Github user franz1981 commented on the issue:
https://github.com/apache/activemq-artemis/pull/1918
This one, together with
https://github.com/apache/activemq-artemis/pull/1916 aims to reduce the memoty
footprint (improving scalability by different angles) of AMQP on the broker.
I need a review from @clebertsuconic and @tabish121 for the mechanics that
I've used to reuse the same send buffer: the mechanics is solid but if
`ProtonServerSenderContext.deliverMessage` would be already
uncontended/single-threaded it could be simplified avoiding the
`AtomicReference` usage.
---
