skip to navigation
skip to content

ntpdshm 0.2.1

Python interface to NTP Shared Memory


python-ntpdshm provides a Python interface to ntpd’s shared memory driver 28. A single class NtpdShm exposes the fields of the shared memory structure as attributes that can be read and written. In addition there are properties to set the clock and receive timestamps from float values. There is also a convenience update() function for setting the time related fields in a single step.

python-ntpdshm is implemented using Swig.

python-ntpdshm works with the following Python versions.

  • Python 2.6
  • Python 2.7
  • Python 3.3
  • Python 3.4
  • Python 3.5
  • PyPy (but not PyPy3!)


import ntpdshm

ntpd_shm = ntpdshm.NtpdShm(unit=0)

The members of the C struct can be accessed by their original names. These have not been converted into PEP-8 compliant names.

print ntpd_shm.mode
print ntpd_shm.clockTimeStampSec
print ntpd_shm.clockTimeStampUSec
print ntpd_shm.clockTimeStampNSec      # only ntpd 4.2.7p303 or later, probably random value otherwise
print ntpd_shm.receiveTimeStampSec
print ntpd_shm.receiveTimeStampUSec
print ntpd_shm.receiveTimeStampNSec    # only ntpd 4.2.7p303 or later, probably random value otherwise
print ntpd_shm.leap
print ntpd_shm.precision
print ntpd_shm.valid

In addition there are two pseudo properties that combine the second and microsecond attributes into “float” timestamps. These don’t support nanosecond precision as (as far as I know) it is not possible to detect whether ntpd does support nanosecond resolution.

print ntpd_shm.clockTimeStamp          # clockTimeStampSec.clockTimeStampUSec
print ntpd_shm.receiveTimeStamp        # receiveTimeStampSec.receiveTimeStampUSec

The process to feed ntpd an external reference time is shown below.

import time

clock_time = get_clock_time()          # `get_clock_time` must be implemented somewhere else and
                                       # return a float.
recev_time = time.time()
ntpd_shm.valid = 0                     # don't use Python boolean
ntpd_shm.clockTimeStamp = clock_time
ntpd_shm.receiveTimeStamp = recv_time
ntpd_shm.precision = -5                # 2^-5 = 0.03125 seconds in this case
ntpd_shm.count += 1
ntpd_shm.valid = 1

As this is somewhat cumbersome, there is a convenience method update() that achieves the same in a single line. It requires the clock_time as mandatory argument and accepts several optional arguments.

ntpd_shm.update(clock_time, recv_time=recv_time, precision=-5)

# Or simply, if no other fields are to be changed. The receive timestamp is set
# automatically.


“Off by one second” reference time

A just for fun example of using python-ntpdshm is to implement an “off by one second” reference time source for ntpd. While this example makes no sense at all for practical purposes it provides a useful template for how it all fits together.

First we write the code for the reference clock.

import time
import ntpdshm

def get_clock_time():
    return time.time() - 1.0     # always be exactly one second behind.

def main():
    ntpd_shm = ntpdshm.NtpdShm(unit=2)
    ntpd_shm.mode = 0            # set mode
    ntpd_shm.precision = -6      # set precision once
    ntpd_shm.leap = 0            # how would we know about leap seconds?

    while True:
        clock_time = get_clock_time()

if __name__ == '__main__':

Then add the shared memory reference clock to ntp.conf:

# ntp.conf
server noselect     # unit=2, never select this reference
fudge refid PYTH stratum 10

Restart ntpd and monitor the output of ntpq -pn. The offset should be exactly -1000 msec:

$ ntpq -pn
     remote           refid      st t when poll reach   delay   offset  jitter
...    .PYTH.          10 l    9   16  377    0.000  -1000.0   0.017


0.1.0 (15-Dec-2015)

  • Initial version
File Type Py Version Uploaded on Size
ntpdshm-0.2.1.tar.gz (md5) Source 2016-11-12 41KB