[ntp:questions] Re: NTP stepping issue
David L. Mills
mills at udel.edu
Wed Oct 20 13:12:07 UTC 2004
The ntpd engineered impulse response limits overshoot to about seven
percent, with either the kernel or daemon feedback loop. What I see with
Solaris is more like 100 percent for large initial offsets. The
transients don't die out for many hours to a day.
I have removed questions at lists.ntp.isc.org from this reply, as that list
complains that I am not authorized to send to it.
Brian Utterback wrote:
> David L. Mills wrote:
>> You are not going to like this answer. I mentioned awhile back that
>> Solaris, at least, had fiddled with the adjtime() syscall to speed up
>> convergence for large adjustments, effectively adding an additional
>> pole to the carefully crafted impulse response built into ntpd. The
>> result expected is serious overshoot at large offsets, just as you
>> describe. There is no help for it other than to use the kernel
>> modifications, but the existing modifications were designed only for
>> offsets less than 0.5 second.
>> I don't know what system you are using, but if it slews faster than
>> 500 PPM, expect trouble.
> Dave, correct me if I am wrong, but isn't a 50% overslew expected even if
> the slew rate is less than 500PPM as required? IIRC, this is one of the
> to see if the kernel PLL is working correctly.
More information about the questions