[ntp:questions] Re: Server sync changed from 3ms to 75ms
Richard B. Gilbert
rgilbert88 at comcast.net
Fri Jan 7 20:17:06 UTC 2005
holmezy at yahoo.com wrote:
>Dave,
>
>Thanks for the response! Yes, we are aware of the demise of TrueTime.
>However, you might like to know we purchased 2 TS2100s in the past
>year. One configured to receive IRIG-B, the other GPS.
>
>As for our configuration, this lab architecture was inherited and out
>of our control. We are working toward making our fielded configuration
>more robust and able to take advantage of NTP's built-in redundancies.
>
>The differences are being measured between our Regatta servers using
>the timedc and ntpq commands. Each Regatta is on its own LAN and
>communicates to the other Regatta's via an emulated WAN using SHUNRA's
>to impose real-world long haul comm constraints (packet loss, delay,
>jitter, etc).
>
>Currently, the lab where these machines are located is in configuration
>lock-down. We'll be able to investigate the jitter between the servers
>sometime next week.
>
>Based on this information, when the software we're testing performs the
>timedc command between the servers, the differences are actually
>between the Regatta system clocks, which are supposed to be in sync
>from their NTP requests to the NTS100i units. It's strange because
>sometimes we're 1-3ms and then it will suddenly grow to +/-75ms. It's
>also inconsistent when you are sitting on each machine and checking the
>clock differences. We're also discussing reducing our polling time if
>we conclude that the machines are experiencing too much drift between
>the polling intervals.
>Once again, thanks for the response,
>
>John Holmes
>
>
>
If you are controlling your polling time, try letting NTP control it
instead. NTP should start at 64 second intervals and increase
gradually to 1024 seconds as the clock synchronizes and stabilizes.
Forcing short intervals would mean that you never had an interval long
enough to measure a small error accurately and could not correct that
small error. Forcing longer intervals means that correcting larger
errors will take much longer. NTP will set a poll interval appropriate
to the circumstances if you allow it to do so.
More information about the questions
mailing list