[time-nuts] Chooses for a desktop/server NTP external 1PPS reference

[Bruce Griffiths]

Hal Murray wrote:
>    
>> Right, I would assume the oscillator on a typical motherboard will
>> have too much variance to get down into the nanosecond range?  When
>> you say system has special hardare, do you mean like an Oregano PCIe
>> card that the system could use as a reference or something else.
>>      
> The crystals in most PCs are good thermometers.  If you hold the temperature
> stable the frequency will be pretty stable.
>
> Most PCs use spread spectrum clocking to dance around the EMI rules.  That's
> an interesting monkey wrench.  I'll have to work out the numbers sometime.
> (If the data sheet tells you the spread and the modulation frequency, you
> should be able to work out the max deviation from the long term average.)
>
> Sometimes, the BIOS lets you turn that off.
>
> There are 2 types of special hardware I was thinking of.  One is a counter
> that gets latched by the PPS signal.  The other is a more stable clock.
>    
The Soekris NET4501 allows this to be done.
You can also replace its 33.3333Mhz clock with a more stable one.
> Another idea would be an add-on board, probably with a FPGA.  Details TBD,
> but the general idea is to feed it a good clock and a PPS signal.
>
>
> pisymbol at gmail.com said:
>    
>> I'd like to go to the next step.  I'd also want to start graphing a
>> bit some of my ntpd usage (I assume folks basically grok peerstats and
>> look at the offsets over time?).
>>      
> Look at loopstats and clockstats too.  In particular, the drift in loopstats
> is a good indication of system health.
>
> There is another large rathole you can get sucked in.  If you have good
> clocks at both ends, you can measure one-way network times rather than just
> round-trip times.  If you want that info, look at rawstats.
>
>
>    
Bruce