[ntp:questions] Computing RTC drift

Kalle Pokki kalle.pokki at iki.fi
Thu Jun 4 19:54:48 UTC 2009


We have a system in which we normally synchronize computers to GPS
time through a couple of receivers. There may also be long time
periods when the GPS is not available and the system gets isolated.
There is no internet connection to get the time from. When the system
is up and running, it is synchronized to an orphan server which we
believe will do just fine, as long as the frequency offset NTP has
calculated is close enough to the real errors. It seems the
oscillators we have are pretty stable through a wide temperature
range, so I'm not too worried about that for now.

However, there can also be long down times during which the time is
only kept by the real-time clocks backed up by batteries. The
frequency error of the RTC is way too much - measured worst case was
about 40ppm. If we rule out getting a better RTC, what are the
options? Is there a way within NTP to measure the drift of the local
RTC? Maybe write a reference clock driver for it, if NTP is that way
able to extract useful information from it? Then we could store the
time at shutdown and make a correction to it at boot according to the
latest measured drift value.

Has anyone tried something similar before?

More information about the questions mailing list