[ntp:questions] NTP makes a time jump
david-taylor at blueyonder.co.uk.invalid
Thu Jul 11 14:51:54 UTC 2013
On 11/07/2013 15:32, Igor wrote:
> Hi Brian,
> 15 secs was just a sample, I hope it would be less. The problem is that the system consists of two parts - mission critical (that shall not depend on anything except itself, which does not really care about timestamps) and data logging that does care. They share same hardware devices during operation. It could be the case system is launched without NTP, in case our logging part is not needed. But suddenly, after some time of operation, logging part is added and a NTP server is configured. We need to re-align "floated away" critical part to the true time without interruption on the data transfer in critical part. Since we have 100ms polling rate, it shall not jump more than that.
> What I conclude from my tests and all your advises is that we will not be able to cover 100% of time jump cases by means of NTP network - clients follow, but not fast enough, or make jumps, or get lost if the difference will be too big.
> The feasible solution may be to use a Single NTP server(that behaves like on the graph), and SNTP clients that will blindly make small jumps each second, following the server as close as possible. Or - use completely unsync clock on critical part and compensate the difference on logging part by periodical detection of delta between them.
> Any other ideas will be greatly appreciated!
> Thanks for everyone for your help!
Igor, could you get away with logging two lots of timestamps - from the
client system clock and from a true UTC source?
I do something vaguely similar when geo-tagging photos - once a day I
take a picture of the GPS-locked clock with my camera and can then
estimate how many seconds offset the camera is from true UTC. As the
camera drifts by less than a second a day, that's good enough. My other
camera has a built-in GPS. Your error rate seems much greater than
More information about the questions