edge |
|---|---|---|---|---|
| OS | Amazon Linux 2 | CBL-Mariner / Linux | Debian Linux | Linux Debian 12 |
| Supported Languages | Python, Java, Node.js, C++ | Python, C#, Java, Go | Python, Node.js, Java | Python 3.11, Node.js, Node-RED, WASM |
| Native OT Protocols | No (using lambdas) | No (third-party modules) | Modbus, DNP3, SNMP | Modbus RTU/TCP, OPC-UA, BACnet, SNMP, IEPE, 4–20 mA |
| System Requirements | AWS Account + CloudFormation | Azure Account + ARM Templates | Local Console + SSH | Intuitive Web Interface + ZTP |
| Built-in 4G LTE | No (separate hardware) | No (separate hardware) | Optional (module) | Yes (built-in dual SIM) |
| Built-in VPN | AWS Site-to-Site VPN | Azure VPN Gateway | Optional OpenVPN | Built-in OpenVPN, zero configuration |
| French Support | No (EN support) | No (EN support/level) | No (Asia/US support) | Yes — FR team, office hours |
| Estimated cost | Variable + IoT Core + EC2 | Variable + IoT Hub + VM | 800–1,500 EUR + software | All-inclusive monthly plan |
Incoming from the field, outgoing to the cloud or monitoring tools.
Questions that systems architects, IIoT project managers, and R&D engineers ask before choosing an edge solution.