Direct Answer
Why can't the meters connect straight to the cloud? The two protocol segments each serve their own master: Modbus is the field bus master-slave polling protocol — reliable but unfit for massive long-lived connections; MQTT is the Internet publish/subscribe protocol — lightweight and NAT-piercing; the gateway in the middle performs protocol conversion, data-model mapping and caching, letting "the field" and "the cloud" each speak their own language — the two cannot replace each other. The ESX-0223-GR, CW-C2 and FG-0221-ER ship with "Modbus down, MQTT up" as the default protocol stack.
A scene: a park had 200 meters, and someone asked "can't the meters just connect to WiFi and go straight to the cloud?" — try it: Modbus master polling takes about 50 ms per meter, 200 meters is 10 seconds per full round; with meters going straight to the cloud, the cloud has to hold a long connection to each one — 200 meters is 200 connections, and reconnect and catch-up logic becomes a mess — the field bus cannot do the cloud's job, and the cloud protocol cannot do the field's job.
A second scene: a project wired meter RS485 cables to serial-to-WiFi modules and sent them "to the cloud" — the cloud connected, but a polling round took 40 seconds (300 meters), every disconnection lost 40 seconds of data, and field electrical noise corrupted the bus — the project nearly died. After switching to gateways: local polling (1 second per round), MQTT push (real-time), disconnect caching with catch-up — the same site, two different outcomes.
1. Modbus: The Field Bus Master-Slave Polling
Master-slave mechanism — one master (the gateway) polls each slave (meters, instruments) in turn, one question one answer, slaves respond passively; Register mapping — data lives at register addresses on the slave (e.g., 40001 voltage, 40003 current), and the master reads/writes by address; RTU vs TCP — RTU over RS485 (binary frames with CRC), TCP over Ethernet (frames with IP headers); Reliability — the question-answer pattern is naturally acknowledged, retries are simple, but the round time grows with the number of slaves.
In one sentence: Modbus is "roll call" — reliable, simple, mature, but one at a time; it slows down as the device count grows.
A register example: a switch-room meter had voltage at register 40001, current at 40003 and power at 40005 — the gateway polls once per second and reads out the register data. In the gateway's eyes, the field is "a set of addresses plus a set of values" — Modbus's value is simplifying complex devices into addressable data tables.
Why polling is enough: field data changes at minute-level rates (temperature, power, voltage), and 1-second polling is far faster than the rate of change; polling only "feels slow" when the device count is large (200 meters, 10 seconds a round). Modbus's design philosophy is "good enough" — field data does not need millisecond refresh in the first place.
2. MQTT: The Internet Publish/Subscribe
Publish/subscribe — devices (clients) publish messages to topics, and the cloud subscribes to receive — decoupling publishers from subscribers; QoS levels — QoS0 at most once (may drop), QoS1 at least once (may duplicate), QoS2 exactly once (most reliable); Lightweight and NAT-piercing — a TCP long connection with heartbeat keep-alive pierces NAT/firewalls, and massive devices each connect without interfering; Topic design — e.g., park/switchroom/meter_01/voltage, organized by "park/switchroom/device/data", and the cloud subscribes by topic.
In one sentence: MQTT is "broadcasting to whoever listens" — publishers do not care who is listening and subscribers do not care who is publishing; massive devices naturally suit long connections.
A topic example: 200 meters in a park, topics named park01/roomA/meter001/data — the cloud subscribes to park01/# to receive everything and subscribes precisely to meter001 to watch a single meter. Topic design is the data's "postal code" — designed well, cloud queries and alarm routing are all easier.
How to choose QoS: routine telemetry (voltage, current) is fine at QoS0 (a dropped frame is harmless); important events (alarms, threshold crossings) use QoS1 to guarantee at least once; metering data (cumulative energy) uses QoS2 to avoid double-counting. Choose QoS by data value — QoS2 everywhere is not the safest; QoS2 has the highest overhead and needs extra deduplication on duplicate delivery.
3. Inside the Gateway: Protocol Conversion and Data-Model Mapping
Protocol conversion — read Modbus frames down the stack, wrap MQTT messages up the stack, unified into an internal data model in the middle; Data-model mapping — register addresses (40001) map to semantic fields (voltage), device IDs map to topic paths; Caching and catch-up — when the network is down, data is stored locally first and retransmitted in order after recovery — local caching is the gateway's core value; Edge pre-processing — some alarms are computed locally (e.g., threshold checks), independent of the cloud — alarms still fire while disconnected.
In one sentence: the gateway is an "interpreter plus warehouse manager" — translating protocols, mapping models and caching data, so the field and the cloud can cooperate while speaking different languages.
Why cache limits matter: the longer the outage, the more data piles up — gateway storage is finite, so cache policy must prioritize: important alarm events are kept first, routine telemetry rolls over by time window. The cache is not an "unlimited ledger"; it is a "ledger that prunes by value" — life-saving data survives an outage intact, and routine data is caught up after.
A conversion example: a meter's register 40001 holds a raw voltage value of 2310 (actually 231.0 V) — the gateway reads it, divides by 10 into 231.0, and wraps it into an MQTT message published to the topic park01/roomA/meter001/voltage. The cloud receives not "40001=2310" but "voltage=231.0V" — the end of protocol conversion is semantic consistency.
