
My problem appears to be solved. It's been nearly 24 hours and ntp is latched very well - 2-3ms offset all day! This is a brief summary of things in the hope of helping any one else who has trouble. I am not sure if my solution was exactly ideal : I noticed that my frequency offset was exactly 500 - which is mentioned as an upper limit for the ntp to fix the clock drift. It just said it didn't perform well with such large offsets not that it wouldn't work at all. I read this page: http://lists.debian.org/debian-user/2003/01/msg05870.html and stopped ntp, removed the drift file and tried to calculate the adjustments to correct the clock drift and fix up it up by using the calculator on this page: http://www.ep.ph.bham.ac.uk/general/support/adjtimex.html I stopped ntp from synchronising with clocks by using the noselect flag on the ntp server statements (as mentioned in it's debugging notes on large drift values) and then run adjtimex. I did that several times first getting a large +ve drifts and then next -ve drifts. Each time I zero-ed the clock by running ntpdate and started again. I think actually adjtimex may take a while to cause an effect so I am not sure if I am waiting long enough or undoing the effect of the previous one. After doing that a few times, I restarted ntp normally, having set the date with ntpdate and it set the frequency to -467 (previously it was +500) and then it has slowly settled down to -453 and the offset is around 2-3ms. Really good. It might have worked even simpler if I just followed the instructions, stopped ntpd, removed the drift file, and ran ntpdate every 10mins ntpdate -s -b ntpserver which will set the time instantly (rather than drift it and log it to syslog). Then use the size of the adjustments to calculate parameters for adjtimex using the above link. Then run ntpd Another few notes for other people: You can enable more logging to syslog via logconfig +allsync +allclock +allpeer Still doesn't log that much and as it is not losing sync I don't know if it would be helpful. /usr/share/doc/ntp-doc/html/monopt.html - using the debian documentation package ntp-doc explains the stats when you enable as mentioned in previous e-mail. Also the above link mentions some interesting issues about the clock source - it found the hpet clock source was 10x better than tsc. e.g. - cat /sys/devices/system/clocksource/clocksource0/available_clocksource echo hpet > /sys/devices/system/clocksource/clocksource0/current_clocksource I haven't tried that because I am very happy with things currently. Hopefully this is useful to some one else having troubles with ntp which is really quite complex under the hood. Andrew