[time-nuts] Is it sensible to update every few seconds from NTP server?
Magnus Danielson
magnus at rubidium.dyndns.org
Sun Nov 11 13:37:22 UTC 2012
On 08/11/12 09:14, Poul-Henning Kamp wrote:
> --------
> In message<CANX10hBat1ie0Rx_bS+X1dqGGLTPx5xCz2=tYPP_uvwGu-mosg at mail.gmail.com>
> , David Kirkby writes:
>
> If the local NTP implementation on the machine has a marginally competent
> PLL, updating more than once per minute will just increase the noise.
>
> If the local NTP implementation is really SNTP which steps the clock,
> then you should find a better one, if what you need is good timekeeping.
>
In addition, if many people ask the same server at a relatively high
rate, it's approximating DDOS.
I'd expect that kernel FLL/PLL be included and used. Shifting from the
default SNTP to NTP with set-only is of only marginal improvement at
best, but with the dangerous notion that you "fixed it".
If you read up on NTP, there are articles (you find them through
http://www.ntp.org) describing the properties of some (now old) OS
clocks and their behaviour. Similar is to be expected for default
Windows behaviour as well as many virtual machines. I think the old
articles is good to raise the awareness, but I can't recall similar for
the modern situation, but I am sure someone here have seen them.
Make sure the OS (kernel-support + NTP configuration) you are using
provides you with the stability you need.
Does someone holds a Best Current Practice for various OSes?
Cheers,
Magnus
More information about the time-nuts
mailing list