[ntp:questions] NTP 4.2-dev snapshot - increases PLL freq linearly, decreases with a curve?

Joshua Coombs jcoombs at gwi.net
Thu Mar 31 16:16:59 UTC 2005


Playing around with a couple of the latest NTP dev snapshots on FreeBSD 
4.11, I noticed a bit of odd behavior.  If I arbitrarily set the PLL drift 
setting low, NTP very quickly ramps it up.  On the graphing I've been doing, 
it is a linear increase, looking like a max rate of change of 30ppm per 
hour.  If I set it high, there is a curve showing up for the decrease, it 
looks like it's not quite aggressive enough to get the PLL drift back to 
spec in 48 hours.

The data I have has the following interesting data points.

The box normally hovers at a correction value of -9.5ppm
I set it to 32 and observed it from there, starting at 12:15pm yesterday
The rate of adjustment is linear for the first 3 hours, dropping down to 16 
by 3pm
>From this point forward, the curve shows up
12:15pm - 32
1pm - 28
2pm - 22
3pm - 16
4pm - 12
6pm - 6
8pm - 2
10pm - 0
12am - -2
2am - -2.5
4am - -3
6am - -3.75
8am - -4
10am - -4

The drift is still steadily dropping, but on the order of .001 every 64s.

Is this the anticipated and intended behavior?  It seems odd to allow a high 
rate of adjustment in one direction, but not the other.

Joshua Coombs 





More information about the questions mailing list