Technical detail · AMQP Core Architecture: Exchange Types
A 50-machine routing worksheet
A second factory worksheet has 50 CNC machines. Maintenance binds factory.*.*.error and factory.*.*.warning; the production dashboard binds factory.#; machine-learning analytics binds factory.line3.*.metrics. Example publications are factory.line1.cnc001.metrics, factory.line3.cnc042.error and factory.line2.cnc015.warning.
For factory.line3.cnc042.error, maintenance’s two stars match line3 and cnc042, and its error suffix matches. Its warning binding fails. The dashboard matches because # accepts the remaining words. Analytics fails because error is not metrics. One publish therefore reaches two queues instead of fanout’s three, a 33% reduction for that message.
At 120 messages/hour per machine, 50 × 120 gives 6,000 publishes/hour. The worksheet assumes a traffic mix averaging 1.4 queues per message, giving 8,400 topic deliveries. Fanout gives 6,000 × 3 = 18,000 deliveries, about 114% more than topic routing. Topic reduces broker-to-consumer traffic by about 53% under that mix; the single error example alone does not establish the hourly average.