[ntp:questions] Re: How often should server hop happen?

David Schwartz davids at webmaster.com
Thu Feb 3 03:05:05 UTC 2005

"David L. Mills" <mills at udel.edu> wrote in message 
news:cts22p$c18$1 at dewey.udel.edu...

> However, if there are multiplie sources all with similar jitter and 
> dispersion, it makes little difference which one is selected, even if a 
> clockhop does occur.

    I don't follow the logic here. Even if the jitter and dispersion is the 
same, won't a clock hop require the server to slew its clock to the new 
clock rather than the old one, resulting in a reduction in stability after 
each hop?

    Consider, for simplicity, two equally good clocks. One a millisecond 
ahead of the other. Both flow perfectly smoothly. So long as the computer 
stays on any one clock, it will have perfect stability, or nearly so. But 
each time it changes clocks, for however long it takes it to move that 
millisecond, it will be flowing either too fast or too slow.

 >And, in a very real sense, it is not the system peer
> that determines the offset, but a weighted average of all cluster 
> survivors. Even with a hop the average doesn't change much.

    Well that starts a whole new set of arguments. ;)

    All too often, people seem incredibly concerned with accuracy (how 
little can we make our local clock average's error relative to UTC be) and 
they ignore stability (how smoothly can we keep time flowing).


More information about the questions mailing list