Github user franz1981 commented on the issue:
https://github.com/apache/activemq-artemis/pull/2494
Most of the benefits on memory footprint are already explained on
https://github.com/qihongxu/activemq-artemis/pull/1, while about the contention
I have already built several contention graphs that shows that this
implementation scale linearly with the number of producers on a topic, without
any contention.
In addition, several JMH microbenchs (not attached here) shows that is
always fastest (if single-threaded too) then the original one for any
operations and in any condition.
The only missing bits is an end 2 end test after applying #2484
---
