Re: [gpsd-users] ntpd time accuracy with gpsd and 1PPS

[Greg Troxel]

Yakov Shmulevich <address@hidden> writes:

I have recovered formatting; your mail client incorrectly wrapped the
ntpq output.

> +SHM(0)          .GPS.            0 l    2   16  377    0.000   78.510  4.351
> *SHM(1)          .PPS.            0 l   11   16  377    0.000   -0.017  0.057
> xmivzakim.net    64.236.96.53     2 u   28   64  377   18.184   -9.893  2.426
>  LOCAL(0)        .LOCL.          10 l   63   64  377    0.000    0.000> 0.001

>    1. Did the offset (17 microseconds) give the system time accuracy
>    relatively to the time provided by GPS receiver? Or we have some
>    additional delay related to the time that take gpsd for
>    interpreting 1PPS signal? If yes, what is this time delay value?

No.  It simply says that a pps measurement was made that showed system
time 17 us ahead of the second.  By seeing 0.057 in the jitter column we
can see that generally there is 57 us of fuzz in these measurements.

It looks like the fudge value on the timecode-based time is 70ms high.

ntpd will steer the clock to the signal, so the SHM(1) line being -0.017
does not tell you about accuracy, just that ntpd is tracking the pps.

The mivzakim.net line indicates an offset of -9.8 with delay of 18, so
either it's off, you have massive asymmetric delay, your PPS is off, or
some combination.

Do you have a fudge set for the pps observations?

>    2. Did you know any tools that give possibility to measure system time
>    accuracy relatively to the time provided by GPS receiver?

See RFC2783, and the part about reflecting signals.  You can then look
at the incoming edge and the reflected edge with a storage scope
(perhaps a conventional one) and get a sense of the total latency.

But seriously, if you have serial PPS set up, you're sufficiently close
that absent using test equipment you are unlikely to ever be able to
perceive any errors.

pgpWWedVe2Xs3.pgp
Description: PGP signature