You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A-runtimeArea: std's runtime and "pre-main" init for handling backtraces, unwinds, stack overflowsI-slowIssue: Problems and improvements with respect to performance of generated code.
I did a quick test on a message passing benchmark and avoiding the event loop was a 25% speedup. To do this right we need to keep our own I/O ref count in order to know when it's ok not to hit epoll.
The text was updated successfully, but these errors were encountered:
Is the goal to never hit the event loop while there is no IO, or do we want to have some strategy for avoiding the event loop if there is only a little IO? The former should be really easy, the latter could be complicated.
Something that might not be obvious from the short description here is that doing this means we miss all async callbacks, but that is entirely ok. Not receiving them is fine as their only function is to make sure the scheduler is active doing work, which is the case if the scheduler doesn't receive them due to being active doing work.
A-runtimeArea: std's runtime and "pre-main" init for handling backtraces, unwinds, stack overflowsI-slowIssue: Problems and improvements with respect to performance of generated code.
I did a quick test on a message passing benchmark and avoiding the event loop was a 25% speedup. To do this right we need to keep our own I/O ref count in order to know when it's ok not to hit epoll.
The text was updated successfully, but these errors were encountered: