[ntp:hackers] A stop-gap authenticated time service
mlichvar at redhat.com
Wed Nov 11 06:55:10 UTC 2015
On Tue, Nov 10, 2015 at 06:27:02PM +0000, Poul-Henning Kamp wrote:
> I have a lot of respect for Dave Mills, but one of the the things
> he never updated was the "rules of thumb" derived from the PDP/11,
> and that's why NTPD is far too eager to yank the clock around.
Interesting. I'd say it's exactly the opposite now. ntpd is too slow
in adjusting the clock in modern (wired) networks.
> Modern motherboard crystals seldom move more than 1 ppm/K, and 1 ppm
> for a day is only 86 milliseconds.
The temperature sensitivity may be smaller now, but the range in which
the temperature is moving is likely larger due to modern power-saving
features, so the combined effect is probably not very different from
it was before.
What has changed significantly since the old days is the network
jitter. It went down by orders of magnitude, so the Allan intercept is
much shorter than before and ntpd is too slow now.
Of course, if you only need the clock to be accurate to a second or
two, or you are dealing with sources like the HTTPS server, this
doesn't really matter. The loop can simply assume the clock is stable.
More information about the hackers