Edge Computing · IIoT · Latency < 10ms

Industrial edge computing: Intelligence: Where Are Your Machines?

Process the data directly on the field gateway: React in milliseconds, reduce your 4G bandwidth by 90 %, and ensure continuity even without a network.

< 10ms
Local latency
-90%
Cloud Bandwidth
Linux
Native OS on the gateway
0 dep.
Cloud for Critical Alarms
⚙️
Machine / Sensor
Modbus RTU · 4–20 mA · IEPE
🖥️
Edge Gateway
Python · Docker · Node-RED
☁️
Cloud Eziwan
Aggregates · Alerts · API
95% of the data is filtered locally
Architectural Comparison

Cloud Only vs. Edge + Cloud: Which Architecture Is Right for Your Factory?

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.

☁️
Cloud only
Traditional Architecture
Send All = Bandwidth Overloaded
100 sensors at 100 ms = 10,000 points/s → 4G will be full in minutes
Latency: 200 to 500 ms minimum
It is impossible to shut down a machine in less than 500 ms via the cloud
Loss of connectivity = loss of data
A 2-hour 4G outage = 2 hours of gaps in your history
Data-Proportional Cloud Cost
1 TB/month of raw telemetry data = a hefty cloud bill
Complex processing = increased latency
Vibration FFT calculates cloud dimension: 800 ms to 2 s response time
Edge + Cloud — Eziwan
Recommended Architecture
Local filtering: Only essential data is sent to the cloud
10,000 points/s collected → 200 events/s sent = -98% bandwidth
Local response time of less than 10 ms
Local Python script: anomaly detected → relay cuts off in 8 ms, without network connectivity
Guaranteed offline continuity
4 GB local eMMC buffer: 30 days at 1 point/min × 100 variables
Selective cloud storage = cost-effective
SIM bill reduced by a factor of 5 to 10 compared to a pure cloud architecture
FFT and ML run locally
FFT 1,024 points in 12 ms · Random Forest inference in 8 ms on Cortex-A53
< 10 ms
Local latency
vs. 200–500 ms cloud
-90 %
Bandwidth
data sent to the cloud
30 days
Offline buffer
100 variables × 1 min
x5
Cheaper
on the 4G SIM bill
Real-World Use Cases

When Edge Computing Changes Everything in Your Factory

Four specific scenarios in which processing the data at the source is the only reasonable option.

🔔
< 50 ms response time

Real-Time Anomaly Detection

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.

Local PythonGPIO RelayModbus polling at 50 msWatchdog
Machine Shutdown
< 50 ms
vs. 500 ms with cloud only
📊
Edge analytics

Vibration Preprocessing (Local FFT)

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.

FFT 1,024 ptsIEPE / accelerometerPython NumPySpectre sends only
Data Reduction
-99,7 %
raw → transmitted spectrum
🔌
Store & forward

Offline Continuity — Off-Grid Plant

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.

4 GB eMMC32 GB microSD cardStore-and-forwardNo loss
Buffer capacity
30 days
100 vars × 1 min
💳
Immediate ROI

Reducing data costs (SIM bill)

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.

Local filteringDedoublonnageHourly aggregationDelta compression
SIM Bill
divide by 5
200 MB vs. 1 GB/month
Edge Runtime

What the Eziwan Edge Gateway Can Do Locally

A true embedded Linux computer, not just a router—with all the tools an R&D engineer would expect.

🐍

Python 3.11 on the gateway

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.

import modbus_tk.modbus_tcp as modbus_tcp master = modbus_tcp.TcpMaster("192.168.1.50") val = master.execute(1, cst.READ_HOLDING_REGISTERS, 0, 10)
🔀

Node-RED for Visual Workflows

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.

Modbus-Read → [seuil > 80°C?] → oui: Relais-GPIO + MQTT-Alert → non: Buffer-Local (1 min)
🐳

Lightweight Docker containers

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.

docker run -d --name influx \ -p 8086:8086 \ influxdb:2.7-alpine # ~45 Mo RAM, ~120 Mo disque
🚨

Local Alarms Without the Cloud

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.

rules: - name: "surchauffe_moteur" condition: "temp_moteur > 80" action: - set_relay: {id:1, state:OFF} - mqtt_publish: "alerte/critique"
💾

Local buffer storage for 30 days

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.

# eMMC interne : 4 Go # micro-SD : 8/16/32 Go # SQLite WAL : 5 000 pts/s # InfluxDB Alpine: 3 000 pts/s
🔌

Local REST API (port 8080)

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.

GET http://192.168.1.10:8080/api/variables → {"temp_moteur":72.4,"freq_roteur":49.8} POST /api/control {"relay_1": false} ← coupe relais
Technical Architecture

Eziwan Edge Computing Architecture: 3 layers, a single configuration

OT, edge, and cloud work together seamlessly. You configure it once, and the entire pipeline is up and running.

