[ntp:questions] Re: Server sync changed from 3ms to 75ms

David L. Mills mills at udel.edu
Thu Jan 6 20:27:52 UTC 2005


John,

As you probably know, TrueTime is no longer in business, as it was eaten 
by Symmetricom. The NTS100 is no longer manufactured and all of 
TrueTime's network time products other than GPS have been abandoned.

I find your configuration very curious, as apparently you are not using 
the built-in redundancy of the protocol. I assume there is one GPS 
receiver delivering IRIG to multiple NTS100s, each with a dedicated 
xntpd client.  I also assume the differences you mention are between the 
xntpd servers as measured by their clients. The most likely cause is 
large jitter in either the GPS receiver, IRIG signal or NTS100s.

The first thing to do is check the NTS100 client xntpd for jitter. 
Unlike the NTPv4 client (ntpd), the NTPv3 client (xntpd) has no jitter 
statistic, so you will have to eyeball that from the peer variables 
and/or peerstats statistics. You should see jitter much less than 3 ms. 
If all NTS100 clients see substantially the same jitter, go after the 
IRIG and/or GPS. Unplug the IRIG at the radio and see if the jitter 
subsides, as the NTS100s will coast, at least for awhile.

You should understand that what you see with the xntpd client is not the 
actual local clock offset but a synthetic offset relative to the 
disciplined clock time. This was a terrible idea and one of the several 
reasons why ntpd was born. Also, somebody must have modified the code, 
since the original xntpd code that left here several years ago prior to 
ntpd would not have survived the 34-year rollover that occured in 2004. 
I have no idea what else might have been done to it.

Dave

olmezy at yahoo.com wrote:
> Hello everyone,
> 
> Some of our IBM Regatta P690 servers recently started displaying time
> differences of +/-75ms in the past six weeks.  Prior to this the time
> differences were +/-3ms.  These systems are managed under very strict
> CM guidelines!
> 
> Each Regatta has its own TrueTime NTS100i timeserver and each one is
> running xntpd.  The TrueTime servers receive an IRIG-B input signal
> from one upstream GPS unit.  Prior to the start of our software a
> "timedc clockdiff hostname1 hostname2 hostname3" is performed to verify
> the that the Regatta sytem clocks are within a +/-3ms tolerance.
> 
> We've been running the same config in the lab for a few years and noone
> has ever seen this problem before.  The only change to the config in
> recent memory is the addition of 1 more Regatta, with its own
> timeserver to the mix.
> 
> Any thoughts or suggstions on where to look for some better info?
> We've tried various commands such as ntptrace, ntp -q, etc.  We're in
> the process of verifying the IRIG-B input signal to the TrueTime
> machines.  I don't believe this will yield much as all the TrueTime
> machines are connected to the same signal amplifier.  I have a call
> into IBM with a system clock question regarding a similar problem we've
> had in the past.  Apparently, there is some validity to my concern but
> haven't heard back from the Development Engineers.
> Thanks in advance,
> 
> John Holmes
> 



More information about the questions mailing list