[LinuxPPS] Interrupt problems?

[Robert Jenkins]

Hi,

I am still seeing some jitter, though with the Atom refclock handling PPS
(rather than using NMEA for both) it's now a fraction of a millisecond
rather than several milliseconds.

This seems to be partly related to machine loading, ntp just loses stability
and all the poll intervals drop. (I run the PPS setup on my fileserver, I
can't justify having yet another machine solely for NTP.)


I'm guessing it's due to either missed clock interrupts or simply delays in
processing the PPS interrupts when another interrupt routine is tying up the
CPUs.

I also suspect some bug in NTP or the kernel, as the present setup keeps
showing 'reach' other than 377 even on peers on the same local subnet, but
the ethernet interfaces involved on all three machines show zero errors.

Does anyone have any general machine setup tweaks that could reduce or
eliminate the remaining jitter?
The present system (on an Athlon64 dual core CPU) is running at 1000Hz with
irqbalance loaded.


Thanks,
Robert.