[ntp:questions] Antwort: Re: Meinberg software running under Vista/Windows 2008
unruh at wormhole.physics.ubc.ca
Wed Jul 6 18:50:46 UTC 2011
On 2011-07-06, Radu.Popa at technomatic.de <Radu.Popa at technomatic.de> wrote:
> Ok,, but windows time server would synchronize according to the parameters
> I set in the registry. Can I do the same with NTP? I mean can I "force" it
> to update the system clock as soon as it sees it's offset? In my config
> file I have set the polling interval at 16 seconds, iburst. The servers in
> question are mission critical machines and have to be synced all the time.
> Unfortunately for this task the windows time service is not accurate.
You first need to understand how ntp works. It's primary mode of
operation is to change the rate of the clock to get rid of any offset
that it detects via its algorithm. That algorithm goes out and gets time
from a variety of servers (set by you). In order to try to get around
the possibility that the time from your machine to the servers is
sometimes assymetric, it onlyuses those measurements which have the
smallest roundtrip time to that server of the last 8 requests. It then
uses the offset it gets from the various servers to try to determine
what the best estimate of the offset of your clock is. It then alters
the rate of your clock to get rid of the offset-- if the offset if
positive (your clock is ahead) it slows down your clock. If negative it
speeds up your clock.
If the offset is greater than something like 125ms it steps the clock
rather than try to get rid of the offset by rate changes. ("The ntpd
algorithms discard sample offsets
exceeding 128 ms, unless the interval during which no sample
less than 128 ms exceeds 900s") If the offset
is greater than about 1000 sec, it gives up and quits (unless you run it
with the -g option when it will correct such a large error only once.)
ntpd will keep your clock on time quite well, but it is NOT designed to
handle your suddenly altering the clock by 120 sec. If your system is
such that something on your computer can alter the clock suddenly by
hundreds of seconds, ntp is NOT for y ou. (I know of nothing that can
help in that case).
If you are simply "testing" then your tests are unrealistic-- that 120
sec shift will not happen in real life. It is like taking your computer
up into an airplane and testing it by dropping it out the windown. It
will fail, but I would not base any decision on that failure.
> Von: unruh <unruh at wormhole.physics.ubc.ca>
> An: questions at lists.ntp.org
> Datum: 07/06/2011 06:31 PM
> Betreff: Re: [ntp:questions] Meinberg software running under
> Vista/Windows 2008
> Gesendet von: questions-bounces+radu.popa=technomatic.de at lists.ntp.org
> On 2011-07-06, Radu.Popa at technomatic.de <Radu.Popa at technomatic.de> wrote:
>> I have an issue with the Meinberg software:
>> I have defined a simple configuration file
>> Server time1
>> Server time2
>> I can see be running the ntpq -p command that the client is using the
>> servers: one as preferred (*) the other one on standby (+) and also the
>> offset which is around 120 seconds. The problem is that Meinberg is not
>> setting up the system clock by itself. However, if I restart the
>> it does it at the start, but if I modify the clock afterwards to
>> an offset, it still does not sync system clock.
> Exactly why would you be modifying the clock? NTP drives the clock to to
> a small offset (millisaconds to microseconds) and keeps it there. That
> is its purpose. It does so by reading a remote server occasionally (
> from 1 min to 20 min) It will then throw away 7/8 of the times its gets,
> (depending on the round trip delay of the measurement).
> so you need to wait for about 2 hours to see what ntp will do with your
> offset (3hr=8x20 min. 20 min is the time between polls at level 10)
> But once the clock is disciplined that 3 hr wait is "fine", since the
> clock is ticking along reasonably disciplined to keep good time during
> that 3 hours. Ie, I have no idea how long you waited to see what happens
> to the 120sec.
> Speedyness is not one of ntpd's strong points.
>> Thank you!
> questions mailing list
> questions at lists.ntp.org
More information about the questions