[ntp:questions] Re: ntp only support sys_precision to microsecond?

Dongling Duan duan at cisco.com
Wed Dec 24 20:02:19 UTC 2003


When you say you get accuracy of 10ns, 2ms or 20 ms etc, how do you get that
number? Is there any tool I can run to get the number? or which variable or
field indicates ntp accuracy?

Thanks
-dongling

"Richard B. Gilbert" wrote:

> If you don't specify your configuration, I don't believe that anyone can
> offer even a good guess.
>
> The accuracy you can achieve depends on what you use for a time source,
> the round trip network delay if you are using network servers to get
> your time, network jitter, the stability of your local clock, the
> version of ntpd that you are using, the speed of the computer it runs
> on, etc.  These are just a few of the factors that could affect the
> accuracy you could hope to achieve.
>
> I've read, I think, of accuracies of 10nsec or better using a local
> "atomic clock" as a time source.  Using a somewhat worse than average
> network connection,  and ntpd 3.5-91on an Alphserver 4000 5/300 running
> VMS V7.2-1, I've achieved accuracies of 2msec when the network was
> idle.  Accuracy with a busy network degrades to 20msec.  Some of the
> clients served by this machine track it within 100 or 200 nsec; others
> don't do nearly as well.  The machines that do best are Compaq D530
> workstations with Pentium Processors of 1.6GHz or better; these machine
> are seldom required to do anything else!
> The machines that do the worst are running a brain-dead implemenation of
> ntpd on 233 or 266MHz processors.
>
> Other things being equal, faster computers do better, lower network
> delays do better, lower network jitter  does better, newer versions of
> ntpd do better, etc.
>
> Dongling Duan wrote:
>
> >Piotr Trojanek wrote:
> >
> >
> >
> >>On 2003-12-19, Piotr Trojanek <ptrojane at mion.elka.pw.edu.pl> wrote:
> >>
> >>
> >>>>   Does that mean that ntp protocol only has miscroseconds accuracy?
> >>>>
> >>>>
> >>No, timestamps in NTP have much better resolution. Note, that in many
> >>OS you can't do better than microsecond accuracy at syscals.
> >>
> >>
> >>
> >
> >  Ok. That's api is mainly for os which has lower resolution (microsecond).
> >
> >   How do we know what accuracy we could get if we run ntp on a box?
> >
> >Thanks
> >-dongling
> >
> >
> >
> >>--
> >>Piotr Trojanek
> >>
> >>
> >
> >
> >




More information about the questions mailing list