[ntp:questions] Leap second events
David J Taylor
david-taylor at blueyonder.co.not-this-bit.nor-this-part.uk.invalid
Sun Jan 1 08:56:38 UTC 2006
Similar to other reports, a had a mixed bag of results here due to some
external servers correctly inserting the leap second and others not. I
have only external servers an no reference clock.
What I noticed was that the drift values stored in the ntp.drift files
became very high (values near to 500), which were grossly incorrect.
Under that situation, whilst ntp on two PCs recovered slowly towards
normal (Bacchus and Stamsund), by 07:40 UTC the two other PCs had not, and
were still showing the very high drift values. Consequently I restarted
ntp on those PCs at 07:40, deleting the ntp.drift files before the
restart.
You can see the results here, although only offsets within +/- 100ms are
recorded, so it's just the general disturbance which can be seen. As I
write, Hermes and Odin have just written more reasonable (although not
correct) values into their drift files (been operating for about one
hour), and appear to be gradually recovering, restarting when the offset
exceeds 128ms but with a smaller offset drift rate each restart.
http://www.david-taylor.myby.co.uk/mrtg/daily_ntp.html
It seems to me that, where a single server is in error ntp copes nicely by
ignoring it, but where many servers are in error, some sort of instability
may result.
Happy New Year
David
More information about the questions
mailing list