Question about the technique for converting PHC timestamps into
monotonic deadlines.

   clock_gettime(CLOCK_MONOTONIC, t1);
   clock_gettime(phc, t2);
   clock_gettime(CLOCK_MONOTONIC, t3);
   offset = ((t3 - t1) / 2) - t2

Using code like this, I seem to observe a drift of about ~1
microsecond per second. That is, the offset from the system monotonic
clock to the PHC clock grows by ~1000 nanoseconds each second.

That is not what I would expect .. both domains have nanosecond
precision. That's quite a lot of drift to compensate for. Any ideas?

-Todd

On Sun, Oct 2, 2022 at 1:07 PM Richard Cochran <richardcoch...@gmail.com> wrote:
>
> On Sat, Oct 01, 2022 at 08:47:53PM -0700, todd freed wrote:
> > Could you say just a bit more about why it shouldn't be implemented?
>
> I'm only saying that the potential benefit is very meager and does not
> justify the implementation effort.
>
> > I'm a professional software engineer, and I had thought I might take a
> > look at implementing it myself. But if you wouldn't recommend such a
> > thing, I will avoid spending any time on it.
>
> Before you do anything, I recommend reading these threads from
> archives from 2016:
>
>  31.Aug'16 Kieran Tyrrell       Re: [Linuxptp-users] one-shot alarm
>  31.Aug'16 To Kieran Tyrre      └─>
>  31.Aug'16 Dale Smith             ├─>Re: [Linuxptp-devel] [Linuxptp-users] 
> one-shot alarm
>  09.Sep'16 Kieran Tyrrell         └─>Re: [Linuxptp-users] one-shot alarm
>  09.Sep'16 Kieran Tyrrell       [Linuxptp-devel] [PATCH] PTP subsystem: 
> implement POSIX timer interface
>  12.Sep'16 To Kieran Tyrre      └─>
>  12.Sep'16 Kieran Tyrrell         ├─>
>  12.Sep'16 To Kieran Tyrre        │ └─>
>  15.Sep'16 Kieran Tyrrell         │   ├─>
>  15.Sep'16 To Kieran Tyrre        │   │ └─>
>  15.Sep'16 Kieran Tyrrell         │   │   └─>
>  18.Oct'16 Kieran Tyrrell         │   └─>igb tsync int handler double 
> acknowledge? (was: Re: [Linuxptp-devel] [PATCH] PTP subsystem: implement 
> POSIX ti
>  18.Oct'16 To Kieran Tyrre        │     └─>Re: [Linuxptp-devel] igb tsync int 
> handler double acknowledge? (was: Re: [PATCH] PTP subsystem: implement PO
>  18.Oct'16 To Kieran Tyrre        │       ├=>Re: igb tsync int handler double 
> acknowledge? (was: Re: [Linuxptp-devel] [PATCH] PTP subsystem: implement
>  19.Oct'16 Keller, Jacob E        │       └─>Re: [Linuxptp-devel] igb tsync 
> int handler double acknowledge? (was: Re: [PATCH] PTP subsystem: implement
>  13.Sep'16 Kieran Tyrrell         └─>
>  13.Sep'16 To Kieran Tyrre          └─>
>  15.Sep'16 Kieran Tyrrell             └─>
>  15.Sep'16 To Kieran Tyrre              └─>
>  09.Sep'16 Kieran Tyrrell       [Linuxptp-devel] [PATCH] igb: add timer 
> (alarm) functionality
>  12.Sep'16 To Kieran Tyrre      └─>
>  13.Sep'16 Kieran Tyrrell         └─>
>  13.Sep'16 To Kieran Tyrre          └─>
>  14.Sep'16 Kieran Tyrrell             └─>
>  15.Sep'16 Kieran Tyrrell       [Linuxptp-devel] [PATCH V2] ptp and igb: 
> implement POSIX timer (alarm)
>  03.Oct'16 To Kieran Tyrre      └─>
>  06.Dec'16 To Kieran Tyrre        └─>
>
> Especially this last message ^^^ shows the poor performance of a PCIe card:
>
> TL;DR graph...
>
>    https://linuxptp.sourceforge.net/phc-timer/phc-timer-vs-nanosleep.png
>
> Good luck,
> Richard
>


_______________________________________________
Linuxptp-users mailing list
Linuxptp-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linuxptp-users

Reply via email to