[ntp:hackers] A stop-gap authenticated time service
mlichvar at redhat.com
Wed Nov 11 17:39:59 UTC 2015
On Wed, Nov 11, 2015 at 11:42:54AM -0500, Majdi S. Abbas wrote:
> What do you think a cellular connection looks like today?
> Or bad airport WiFi?
I think the distribution of the delay there is very different from
ordinary networks and only a small part of the measurements is good
for synchronization. The NTP implementation should be able to drop a
large number of consecutive samples. In the ntpd design it seems
that's not possible. The clock filter can drop only a limited number
of samples before the PLL becames unstable.
> I'm not so sure I'd write off some of those assumptions
> just yet, at least not without some hard data around it. The
> implementation has to be prepared to handle anything out there,
> not just the common case.
I fully agree. That's why I prefer a more adaptive approach, even
when it can't always work better than something optimized for more
More information about the hackers