← Back to the chapter · Previous detail

Technical detail · AMQP Core: Routing and Message Lifecycle

Cost assumptions behind the protocol comparison

The illustrative client-library ranges are 10–50 kB for MQTT and 100–500 kB for AMQP. These depend on the implementation and are not device memory budgets. MQTT’s fixed header can start at 2 bytes. The source’s 8-byte AMQP framing allowance comprises a 7-byte AMQP 0-9-1 frame header and a 1-byte frame-end marker; neither number is a complete connection or message cost.

The e-commerce example routes orders.us.priority-high to US fulfillment with priority handling. A deliberately broad MQTT subscription orders/# sends all orders to each subscriber for local filtering: assumed 10 ms latency and 100K messages/s per subscriber. AMQP bindings orders.us.* and orders.eu.* select US and EU queues: assumed 5 ms routing and 10K messages/s per queue. That is a 90% per-destination reduction for the assumed geographically partitioned workload.

These are example assumptions, not benchmark results or an inherent MQTT penalty. More selective MQTT subscriptions would change its delivery count too. The architectural question is which broker-side selection and queue behavior the application needs.

← Back to the chapter · Previous detail