2500x slowdown after switching to `lnm` #42

New issue

Closed

opened 2023-12-08 09:49:00 +01:00 by Jef Roosens · 1 comment

Jef Roosens commented

2023-12-08 09:49:00 +01:00

Owner

After switching to lnm, I've noticed an enermous decrease in requests/second when trying to benchmark the server application. Upon further inspection, this slowdown stems from the call to poll. In the old version, this call's runtime would scale along with the load of the server, with it being as low as 10 microseconds when on full load for a single connection.

With the new implementation, this poll call seems to always be at least 10ms (usually around 40ms) after a call has been performed. I have absolutely zero clue why this is happening, as I can't find anything different about how I call poll.

After switching to `lnm`, I've noticed an enermous decrease in requests/second when trying to benchmark the server application. Upon further inspection, this slowdown stems from the call to `poll`. In the old version, this call's runtime would scale along with the load of the server, with it being as low as 10 microseconds when on full load for a single connection. With the new implementation, this `poll` call seems to *always* be at least 10ms (usually around 40ms) after a call has been performed. I have absolutely zero clue why this is happening, as I can't find anything different about how I call `poll`.