C1's latency is still way too high for our latency-sensitive deployment

[Konstantin Belousov]

On Mon, Oct 26, 2020 at 09:29:02PM +0800, YUAN RUI via freebsd-arch wrote:
> C1's latency is still way too high for our latency-sensitive deployment
> I want a generic interface that allows applications to prevent entering deep
> C-state
> 
> 
> Similar:
> 
> linux api:https://access.redhat.com/articles/65410
So what do you want that we do not provide ?
There is cpu idle spin method that you can switch to, using
machdep.idle sysctl.

> 
> windows api:https://docs.microsoft.com/en-us/windows/win32/api/timeapi/nf-timeapi-timebeginperiod
> 
There is no mention that windows HALs tie timer precision with Cx latency.
That said, FreeBSD has a knob kern.timecounter.alloweddeviation, that specifies
how much precision is taken when setting up process timers or when sleeping
by explicit user request.

> 
> In fact, a better way is to provide windows to applications in the form of
> APIs, But Linux can be fine-tuned.
> 
> _______________________________________________
> freebsd-arch at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-arch
> To unsubscribe, send any mail to "freebsd-arch-unsubscribe at freebsd.org"