Re: [Linuxptp-devel] [PATCH RFC 0/6] Improve accuracy with software timestamping

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

On Wed, Mar 11, 2015 at 03:32:16PM +0100, Richard Cochran wrote:
>    https://mlichvar.fedorapeople.org/tmp/ptp/ptp4l_error2.png
> 
> I would think simply comparing the sampled offset with the average
> offset and scaling (or pruning) it would be at least as effective as
> looking at the delay.  Plus it would not depend on "balanced" rates.

I think that would be a filter working with already calculated
offset/delay samples, not time stamps directly. Spike filters work well
if the packets are delayed occasionally, but without looking at delay
it wouldn't be able to tell if the network is still congested or the
clock is really off that much.

> > - port.c can use this too, deduplicating the delay calculation and
> >   delay filtering
> 
> (The delay must be per port, for P2P mode.)

Each port would have its own time stamp processor.

> > Does it make sense?
> 
> Yes, that is the right direction.

Ok, I'll see if I can update the patches for this.

Thanks,

-- 
Miroslav Lichvar