🚀 go-pugleaf

In article <1992Nov10.200220.27203@kakwa.ucs.ualberta.ca> sherwood@space.ualberta.ca (Sherwood Botsford) writes:

>Because the amount of drift depends on the usage of the machine.  The  
>clock issues an interupt every nth of a second ( 50 on most unix boxes)  
>however, the priority of the interupt is fairly low, so a lot of the  
>time it has to wait for service.  If it doesn't get serviced before the  
>next interupt, then a tick is lost.  A quiestent machine keeps pretty  
>good time.

This is (usually) incorrect.  You have rather serious problems if you
machine is losing clock interrupts.  The NeXT, in Release 0.8 had a
problem that I found which attempted to periodically resynchronize the
time of day clock from the cheap battery-backed-up clock/calendar
chip; it no longer does this.  Sun platforms also have this problem,
but it can be turned off.  By far the worse problem with timekeeping
these days seems to be on some versions of SunOS on some Sun4
platforms that disable interrupts during kernel writes to
/dev/console.  I believe there is a fix for that.  I am not aware of
*any* problems on the NeXT related to missing clock interrupts and
load.

>I never have figured out why Casio can make a watch that keeps time to  
>a second per month, but no unix box can do so.  Why can't they put a  
>casio set of guts in there, and periodically update the system clock  
>from it...

The advantage the Casio has is that it's in a relatively constant,
temperature controlled environment (your arm, next to your skin).
Also, most computer manufacturers have no motivation to spend a little
extra money to design a robust, temperature-compensated crystal
oscillator that's required for good timekeeping.  It's not
particularly difficult to do, but it costs money.

louie