Re: [chrony-users] Chrony not taking SOCKET data from Application

[ Thread Index | Date Index | More chrony.tuxfamily.org/chrony-users Archives ]



On Fri, 24 Feb 2023, sarveshwar k wrote:

[CAUTION: Non-UBC Email]Hi Miroslav Lichvar,
Can I use pps directive to a PPS refclock?
refclock PPS /dev/pps1 poll 0 pps lock GPS refid PPS trust prefer

Will this increase the accuracy?

I am using below refclocks:
refclock PPS /dev/pps1 poll 0 lock GPS refid PPS trust prefer
refclock SOCK /var/run/chrony/udssocket.sock poll 0 refid GPS noselect

It is not going to poll at GHz rates (that is the speed of the computer) I am
not sure but suspect polling would be better, but could also be more subject
to delays if the computer is busy. Do not know if anyone has compared polling
to interrupt.


As we are using GPS and 1PPS, our assumption is chrony synchronizes system clock with nanosecond
level accuracy. Am I right on this? But we see microsecond level accuracy even after using
makestep directive.

No. It cannot. The clock does not even send out the PPS with ns accuracy. Most
gps receivers put out a pps signal which is a sawtooth accuracy of tens of ns
displacement which can be corrected later, if you have a timeing gps. Furthermore one needs to set up the PPS cabling so as to impedence match
between the gps receiver, the cabling and the computer interface. or the pps
edge will get rounded off and smeared out with an unknown relation to the
actual time to probably hundreds of nanoseconds. Futher more the computer has
to notice that the pulse has come in and to read the system clock which is of
the order of a microsecond. Now if you had a super accurate clock that you
could compare the computer time with the UPS time, then some of that error is
systematic and could be subtraced off. The random fluctuation can be averaged
out in chrony to maybe 100ns, but of course systematic errors cannot be
averaged out. So you are probably goot to a microsecond, not nanoseconds.


Thanks & Regards
Sarveshwar.K

On Mon, Feb 20, 2023 at 3:56 PM sarveshwar k <sarveshwar.kt@xxxxxxxxx> wrote:
      Hi Miroslav,
Thanks for the information.

>My question would be, why do you need the system clock to be so
>accurate? The CPU is connected to the outside world over PCIe, which
>has latency and possibly also asymmetry in hundreds of nanoseconds.
>What do you do on that computer that this accuracy would make a
>difference?
This system clock is being used by other applications like IMU sensor data timestamping. 
We are having this chrony running in an ECU (Electronic Control Unit) along with PTP stack
to send PTP packets to other ECUs in the network to get synced to Master ECU.

Thanks & Regards
Sarveshwar.K

On Thu, Feb 16, 2023 at 1:59 PM Miroslav Lichvar <mlichvar@xxxxxxxxxx> wrote:
      On Thu, Feb 16, 2023 at 10:04:02AM +0530, sarveshwar k wrote:
      > Is it not possible to achieve synchronization accuracy less than 100ns with
      > chrony (with GPS and 1PPS).

      With HW timestamping of the PPS signal (e.g. on the I210) it is
      possible, but difficult to verify.

      My question would be, why do you need the system clock to be so
      accurate? The CPU is connected to the outside world over PCIe, which
      has latency and possibly also asymmetry in hundreds of nanoseconds.
      What do you do on that computer that this accuracy would make a
      difference?

      > Can we configure chrony behaving as a measurement tool (not to make any
      > corrections)?

      There is the chronyd -x option or the refclock noselect option for
      that.

      --
      Miroslav Lichvar


      --
      To unsubscribe email chrony-users-request@xxxxxxxxxxxxxxxxxxxx
      with "unsubscribe" in the subject.
      For help email chrony-users-request@xxxxxxxxxxxxxxxxxxxx
      with "help" in the subject.
      Trouble?  Email listmaster@xxxxxxxxxxxxxxxxxxxx.




Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/