Messages in this thread |  | | Date | Tue, 14 May 96 04:04 EDT | From | (Steven S. Dick) | Subject | Re: Why touch the CMOS clock? |
| |
In article <19960512144353.16753.qmail@Mail.UTexas.EDU>, lilo <alfred!ucf-cs!mail.utexas.edu!TaRDiS> wrote: >On Sat, 11 May 1996, Herbert Rosmanith wrote: >> If you dont have the chance to update your clock via ntp, then the >> RTC is *much more* accurate than the internal, interrupt driven timer-clock. [...] >You've tried using adjtime to correct the problem?
Actually, I spent about a month or two calibrating my system's interrupt driven clock. I wrote a command line utility that calls adjtimex directly to set the system tick length.
My system's clock is accurate within one or two seconds a MONTH. As long as it is on and running Linux, the clock doesn't drift significantly!
Prior to calibration, it drifted by as much as 30 seconds a day. The RTC doesn't drift quite that much when the system is off, but it does drift enough that I have to check the clock if I have the system off for more than a few hours.
I don't adjust my clock with an external source, but the calibration done with adjtimex seems to trigger the kernel updating the RTC. I'm perfectly happy with this. I'd be unhappy if it was changed.
If you don't like the RTC fiddled with unexpectedly, either don't synchronize your system clock, or add a flag to the adjtimex stuff to not update the clock... Don't just remove the code because you personally have no use for it.
Steve
|  |