Re: [Linuxptp-users] clockcheck - need to filter large spurious phase jumps?

[Richard C. <ric...@gm...>]

On Wed, Nov 06, 2013 at 05:06:57PM -0500, Rich Schmidt wrote:
> I redefined CURRENT_UTC_OFFSET to zero for testing ptp4l without phc2sys:
> 
> nice --20 $PWD/ptp4l -S -i eth1 -l 7 -s -p /dev/ptp1
> 
> The ptp4l master offset of the CLOCK_REALTIME shows a high jitter,
> 26.5 microsec peak-peak, 5.5 microsecs s.d.  Smoothing would greatly
> improve this;  I would like to write an NTP driver based on ptp4l and let
> it filter the results rather than having it steer CLOCK_REALTIME.

This jitter sounds about right for SW time stamping. You can make it
worse by adding a high system load. Also, the nice command probably
will not improve things as the time stamps are made in kernel
context (but it doesn't hurt either).

Before implementing a filter, you can simply use smaller PI weights as
a first attempt. In a project a few years back, we came to the
conclusion that the filter in the original ptpd code was not really
helping too much.

In any case, I only suggested to try SW time stamping in order to
characterize the source of the time jumping problems you report. If
you see the same effect with SW time stamping, then this indicates a
problem in ptp4l or in your grand master. If you see the effect only
with HW time stamping, then it is likely to be a driver issue.

Thanks,
Richard