On Wed, Jan 13, 2021 at 12:42:36PM +0200, Vladimir Oltean wrote:
> So I might expect that what you mean by "different requirements" is
> simply that some might be ok with simply a one-time measurement of
> offsetFromMaster that falls below (in absolute value) than the
> user-provided threshold, while others might want to compare a time
> average of the offsetFromMaster with the threshold. Maybe we can do
> both?

There are many more than two variants.  For example you could want a
moving window of mean or median below X, or standard deviation below
Y, or frequency offset not exceeding Z.  The list goes on and on...
 
> So this is where I was trying to get at. What should this management
> interface look like? Should it be compliant to IEEE 1588 clause 15
> Management, or should it be a custom implementation?

If we use the standard TLVs, then we can re-use code.
 
> If the former, I guess we would have to fake a portIdentity for
> CLOCK_REALTIME, and we should make it answer the managementIDs that make
> sense (CLOCK_ACCURACY, CLOCK_DESCRIPTION, TIMESCALE_PROPERTIES,
> TIME_STATUS_NP, TRACEABILITY_PROPERTIES).  Or we could treat it like a
> full-blown ordinary clock and register all the clock data sets for one:
> defaultDS, currentDS, parentDS, timePropertiesDS, portDS.
> 
> If the latter, any ideas?

IIRC, not all management IDs need be implemented.  So we could let
ptp4l receive requests for portIdentity.portNumber = 0xffff and
forward them to phc2sys over UDS.

Thanks,
Richard



_______________________________________________
Linuxptp-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linuxptp-users

Reply via email to