← Back to the chapter · Next detail →

Technical detail · MQTT Production Operations

Memory worksheet and two-node availability assumptions

An illustrative MQTT deployment scaled from the house’s client roles sizes broker memory and compares a two-node availability estimate with measured recovery.

A broker accepts publications and routes them to subscribers; two nodes are two running broker instances. The scaled workload’s memory worksheet divides 4,000 device connections across two nodes. It assumes four decimal kilobytes per connection, a 100-message buffer of 200-byte payloads, an allowance for the routing table that maps subscriptions to recipients and illustrative runtime overhead. Its subtotal is 248.256 megabytes per node before omitted costs. Testing two 1 GB nodes is a starting point, not evidence of an operational margin.

A single node might carry the workload. A second becomes an availability candidate only when required state and delivery recover through it. The idealized 99.9975 percent two-node availability figure assumes 99.5 percent availability for each node, independent failures and instantaneous failover. Shared power, network or storage dependencies and recovery delay undermine those assumptions. The worksheet target and calculated result must remain distinct from a measured service result.

Record connection limits, incoming publications, outgoing deliveries, the start and end events used to measure latency and its percentiles, memory and queue behavior. Include a reconnect burst and surviving-node saturation. Keep Transport Layer Security (TLS), payload, delivery level, fan-out (outgoing copies per publication) and persistence (stored-state retention) settings with every benchmark so another reviewer can interpret it.

← Back to the chapter · Next detail →