[LinuxPPS] bouncing primary signals in ntpq -p output, signal loss in pps0 when pps1 is attached.

[Felix Joussein]

Hello List,

When I attach 2 pps sources to my machine, and run ppstest /dev/pps0 and 
simultaneously on /dev/pps1, the following happens: (I used gnu screen 
to monitor the output simultaneously)

First, I plug the pulse generator to pps0 - everything is perfect, the 
pulse is retrieved regularly, no loss.
After a minute or so, I have the insurance, pps0 works fine, so I attach 
a second pulse generator to pps1: suddenly pps0 looses some pulses, 
and/or come very irregular.
Both pulse signals come from an HP atom clock.
pps1 also retrieves some pules, but far less then pps0 and both sources 
are irregular and have lot's of loss. As soon as I disconnect the pps1 
pulse, pps0 returns to normal.

When I increase the pulse length, to half second up/down, ppstest works 
great.

The result in ntpd is as follows:

If I enable 2 pps sources in ntpd.conf with flag3 1.:
pps pps1: another kernel consumer is already bound

Does that mean, that I can only use kernel consumer for one attached pps 
source?

Apart from that, the output from ntpq -p looks suspicious: all peers, 
whether they are other internet peers or pps attached sources, have the 
same stereotype jitter: 7.812, and it will not change anytime.

Also: primary source is constantly bouncing due the signal loss when a 
second signal is enabled.

When I remove the flag3 1 from the second source, and - to be sure, 
reboot the entire machine - the ntpq -p output differs:
Jitter is different for every peer, and lower 0.500 for the internet 
peers, the 2 attached pps sources are at +/- 0.001

Regarding the peers: When I add more then 3 peers (6 in my case), the 
bouncing is again here.

Software versions: OS: ubuntu 10.04, ntpd: manually compiled ntp4-2.6p4.
Kernel: initially 2.6.35.7, no_hz=y config_hz=100mhz
All the described issues where very intense.

After reading the mailing list I realized, no_hz must be n, and I set 
config_hz=250mhz (due description in the kernel help) . (I build 6 
different kernels: all config_hz options with no_hz=no and no_hz=y.
The best result was with no_hz=no and config_hz=250.)

Right now I have 2.6.37 no_hz=n and config_hz=250.
Here I experience slightly different behavior:
simultaneously ppstest on pps0 and pps1 work great, but the ntp bouncing 
is still there when I have more then 3 peers in ntp.conf.

I'd appreciate any help to resolve those issues/questions!

Thank you very much,

Felix