Subject: Re: hardware clock 'corrupted' by ddb (and possibly other
To: None <jonathan@dsg.stanford.edu>
From: None <cgd@broadcom.com>
List: tech-kern
Date: 07/10/2002 18:53:17
At Wed, 10 Jul 2002 22:26:55 +0000 (UTC), "Jonathan Stone" wrote:
> If you really care about clock accuracy, you are running NTPD.

Everybody really cares about accuracy.  They just require different
amounts.  Sometimes they care if they have to pay for too much.  8-)

If you want sub-second, you want ntp.

If you want sub-minute, you may be OK with the kernel keeping time w/o
NTP, and (perhaps) periodically syncing up with 'correct time.'

(If you want sub-hour something larger, you may be OK over the
lifetime of your box to just set the clock once and never set it
again... but that renderes the di


IMO, probably "The Right Thing" is to write the RTC "periodically" if
running NTP and when shutting down the system (e.g. when your NTP
daemon is killed after having been functioning correctly), and to
write it whenver the system clock is explicitly set.

NTPheads lose slightly on system crashes, because they have RTC drift
since last periodic sync... but they're just gonna adjust themselves
anyway and if they trust the RTC that little... why trust it to DTRT
for the time it takes to reboot?  8-)



cgd