[ntp:questions] Installing more stable oscillator?

jlevine jlevine at boulder.nist.gov
Mon Jul 16 13:45:28 UTC 2007


On Jul 13, 7:43 pm, Pete Stephenson <pete+use... at heypete.com> wrote:
>
> Surely many of the stratum 1 servers (say, time.nist.gov) that get and
> distribute time directly from atomic clocks aren't just off-the-shelf
> servers with cheap, unstabilized system clocks, right? I know that many
> of the public stratum 1 servers deployed by individuals and
> organizations get their time from GPS, and are probably ordinary
> computers, but I have this (again, perhaps incorrect) assumption that
> the servers that supply the time /to/ the GPS system are not ordinary
> computers.

   time.nist.gov (and all of the other NIST time servers) use
standard
servers. The older systems are built using Tru64 Unix and the newer
ones are built using FreeBSD. The systems are synchronized using a
telephone connection to the NIST ACTS time service.
I have modified both the kernel and the serial port driver to reduce
the
jitter and latency, but the systems are basically off-the-shelf in
terms of
hardware.
Some of the systems have external rubidium atomic frequency standards.
These devices improve the short-term stability and the reliability
("Holdover
performance") of the system, but do not improve the accuracy very
much,
since the external clock must be set on time using some external
means.
The servers are synchronized using a separate process that is
independent
of NTP. It is an adaptive frequency-lock loop that adjusts itself for
the statistics
of each individual server The details are a bit complicated and have
been
published in the literature.
Once it has been characterized, a cheap computer clock can probably
provide
timing accuracy of about 50 milliseconds for 1 day with no external
calibration.
Cheap no-name hardware is typically worse, but your mileage may vary.
An
external quartz crystal oscillator might improve this to about 10
milliseconds
per day (characterized but free-running), but it is hard to do better
than that
without more expensive hardware and/or semi-heroic measures.

Judah Levine
Time and Frequency Division
NIST Boulder




More information about the questions mailing list