On Thu, 26 Mar 2026 17:42:05 GMT, Alan Bateman <[email protected]> wrote:

> This looks like a 1% improvement in ops/sec

it is expected as the most of the cost is the RTT within the kernel (loopback 
ping/pong), but the CPU saving is relevant.
I am checking how to make it cpu-bound so it would translate more easily into 
tps.
That said, the Netty custom scheduler repos show some results: i just need to 
work on it. 
But really, is unexpected to have TPS advantages for RPC like behaviour as the 
latency is dominated by the RTT on loopback. The main advantage is cpu saving 
wise, which i think is still a relevant resource nowadays.

-------------

PR Comment: https://git.openjdk.org/loom/pull/223#issuecomment-4137022621

Reply via email to