[ntp:questions] frequency adjusting only
unruh-spam at physics.ubc.ca
Wed Apr 30 17:06:13 UTC 2008
m.louvel at gmail.com (maxime louvel) writes:
>On Tue, Apr 29, 2008 at 6:27 PM, Unruh <unruh-spam at physics.ubc.ca> wrote:
>> m.louvel at gmail.com (maxime louvel) writes:
>> >I have know run a lot of tests.
>> >Just to let you know what I've got so far.
>> >I have tried NTP, and NTP + PTP (Precision Time Protocol).
>> >I haven't tried Chrony nor TSClock.
>> >I have used the software only implementation of PTP (ptpd).
>> >With NTP only I have got an accuracy around 1ms,
>> accuracy? Do you mean the variation of the offset between ntp and the
>> remote ntp server? YOu need to tell us what you mean by the term.
>By accuracy I actually meant the offset between my nodes.
And that offset is measured how? With a GPS receiver on each of the nodes?
>From teh peerstats of the ntp? From ntpq -p?
And is the offset constant or fluctuates by 1msec.
Note that the ntpq -p offset is in milliseconds, so you would get things
for the offset not .001, if that is how you determine it, for msec jitter.
>> >with one node synchronising to public NTP server, the others nodes
>> >synchronising to it with PTP I got 10 usec (between the nodes).
>> >I haven't used any hardware support (not GPS,PPS, for PTP).
>> Then you do not really know what the accuracy is of the other nodes. For
>> example, your system could have a 100us assymmetry in communication which
>> would make the accuracy 50us, but with a fluctuation of 10us.
>You're right on this point, what I actually get with PTP is delay of 100
>usec, with a variation of 10 usec.
>But I'm only interested in the variation as the 100usec can be removed
>easily because it's constant
>> >I might try the other solutions, if yeI do I'll send an email to complete
>> >the results.
>> As I said, I am completely at a loss to know why you would only be getting
>> 1ms using ntp to synchronize. That is much worse than my experience. (Look
>> at flory on www.theory.physics.ubc.ca/chrony/chrony.html which uses ntp vs
>> the other machine which all use chrony. flory's fluctutations are
>> certainly not 1ms.)
>I have to say that I 'm also surprise of the bad performance of NTP, I will
>try chrony if it is suitable to my nodes (embedded cards, with the latest
>> questions mailing list
>> questions at lists.ntp.org
>0044 7964 5555 80
>43 Allen road
More information about the questions