MQTT, HTTP or CoAP: Choosing an IoT Messaging Protocol

Once a device can reach the network, it needs a way to exchange messages with the services behind it. Three protocols cover most IoT products. HTTP is universal and request-response. MQTT keeps a connection open to a broker and moves messages by publish and subscribe. CoAP brings a compact, HTTP-like model to UDP for constrained devices. This article compares them on the points that decide real projects.

At a glance

8883
The registered port for MQTT over TLS. Unencrypted MQTT uses 1883.
3 levels
MQTT quality of service: at most once, at least once and exactly once.
5683
The default UDP port for CoAP. CoAP over DTLS uses 5684.
2 bytes
The smallest MQTT fixed header, one reason its packets are compact.

The problem

HTTP is the default because every developer knows it, but it suits devices that send occasional reports and rarely need to be told anything. A device that must react to commands has to poll, which wastes energy and adds delay.

MQTT solves that with a persistent connection, but the connection needs periodic keep-alive traffic. For a device that sleeps for long periods, reconnecting and repeating the TLS handshake can cost more than the data itself.

Match the protocol to who starts the conversation and how often. That decides more than message size does.

Reading the three

MQTT runs over TCP and suits devices that stay connected and must receive commands promptly. MQTT 5 adds reason codes, message expiry and user properties, which make error handling clearer than in 3.1.1. QoS 1 suits most telemetry if handlers tolerate duplicates; QoS 2 adds a four-packet exchange few applications need.

CoAP runs over UDP with a 4-byte fixed header and suits constrained devices on lossy or low-power networks, including Thread. It offers confirmable messages for reliability and an observe option for subscriptions. Cloud support for CoAP is narrower than for MQTT and HTTP, so check the platform first.

What good looks like

Describe the traffic: who initiates, how often, payload size, and whether the device sleeps between messages.

Check what the cloud platform and network support natively. A protocol that needs a translation proxy adds a component to operate.

Prototype on the real network and measure energy and data per message, including connection setup, keep-alives and handshakes.

Three protocols at a glance
ProtocolTransportDefault portsPattern
HTTP/1.1 and HTTP/2TCP80; 443 with TLSRequest and response
HTTP/3QUIC over UDP443Request and response
MQTTTCP1883; 8883 with TLSPublish and subscribe via a broker
CoAPUDP5683; 5684 with DTLSRequest and response, plus observe
IANA-registered default ports. Cloud platforms may use others, such as MQTT over WebSockets on 443.
Packets per MQTT publish by QoS
packets
QoS 01
QoS 12
QoS 24
Control packets for one publish: PUBLISH; then PUBACK; or PUBREC, PUBREL and PUBCOMP.

What it changes

Responsiveness

Commands reach devices without polling where the product needs it.

Energy Use

Connection overhead suits how often the device wakes and talks.

Data Cost

Header and handshake overhead are known before a cellular plan is chosen.

Reliability

Delivery guarantees match what the application can tolerate.

Where to go next

Whichever protocol you choose, secure it: TLS for MQTT and HTTP, DTLS or OSCORE for CoAP. Plain-text ports belong only on isolated test benches.

Define topic or resource naming and payload formats as carefully as the protocol. They outlive the first version of the product.

Problems this prevents

  • Devices polling for commands that rarely arrive
  • Handshake costs that dwarf the payload
  • QoS 2 used where duplicates were harmless
  • A protocol the cloud platform cannot ingest

Discuss your project

Email support@aavico.com or use Get in Touch when the website finishes loading.