Major Clock Drift

[alex at pennace.org (Alex Pennace)]

On Mon, Aug 30, 2004 at 09:24:05AM -0400, Josh Pollak wrote:
> On Aug 30, 2004, at 8:56 AM, dsr at tao.merseine.nu wrote:
> >The RTC in your machine is probably rated at 100 parts per
> >million maximum failure. Multiply by 86,400 seconds per day, 100 ticks
> >per second, and you get an assumed clock drift of 8.64 seconds
> >per day.
> >
> >Anything better than that is simply your RTC doing better than
> >specified.
> 
> Thats over a full minute of drift in one week. I find that hard to 
> believe. 

I find that quite easy to believe.

alex at buick:~$ cat /etc/adjtime 
74.119347 1093838485 0.000000
1093838485
UTC

This shows that the RTC on buick is drifting at a rate of 74 seconds
per day.

> Perhaps the RTC is inaccurate as a trade off for providing so 
> many ticks per second, but I've never seen a computer's clock drift 
> this quickly, even when we weren't running NTP.

On most modern operating systems, the RTC (Dallas Semiconductor DS12887
or equivalent) is not the normal run-time clock. Instead, the
operating system arranges to have the interval timer (Intel 8253A or
equivalent) trigger IRQ 0 regularly. This regular interrupt is what
advances system time; intra-interrupt measurments are facilitated by
the RDTSC register on newer Pentium-class processors.