[ntp:questions] userland shm pps, was Re: Discipline performance comparison of SHM(1) and Atom(PPS)

gnu not unix gnu at wraith.sf.ca.us
Thu Jun 11 16:34:20 UTC 2015

Hello from California--

On Thu, 2015-06-11 at 14:18 +0200, f.hiesinger wrote:
>  Hi Folks,
>  I wonder, if a SHM(1) disciplined NTPd can reach the same accuracy than 
>  one with the ATOM(PPS) Driver.
>  [1] says, that SHM(1) is used for "PPS from GPS". SHM(0) is sourced 
>  from NMEA-Messages.
>  I'm currently parsing very precise timestamps via SHM(1) to NTP 
>  (Operation mode=0). Every timestamp in SHM(1) is derived from a 
>  pps-source and tweaked individually. Up to here everything works fine, 
>  including Nanosecond resolution. See ntpshmmon output:

I modified the user mode parallel port/serial port SHM PPS driver for
nanosecond support with NTP (get the code here:

It has worked well under linux. I've been plotting the SHM on host
wraith and linux PPS on host starshine performance here:


The plots show that userland SHM PPS driver gives me 2000 nanoseconds of
flap, and the linux PPS driver about 1000 nanoseconds. 

The userland pulses are simply passed to ntp with no grooming. 

Oh I guess the ntpshmmon is part of the gpsd package. 

I'm still working on the userland SHM driver. I tried to set a few
serial port parameters but got into trouble (cough). I'll probably
rip that out. Its working swell, even though... I'm looking for feedback
if anyone tests it out. I recently built it on 64bit systems also.

gnu at wraith.sf.ca.us

More information about the questions mailing list