[ntp:questions] More on the pinball machine
David L. Mills
mills at udel.edu
Fri Jan 6 23:16:46 UTC 2006
I see there was a bugfix sometime after June 2005 and before 13 December
2005 that directly impacts the pinball behavior. In file
ntp_loopfilter.c on or about line 348 should be
state = S_SPIK;
In the previous version the return(0); line was missing and flow
continued to the next case. After the stepout threshold was reached, a
mistaken frequency correction of one second in 900 seconds, or about
1111 PPM, was made and then clamped to 500 PPM. If this is the case in
your setup, we have found the smoking gun.
I tested with and without initial time steps and time steps after the
daemon was running and had synchronized to sources. I checked the
initial frequency computation with and without a valid frequency file.
While at it I noticed a couple of things that could be and were
improved. One of these was the system jitter, which was unnecessarily
large after the step and resulted in a misadvised increase in poll
interval. The smoke has cleared in the ntp-dev version as of 6 January.
More information about the questions