4. Why the RS485 Bus Is Not Meant to Go Straight to the Cloud
Bus length limit — RS485 bus reaches 1200 meters at most; a park spanning buildings is already out of range; Polling bottleneck — 200 devices take 10 seconds per round, and the cloud can only be as real-time as the polling cycle; No NAT traversal — RS485 is a local bus protocol with no IP address and no routing; it cannot reach the public network by design; Isolation needs — the field bus lives in electrical-interference-heavy environments; sending it straight to the cloud exposes field risk to the public network.
In one sentence: RS485 is an "intercom inside the building" — fine locally, but across buildings, to the cloud, or for real-time, forget it; the cloud's job must go to an IP protocol.
A bottleneck example: a factory wanted real-time current data from 300 devices — an RS485 polling round took 15 seconds, so the cloud could only refresh at 15-second intervals — wanting 1-second? Switch to segmented gateway polling (50 devices per segment) plus MQTT push, and the cloud is at 1-second immediately. Polling is Modbus's nature, not its fault — keep it in the "field segment" and let MQTT carry the "cloud segment"; each does its own job.
About "meters with built-in Ethernet ports": some meters support Modbus TCP over a switch — that still does not mean they can go straight to the cloud: TCP is a transport protocol, not an application protocol; the cloud wants MQTT messages, and the meter only speaks Modbus — an interpreter is still needed in the middle for protocol conversion. Network reachable ≠ protocol compatible — with an IP address but no shared language, cooperation is still impossible.
5. IEC 61850: The Optional Slot for Substation-Grade Scenarios
What IEC 61850 is — the communication standard for substation intelligent devices, built around device modeling and interoperability, far more complex than Modbus; Its place — substation-grade scenarios (protection, measurement-and-control, station-level systems) use IEC 61850; park-grade meters and instruments use Modbus; The gateway's role — an IEC 61850-capable gateway converts station data to MQTT for the cloud, with two languages inside the station and out; Selection judgment — higher grade is not automatically better: Modbus is enough for meter points; forcing 61850 on them is overkill.
In one sentence: IEC 61850 is the "substation language" — park meters use Modbus and only the substation station control needs 61850; the gateway translates both field languages into cloud MQTT.
A selection example: a project needed IEC 61850 at the substation station-control layer and Modbus for 200 meters outside — the ESX-0223-GR (dual-protocol downlink) was chosen: 61850 inside the station, Modbus outside, both converted to MQTT. One gateway speaks three languages at once — however complex the field, the cloud only sees MQTT.
6. Landing Advice: How to Choose the Gateway
Look at the downlink — how many field protocols it supports (Modbus RTU/TCP, DL/T 645, IEC 61850), and whether register mapping is configurable; Look at the uplink business — MQTT QoS levels supported, catch-up duration, local cache size — these decide whether data survives an outage; Look at the data model — whether field mapping is semantic (voltage instead of 40001), and how much secondary development cloud integration needs; Look at edge computing — local threshold checks and alarm pre-processing decide whether the site is still guarded while disconnected.
In one sentence: choosing a gateway is not choosing a "box"; it is choosing the "translation plus cache plus edge" trio — downlink covers the field, uplink MQTT connects the cloud, and local cache backs up outages.
A selection example: a park needed to connect 300 meters, 20 switch-room instruments and a substation station-control system — the checklist: downlink Modbus RTU/TCP (meters and instruments) plus IEC 61850 (station control); uplink MQTT QoS1; cache designed for a 24-hour outage. Before choosing, answer three questions — "what protocols are in the field, what format does the cloud want, how long must an outage be survived" — and selection stops being a gamble.
Back to FEXLINK: the ESX-0223-GR, CW-C2 and FG-0221-ER ship with "Modbus down, MQTT up" — the downlink covers RTU/TCP and mainstream field protocols, the uplink supports MQTT QoS1 with catch-up, and local cache scales on demand. Ask "what does your gateway support down, what up, and how does it catch up after an outage" — a vendor answering "Modbus + MQTT + cached catch-up" treats the protocol stack as engineering; hesitation means the protocol stack was never a first-class concern.
Scope and Limitations
- The conclusions in this article are intended for LV distribution and electrical-safety sites. Standard numbers (DL/T 645) refer to their currently effective editions, and product parameters follow the latest FEXLINK selection handbook and product knowledge base.
- Quantitative criteria (protection zone, discharge and residual voltage, earth resistance, thresholds and grading) hold under the preconditions stated in this article. Where the site voltage level, earthing arrangement, environmental category or load profile differs, the figures must be recalculated. Specific configurations involving ESX-0223-GR、CW-C2、FG-0221-ER should be verified against actual circuit parameters.
- Specific product models and parameters are subject to the latest selection handbook and product knowledge base; this article does not constitute selection or procurement advice.
- Standard numbers cited here refer to the currently effective editions; when a standard is revised, the related quantitative conclusions must be re-checked.
- Deployment and implementation for specific projects must be carried out by qualified engineers based on site survey and design drawings.
Frequently Asked Questions
Modbus Downstream, MQTT Upstream
The two protocol segments each serve their own master: Modbus is the field bus master-slave polling protocol — reliable but unfit for massive long-lived