[LinuxPPS] Problems, but dont know who to blame...

[clemens at dwf.com]

I am testing 2.6.25.1 + tri + IRQ along with the most recent ntp
candidate 4.2.5p113, and an ONCORE PPS source.
I am seeing some problems, but of yet I dont know who to blame.

My experience is that ntpd will start with an 'offset' from the correct time
of some fraction of a second.  NTPD will run for a while and you will see
the 'reach' increase to 377.  Soon thereafter there will be a 'step' decrease
in the offset, the reach will go to zero, and process will repeat.  After 
several
cycles the offset will be on the order of '1us' which is as good as you can
expect with the us clock.

With 2.6.23.1 + bis, this all seemed to work, the 'pull in' took
 longer than I would have expected, but it eventually worked, pulling the
offset to zero.

With 2.6.25.1 + tri + IRQ I do NOT see any step corrections to the
offset, or they occur and then the offset floats upward rather than holding
its own while the reach increases and another step occurs.

At the moment Im not sure what to make of all of this, but the most
significant problem seems to be that the offset is drifting upward each
time you look with ntpdc.  After several hours, the offset is still LARGE,
meaning ms not us. Trying to do some plots just gives garbage
and is of no help.

I guess I should try the tri patch w/o the IRQ patch just to see if its one
or the other,- Ill give that a shot later today.

Puzzled.  Anyone else seeing something like this???

-- 
                                        Reg.Clemens
                                        reg at dwf.com