Re: [Linuxptp-users] question regarding clockcheck_sample() @clockcheck.c

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

On Thu, Oct 26, 2017 at 03:40:27PM +0000, Ruei, Eric via Linuxptp-users wrote:
> To whom it may concern:
> 
> We observed that the PTP reported "clockcheck: clock jumped" events when we  added some CPU/Memory load by using stress-ng.
> It seemed to us that the problem was triggered by the uncertainty of the mono_interval = (int64_t)mono_ts - cc->last_mono_ts at this function where mono_ts is the local time when this routine is called.
> There is some uncertainty when the function is called in particular when the system is busy and therefore there will be a large variance of mono_interval.

> ptp4l[1090.977]: interval  119990608 mono_interval  109828955
> ptp4l[1091.110]: interval  130000707 mono_interval  133090775
> ptp4l[1091.226]: interval  120002438 mono_interval  116894219
> ptp4l[1091.373]: interval  130025626 mono_interval  146662543
> ptp4l[1091.500]: interval  120022987 mono_interval  126504891
> ptp4l[1091.607]: interval  129964127 mono_interval  106861426
> ptp4l[1091.608]: clockcheck: clock jumped forward or running faster than expected!
> 
> Is it expected behavior?

If a very short sync interval is used and the process may run up to 20
milliseconds late, then yes, I think it's expected that the clock
check in default configuration will have false positives.

You may want to increase or disable sanity_freq_limit.

Just curious, on what hardware do you see this?

-- 
Miroslav Lichvar