[ntp:questions] Re: NTP algorithms to predict clock skew and/or jitter?

skillzero at gmail.com skillzero at gmail.com
Sat Jul 30 03:18:17 UTC 2005


I ended up just doing a simple predictor where I calculate a moving
average of the NTP clock offset at each NTP time sync and convert that
to the number of processor cycle ticks to add at each 60 Hz interrupt
to compensate for the skew over the period. This gets time in sync
between the server and clients to within about 10 microseconds, which
is a little better than my goal of 22 microseconds (1 44.1 KHz audio
sample) and is computationally easy (just integer adds, subtracts, and
shifts).




More information about the questions mailing list