[ntp:questions] NTP and GPS

unruh unruh at invalid.ca
Wed May 1 06:00:55 UTC 2013


On 2013-04-30, Miguel Gon?alves <miguel.barbosa.goncalves at gmail.com> wrote:
> Hello fellow NTP users!
>
> I have a few topics I would like to discuss with you guys. I describe my
> configuration first and then I wrote some questions. This is a lengthy
> message but I believe that some topics are interesting for other users.
>
> In my network I have 2 dedicated low powered i386 boxes with GPS receivers
> from different brands (Garmin 18 LVC and Sure Electronics GPS Evaluation
> Board).
>
> At the moment both servers only have their GPS clock as sole reference.
>
> I have a Cisco router that gets the time from these two local stratum 1
> servers and from more 5 stratum 2 servers (close to me) on the Internet. I
> know routers aren't good timekeepers but this is intended as a backup as
> you'll see further down.
>
> Every device in my network runs NTP and requests the time from the 2
> stratum 1 servers and from the Cisco router.
>
> Now the questions...
>
> If one of the stratum 1 goes nuts the Cisco will mark it as a falseticker
> and discard it's data, right? I believe the Cisco will then choose the
> other stratum 1 to sync from. Right?

Nope. It has no idea which the correct clock is. If one disappears
entirely, then it will use the other, but if one wanders off into the
wild blue yonder, then ntp has no idea which one it is. That is why you
need three servers, so two can outvote the bad one. One of them could
even be a network server (pool server) which will not be as accurate,
but will at least tell which of the two has gone crazy. 


>
> If the Internet connection is overloaded (not usual here) the Cisco will
> choose predominately the stratum 1 servers. Right?

It should. The are closer as well ( much lower delays).

>
> Is this circular topology (a client using a server that in turn references
> the other two sources the client is using) a good idea?

Well, as I said, at present it cannot. And you really have no backup.

>
> I know a better approach would be to add another stratum 1 (having 3 in
> total) and ignore completely the Internet servers. Can I really get away
> with the above configuration? I am not looking for the ultimate precision
> but I intend to use these two receivers I have to their fullest.
>
> How good is NTP detecting a problem with a reference clock? Let's imagine
> there's a problem with one particular satellite and makes the time jump a
> few mili-seconds. With only one reference clock will NTP follow blindly or

?? That is no an error likely to happen. The sattelites are very closely
monitored.

> ignore it? Adding another GPS would be pointless but a radio receiver would
> be OK I guess. The problem is that I don't have any radio signals in my
> country (I could get a GLONASS receiver). GPS receivers designed for timing

Glonass would be fine. 

> usually employ the TRAIM algorithm and can discard bad data. Does NTP do
> something like this? I mean, if my NTP is synched and running fine I would
> expect NTP to discard obviously bad data. OTOH, temperature can have a huge
> impact on this and NTP doesn't know how hot the server is running.

 I think you are worried about the wrong things. If GPS goes down, the
world and you will have far worse problems than worrying about the time
on your machines. 
On the other hand, your receivers going down are much more likely (My
two GPS18 have both died after three years. My two sure's one died
because the antenna was crap and died. I have to find a place to get a
new, cheap antenna.

ntpd will more or less take care (slowly) if the servers get hot. It
cannot do anything about about the server dying of course. 

>
> In the end, I might buy a Meinberg NTP server for my network and have 3
> sources and then I will be able to completely ignore the Internet servers
> and validate my servers among each other.

And where would the Meinberg gets it UTC from?

>
> Thanks for reading and I will appreciate your contributions.
>
> Kind regards,
> Miguel



More information about the questions mailing list