[ntp:questions] Time synchronization
nomail at example.com
Sat Feb 7 20:40:45 UTC 2009
David Mills <mills at udel.edu> wrote:
> That would work fine, but does not explain the large jitter reported,
> unless gpsd displays the raw data placed in the shared segment. That
> wouldn't be very useful; the ntpq rv data would be useful and in fact
> the ultimate performance calibration. Also, why is there a PPS claimi
> but the kernel PPS is not disciplining the clock?
gpsd has a thread that watches for changes on the DCD line of the
serial port and puts the timestamps of those changes into an SHM
segment. it is up to ntpd to do something useful with that. the
user (the original poster of this thread) has chosen to call this
clock .PPS. in ntp.conf, this is not something that gpsd does.
PPS processing by gpsd does not require PPS support in the kernel.
There can be jitter when the system is heavily loaded, because there
is user process scheduling involved in the PPS handling using gpsd.
All in all, it works much better than using only NMEA or other serial
protocols with a GPS receiver, but it does not achieve the best result
that could be obtained by using a patched kernel and kernel PPS.
But then, not everybody wants to patch his kernel and some people
want to use their GPS for other purposes besides timing. gpsd
provides an access method to all of the other data from a GPS receiver.
More information about the questions