Need help interpreting RabbitMQ PerfTest results

[dpeger]

I ran several throughput tests on an Amazon MQ hosted RabbitMQ broker (RabbitMQ 3.12.13, clustered setup of 3 mq.m5.large instances). And the test results are somehow hard to interpret (at least for me).

The last tests I conducted had a fixed producer rate limit of 3.000 msg/s with 30 producers and a message size of 512 bytes, publishing persistent messages to a topic exchange for about 5 minutes straight:

[{
  'name': 'producer-limit-3k-512b-persistent-topic',
  'type': 'simple',
  'uri': 'amqps://...',
  'params': [{
    'time-limit': 300,
    'producer-count': 30,
    'consumer-count': 1,
    'producer-rate-limit': 100,
    'min-msg-size': 512,
    'flags': 'persistent',
    'exchange-type': 'topic',
    'exchange-name': 'amqp.topic'
  }]
}]

And the results look very heterogenous (apart from the fact that the message rate is obviously off by 1M)

I also conducted a rate-vs-latency test which left me under the impression that 3k 512 byte messages per second should be fine in terms of latency

There is no activity on the brokers other than my perf test runs, which makes it kind of hard for me to draw any conclusions in terms of acceptable message rate.

Am I missing something obvious or is there anything I could track to make any sense of the above results?

Thanks