[ntp:questions] Meinberg NTP Software--Time Accuracy
martin.burnicki at meinberg.de
Mon Jun 22 10:30:31 UTC 2009
David Woolley wrote:
> W. eWatson wrote:
>> In some science working I'm doing, I need subsecond accuracy for
>> timestamps. I was told the s/w in the Subject will do the trick. Maybe
> It's not the Meinberg NTP Software it is the University of Delaware NTP
> software, more commonly known here as the reference implementation.
> Meinberg simply wrote a Windows installer for it and compiled it. (It
> used to be Dave Mills' NTP software, but the copyright was recently
Right. The installer just has been published to simplify installation under
Windows, but provides binaries built from the official NTP sources.
Hm, maybe we should make this more obvious on our download page.
The NTP service is installed with a reasonable default configuration.
However, the configuration settings can be modified during the
installation, or afterwards by editing the ntp.conf file.
>> someone has used it before. I have little to go on, but installed it
>> successfully yesterday--I think. I gave it the tick.usno.navy.mil NTP
> Although it will work after a fashion, that server will be seriously
> overloaded. In a university, your first choice should be your campus
> servers, then the univerity's ISP's. You should not be using a stratum
> one server if you are leaf node.
>> server name, and that finished the install. After
>> a minute of looking at my clock, there was no change. It happened to be
>> maybe 30 sec off according to my atomic clock. What mechanism do I need
> Make sure that you have the right timezone; ntpd will abort if the time
> is more than 1000 seconds out. (NTP itself users UTC, which will be
> converted to local time by your OS.)
By default the NTP service is installed with the -g option, so if the
initial time offset exceeds 1000 seconds this should be no problem.
However, some people try to "test" whether NTP works by first starting the
service, then changing the system time to some other time or even date, and
then expecting the NTP service to step the clock back immediately.
Of course this kind of testing is not reasonable and lets ntpd stop itself
if the 1000 second offset is exceeded.
>> to use to truly get it started? I don't think what I did showed it was
>> adjusting my clock at all.
If the NTP service encounters any problems then adequate messages are
written to the Windows application event log. The Windows event viewer
program can be used to see if there are any entries from the NTP program.
There should be at least some startup messages.
> Generally at this stage you need to run ntpq and use its peers
> sub-command, then its assoc sub-command and finally run rv for each
> association id (in your case just one) in the assoc output. Post the
> results here.
>> I gave it one NTP server, as above, but probably need more. It allows up
>> to 9. I'm on the west coast, California.
> That would be an installer restriction. If you manually configure it,
> you can have more, although there is a limit to the number of good ones
> that it will actually use. Four is a recommended number. Less
> compromise the fault tolerance.
So if 9 are possible, and 4 recommended, that should be OK. However, if "No
associaton ID's returned" is printed then this looks like a general
problem, and it should not matter if 4 server or only 1 server has been
configured (unless that one server is offline).
>> There doesn't seem like much beginner support at their web site. None?
>> See download page at <http://www.meinberg.de/english/sw/ntp.htm>
> The primary documentation is at http://www.ntp.org/, although that may
> reference the development version. The installer really ought to have
> installed a directory full of html documentation, as the policy is that
> you use the documentation that accompanies the executable.
As David J. Taylor has already mentioned the HTML docs corresponding to the
NTP version are included. There's even an entry for the docs in the Windows
"Start ..." menu.
More information about the questions