Serving millions of clients with Native MQTT and MQTT 5.0
summary
RabbitMQ replaces AMQP proxying with Native MQTT, cutting memory by up to 95%, lowering end‑to‑end latency 50–70% and raising throughput 30–40%, enabling millions of concurrent connections (a 9M connection test used ~56 GB per node). Connection handling shifts from ~22 Erlang processes to one per client; a new qs0 q option skips the queue process for clean‑session QoS0 clients (stateless, higher fan‑out, drops on backpressure). The upcoming MQTT 5.0 adds message/session expiry, reason codes, topic aliases and user properties mapped to AMQP headers.
AI-written summary · about
chapters (7)
- 0:00MQTT introduction
- 3:56RabbitMQ core model
- 8:51Native MQTT in 3.12
- 15:14Optimizing MQTT with QS0
- 20:14MQTT 5.0 features
- 26:20Future MQTT work
- 28:07Q&A
related talks
19mBreaking Storage Barriers: How RabbitMQ Streams Scale Beyond Local Disk
43mRealtime performance at scale (...)
41mPushing RabbitMQ to the limit
33mReaching Ludicrous Speed
40mDistribution, Scale and Flexibility with ZeroMQ
35mVaronis journey deploying RabbitMQ
33mConversational Big Data
58mLong road from oil pipeline sensors up to millions of TV’s