Closed CBenoit closed 2 months ago
We’re spending a lot of time dealing with Bytes and BytesMut in the JMUX sender task.
Bytes
BytesMut
This patch is getting rid of the leftmost drop. We can actually reuse the BytesMut buffer between writes and greatly reduces allocation / de-allocation.
Performance is increased by ~26.3%.
Before this patch:
0.0000-19.0245 sec 25.2 GBytes 11.4 Gbits/sec
After this patch:
0.0000-17.1307 sec 28.8 GBytes 14.4 Gbits/sec
We’re spending a lot of time dealing with
Bytes
andBytesMut
in the JMUX sender task.This patch is getting rid of the leftmost drop. We can actually reuse the
BytesMut
buffer between writes and greatly reduces allocation / de-allocation.Performance is increased by ~26.3%.
Before this patch:
After this patch: