Re: Sv: [chrony-users] Multi-source: No source change from unreachable to reachable one

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




William G. Unruh __| Canadian Institute for|____ Tel: +1(604)822-3273
Physics&Astronomy _|___ Advanced Research _|____ Fax: +1(604)822-5324
UBC, Vancouver,BC _|_ Program in Cosmology |____ unruh@xxxxxxxxxxxxxx
Canada V6T 1Z1 ____|____ and Gravity ______|_ www.theory.physics.ubc.ca/

On Thu, 10 Oct 2019, chuang213 wrote:

This is similar to "chrony switch to low stratum very slow
" https://www.mail-archive.com/chrony-users@xxxxxxxxxxxxxxxxxxxx/msg01974.html

The presumption not to switch clock source often is that the frequency of system crystal is not drift during the period. This might be true
if the temperature does not change or the system crystal uses a good TCXO/OCXO.

But if the presumption is not there, 10 minutes is long enough that the system clock might drift a lot.

a) if there is a temp fluctn problem, a linear fit will be bad and the number
of samples saved for the fit will be small, and the system will transition
faster.
b) A large temp effect would be 10PPM. that would mean the system would lose
or gain 1 sec per day, 40ms/hr, 8ms/10 min. Is that intollerable?



Frank Huang

On Thu, Oct 10, 2019 at 10:12 AM Bill Unruh <unruh@xxxxxxxxxxxxxx> wrote:
      Why such a short fallover demand? Do you really not care which source controls
      your clock? chrony operates by
      accumulating a series of time comparisons, and using a fit to determine the
      current offset. Usually from ntp one is operating a a poll of something like
      6-10, or 64-1000 seconds (1-8min) between samples and it accumulates many
      samples. For an shm it usually accumulates a number of samples (eg 16)
      and uses the median, with some fraction of the samples thown out for each
      chrony sample . thus it accumulates one sample per 1/4 min. 10 min is
      40 samples.

      Why would you have as short a fallover time requirement? Have you really
      thought through the maths of the clock time to determine what the best
      strategy is when one of the clocks diappears? If the clock is stable, then the
      number of samples used is large. If unstable then small.

      Now whether newest from lost source compared to oldest used from good ntp
      source is the best selection is perhaps not entirely clear. Maybe newest from
      unreachable with middle one of the ntp source would be better. It needs to be
      carefully thought through. The problem is that often the unreachable source is
      only unreachable for a while.

      William G. Unruh __| Canadian Institute for|____ Tel: +1(604)822-3273
      Physics&Astronomy _|___ Advanced Research _|____ Fax: +1(604)822-5324
      UBC, Vancouver,BC _|_ Program in Cosmology |____ unruh@xxxxxxxxxxxxxx
      Canada V6T 1Z1 ____|____ and Gravity ______|_ www.theory.physics.ubc.ca/

      On Thu, 10 Oct 2019, Joanna Yurdal wrote:

      > In last test I have made the unreachable source was selected for 10 min.  After that I have stopped the test since it is way
      too long for our failover requirements.
      > I will not have the system access for a while but I can write more description and provide debug logs at later time.
      >
      > Fra: Miroslav Lichvar <mlichvar@xxxxxxxxxx>
      >
      > Sendt: 10. oktober 2019 16:24
      >
      > Til: chrony-users@xxxxxxxxxxxxxxxxxxxx <chrony-users@xxxxxxxxxxxxxxxxxxxx>
      >
      > Emne: Re: [chrony-users] Multi-source: No source change from unreachable to reachable one
      >
      >  
      >
      >
      > On Thu, Oct 10, 2019 at 01:53:03PM +0000, Joanna Yurdal wrote:
      >
      >> chrony.conf:
      >
      >> refclock SHM 0 poll 2 dpoll 0 refid PTP0 precision 1.0e-9 delay 1.0e-04 prefer
      >
      >> server 192.168.3.103 iburst minpoll 2
      >
      >  
      >
      >> When SHM source looses reachability, each second both s-> last_sample_ago and max_reach_sample_ago are increased by one
      >
      >
      >> The condition never becomes true and the unreachable(invalid) source remains the selected source forever.
      >
      >
      >
      > No, it shouldn't be selected forever. How long did you wait?
      >
      >
      >
      > With each sample dropped by the (reachable) NTP source
      >
      > max_reach_sample_ago should get smaller and at some point the SHM
      >
      > source should become "stale" and be unselected, if it doesn't happen
      >
      > earlier due to increased root distance.
      >
      >
      >
      >> I did not really understand the intention for this a comparison.
      >
      >> Was it indented to compare newest sample from unreachable source with the newest sample from reachable source instead?
      >
      >
      >
      > No, it's supposed to compare the newest sample of the unreachable
      >
      > source with oldest sample of a reachable source to determine if the
      >
      > spans of the two sources still overlap.
      >
      >
      >
      > --
      >
      > 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.
      >
      >
      >
      >
      > --
      > 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/