Process latency regression on mqperf AWS run

The run on AWS (using large machines) to replicate the results of https://github.com/enkeys/mqperf/commit/5408da2f3f3210984e6fa43d4893440e25280878 to compare AMQ 7.8.1 to Artemis Upstream master (7.9.0 future GA) has highlighted a huge improvement on throughput with a regression on processing latency yet to be investigated.

Potential reasons could be:

• accumulated processing backlog
• too high number of consumers on the same queue (note: AMQ broker process consumers of the same queue in single-threaded fashion)
• a bug introduced by https://issues.apache.org/jira/browse/ARTEMIS-3045

The test has been performed using 500_000 messages per node, 25 threads per node, 4 producer nodes, 8 consumer nodes, 100 bytes messages.