[LinuxPPS] cross posting [time-nuts] NTP API on Linux 2.6.26

[Hal V. Engel]

On Sunday 11 January 2009 01:17:31 Hal V. Engel wrote:
> Looking back over my logs I see this having values between 0.107usec and
> 1.367usec.

One other thing in the code that creates these numbers.  Any loop iteration 
that is faster than MINSTEP is rejected and MINSTEP is set to 100e-9 which is 
100 nanseconds.  So if you have a clock that had more resolution than this and 
your CPU was fast enough to iterate through the loop faster than every 100 
nanoseconds the routine that does this would reject these faster times and it 
would give a false reading of the precision of the machine.  Notice that the 
first condition is true for my machine (IE. my clock has a 40 nanosecond 
resolution) and it appears that my machine comes close to the second condition 
since some of the time this returns values close to 100ns.  So current state 
of the art machines with processors that are faster than my machine will for 
sure get incorrect results from this.

For those interested this code is located in ntpd/ntp_proto.c int 
default_get_precision(void). 

Hal