[ntp:questions] Onboard Local Oscillator Change Improvements

Jason Rabel jason at extremeoverclocking.com
Wed Oct 10 18:47:01 UTC 2007

> - is the precision of these "messages" (or timestemps) affected by the
> stability of the local oscillator ?

For all practical purposes, no. As Dr. Mills said, there are many other
factors causing various bits latency in a PC that are independent of the
oscillator. What little you might gain from a Frankenstein oscillator
replacement is still outweighed by the random OS and hardware delays.

> - which kind of accuracy can we reach enhancing the stability of
> the oscillator ? (nanoseconds?)

Let's not forget about the PPS source. You mention "GPS", but no details
there. I would focus more on that end first, ensuring you are doing all that
is necessary and calculating out all your ns delays in cabling, splitters,
etc, so that it can be added to NTP. Accurate antenna position is important
too. Keep in mind GPS is not UTC (even omitting the leap seconds). GPS time
does wander in the double-digit ns range.

As long as you have your PPS signal (and setting the minpoll to 4 doesn't
hurt either), changing the oscillator in your PC isn't going to make much
difference. Same reason as before, there are lots of little things going on
inside a PC that cause random bits of uncompensated delay. I wouldn't say
you could get single digit ns, but more in the order of hundreds of ns
depending on the hardware & OS. Even after you pass this NTP info on via the
network then you are jumping up to the ms range for those clients. Depending
on your OS & BIOS there are certain tweaks you can do to help minimize
delays, check the NTP wiki for more info.

You said you were using Linux, what method are you using to read the PPS
signal? Also does Linux even have nanosecond precision?

> The same question at client side is:
> - which accuracy of local clock can we expect using high stability
> oscillators? (we have measured around 60 microseconds  peak-to-peak,
> using standard crystal in *environment test* configuration- see
> below )

Again, oscillator replacement is not going be the big factor here unless you
have random and sharp environmental changes. NTP over Ethernet is going to
jump you to the tens of ms range. The random unknowns of the source +
network delays all add up so the clients can only assume the time falls
somewhere within a certain range. If you need better than that then you
would need to feed a PPS signal directly to the PC. Maybe even try different
networking hardware (both NICs and switches) and see if that affects your
results any.

A better oscillator in the PC would really only mean better holdover when
all NTP sources are unavailable. But even at that, depending on the OS and
load, time could wander off just as quick even with the oscillator upgrade.
Power regulation could be another factor, not just the AC power coming in,
but the quality of the PC's PSU, or heck even the voltage regulators on the
motherboard. You can pick apart literally a thousand different variables for
a PC that are going to affect time keeping. But the complexities of the
hardware itself and the OS are more your problem than the oscillator.

More information about the questions mailing list