Github user NicoK commented on the issue:
https://github.com/apache/flink/pull/4594
Yes @zhijiangW, I kept the original logic.
I guess, the reasoning behind using off-heap netty buffers only was to
reduce the overhead before transmitting messages over the wire: 1) we reduce GC
overhead somewhat and 2) at some point we need the memory to be off-heap and
put into kernel space anyway - depending on netty, this may be optimised if it
is already off-heap.
Also, starting with #4481 we will only be using off-heap network buffers
anyway.
---
