Re: [Linuxptp-devel] Optimal P, I constants

[Miroslav L. <mli...@re...>]

On Tue, May 28, 2013 at 07:16:15PM +0200, Richard Cochran wrote:
> > First results are in:
> > 
> >   http://linuxptp.sourceforge.net/igb-master_bcc-slave_pps/

Very interesting, thank you for doing this.

Is there any specification on the accuracy of the generated PPS signal 
and the accuracy of the PPS capture? Could be the PPS rate increased?

I'd expect the jitter in the PPS offsets to be smaller than in the
offsets measured by ptp4l, which doesn't seem happen here. An
explanation for that could be that most of the error is the clock
instability, i.e. the Allan intercept is extremely short. This could
be on the master or the slave or both.

Could you please try running the test again with a very short sync
interval (-7?), free_running enabled and summary_interval set to the
sync interval, so we can see the measured offsets and try to estimate
the jitter and wander?

> 	# method B at 2^-4
> 	# pi_proportional_const  3.2
> 	# pi_integral_const      0.05

Hm, doesn't the method B give at 2^-4 kp=0.2 and ki=0.8?

Thanks,

-- 
Miroslav Lichvar