Subject: Re: Adding nanotime() API to kernel
To: None <tech-kern@netbsd.org>
From: Ignatios Souvatzis <ignatios@cs.uni-bonn.de>
List: tech-kern
Date: 02/10/1999 11:23:50
On Wed, Feb 10, 1999 at 12:36:58AM -0800, Erik E. Fair wrote:
> To be slightly more precise, sometimes you want the absolutely correct time
> of day, and somtimes you want an accurately measured interval. This
> distinction is key to good OS timekeeping, because you need to whack around
> the ToD to correct for drift in your local system clock, but interval
> timers must be inviolate (it would be evil to set a timer that never goes
> off because the clock it ticks from got stepped beyond the firing time).
I think thats not equivalent to what Chris proposed. I see three time types
here:
- precise, regulated time
---> useful for "date" display.
- monotime, a time guaranteed to never be <= the last reading, but
equal to the above when not called too often.
---> useful for unique timestamps and similar stuff.
- a continuously (uncorrected, or only corrected at a limited pace)
running time
---> useful for interval timers.
-is