Layer 1 — OT / Terrain
⚙️
PLCs
Modbus RTU / RS-485
〰️
Vibration Sensors
IEPE / Modbus TCP
Energy Meters
Modbus RTU 9600 bds
🌡️
Temperature Sensors
PT100 / 4–20 mA
🔄
Frequency converters
Modbus TCP port 502
🖥️
HMI
OPC-UA port 4840
Raw data: up to 10,000 points per second
RS-485
Ethernet
Layer 2 — Eziwan Edge Gateway
Modbus Data Acquisition
Polling interval: 50 ms → 10 ms (configurable)
Python Edge Runtime
Analysis, FFT, Local ML Inference
Node-RED workflows
Filtering, aggregation, rules
Docker containers
InfluxDB, Grafana, custom app
Local Alarms
GPIO relay—no cloud required
eMMC / SD Buffer
4 GB → 32 GB, 30 days offline
OpenVPN VPN
End-to-End Encryption to the Cloud
1.2 GHz ARM CPU · 512 MB RAM · Linux Debian 12
4G LTE
MQTT/HTTPS
Layer 3 — Eziwan Cloud
📊
Real-Time Dashboard
Centralized Multi-Site View
🔔
Alerts & Text Messages
Email, SMS, webhook, push notifications
📈
TS Historization
Unlimited, 3-year default
🔗
Cloud REST API
SCADA, ERP, Power BI
🚀
OTA deployments
Scripts, containers, firmware
🛡️
Audit & Security
Logs, roles, SOC 2
Only aggregates and alerts: -90% data
Market Comparison

Eziwan Edge vs. Other Industrial Edge Computing Solutions

Compare the leading solutions on the market based on the criteria that matter for an industrial deployment in France.

CritereAWS GreengrassAzure IoT EdgeMoxa ThingsPro EdgeEziwan Edge
OSAmazon Linux 2CBL-Mariner / LinuxDebian LinuxLinux Debian 12
Supported LanguagesPython, Java, Node.js, C++Python, C#, Java, GoPython, Node.js, JavaPython 3.11, Node.js, Node-RED, WASM
Native OT ProtocolsNo (using lambdas)No (third-party modules)Modbus, DNP3, SNMPModbus RTU/TCP, OPC-UA, BACnet, SNMP, IEPE, 4–20 mA
System RequirementsAWS Account + CloudFormationAzure Account + ARM TemplatesLocal Console + SSHIntuitive Web Interface + ZTP
Built-in 4G LTENo (separate hardware)No (separate hardware)Optional (module)Yes (built-in dual SIM)
Built-in VPNAWS Site-to-Site VPNAzure VPN GatewayOptional OpenVPNBuilt-in OpenVPN, zero configuration
French SupportNo (EN support)No (EN support/level)No (Asia/US support)Yes — FR team, office hours
Estimated costVariable + IoT Core + EC2Variable + IoT Hub + VM800–1,500 EUR + softwareAll-inclusive monthly plan
Protocols and Integrations

Industrial protocols natively supported at the edge

Incoming from the field, outgoing to the cloud or monitoring tools.

⬇️
Input Protocols — OT to Edge
Modbus RTURS-485, 9,600–115,200 bds
Modbus TCPEthernet, port 502
OPC-UAPort 4840, S&S
BACnet/IPBuildings, HVAC
SNMP v2/v3Network, Infrastructure
4-20 mAAnalog Process
0-10 VGuidelines, Measures
Pt100 / Pt1000Temperature (2-, 3-, or 4-wire)
IEPE / ICPVibration Accelerometers
PulsekWh, m³, kg, cycles
RS-232Analyzers, Legacy Series
CANbusVehicles, Mobility
⬆️
Output Protocols — Edge to Cloud / IT
MQTT 3.1.1 / 5.0QoS 0/1/2, TLS 1.3
HTTPS REST APIJSON, webhooks
InfluxDB v2Line Protocol, Flux
GrafanaDirect data source
AWS IoT CoreGreengrass-compatible
Azure IoT HubDevice SDK integrates
Google Cloud IoTMQTT bridge
OPC-UA ServerEmbedded Edge Server
AMQP 1.0Enterprise Messaging
WebSocketReal-Time Dashboard
Email / Text MessageAlarmes via modem GSM
Kafka (client)Streaming Analytics
Compatible with the ecosystem you're already using
Siemens TIA PortalSchneider EcoStruxureRockwell Studio 5000Kepware OPCIgnition SCADAPower BIGrafana CloudNode-REDHome AssistantThingsBoardDatadogElasticsearch
Frequently Asked Questions

Everything You Need to Know About Industrial Edge Computing

Questions that systems architects, IIoT project managers, and R&D engineers ask before choosing an edge solution.

Edge Computing for IIoT · Available in 1 week

Try out edge computing on your existing equipmentin 1 week

The Eziwan Gateway comes preconfigured and is up and running in 30 minutes. Connect it to your RS-485 bus or industrial Ethernet, define your edge rules and Python scripts—and see the difference in latency and data costs as early as the first week.

Delivery within 3 business daysTechnical support in French includedNo commitment, minimum 12 months

Included: Eziwan Gateway + multi-carrier SIM + cloud platform access + 1-hour onboarding session with a French engineer