On 29/01/2019 10:21, Bernd Eckenfels wrote:
How will that look like on Windows, will it use IO Completion Ports? I
guess scalability becomes much more of an issue with typical thousands
of classic sockets.
What’s the expected performance of this? The blocking IO had a lot
less latency compared to Channels, is there some regression expected
or is a goal to also improve latency compared with the classic
implementation?
There's nothing specific to Windows in this. Reading and writing ends up
doing WSARecv/WSASend on sockets configured blocking, non-blocking with
using timeouts. So no I/O completion ports in the picture, also no
channels either. There is a lot of old code using Socket/ServerSocket so
we don't want to regress performance.
-Alan
