Reference ID: MSF
Driver ID: ARCRON_MSF
/dev/arcu; 300 baud, 8-bits, 2-stop,
This driver supports the Arcron MSF receiver, and would probably
also support the DCF77 variant of the same clock. The clock reports
its ID as ``
MSFa'' to indicate MSF as a source and the use
of the ARCRON driver.
This documentation describes version V1.1 (1997/06/23) of the source
and has been tested (amongst others) against ntpd3-5.90 on Solaris-1 (SunOS 4.1.3_U1 on
an SS1 serving as a router and firewall) and against ntpd3-5.90 on
Solaris-2.5 (on a SS1+ and TurboSPARC 170MHz). This code will probably
work, and show increased stability, reduced jitter and more efficiency
(fewer context switches) with the
discipline/STREAMS module installed, but this has not been tested. For
a to-do list see the comments at the start of the code.
This code has been significantly slimmed down since the V1.0 version, roughly halving the memory footprint of its code and data.
This driver is designed to allow the unit to run from batteries as designed, for something approaching the 2.5 years expected in the usual stand-alone mode, but no battery-life measurements have been taken.
Much of this code is originally from the other refclock driver files with thanks. The code was originally made to work with the clock by Derek Mulcahy, with modifications by Damon Hart-Davis. Thanks also to Lyndon David for some of the specifications of the clock.
There is support for a Tcl/Tk monitor written by Derek Mulcahy that examines the output stats; see the ARC Rugby MSF Receiver page for more details and the code.
Look at the notes at the start of the code for further information; some of the more important details follow.
The driver interrogates the clock at each poll (ie every 64s by default) for a timestamp. The clock responds at the start of the next second (with the start bit of the first byte being on-time). The time is in `local' format, including the daylight savings adjustment when it is in effect. The driver code converts the time back to UTC.
The clock claims to be accurate to within about 20ms of the
MSF-broadcast time, and given the low data transmission speed from
clock to host, and the fact that the clock is not in continuous sync
with MSF, it seems sensible to set the `precision' of this clock to -5
or -4, -4 being used in this code, which builds in a reported
dispersion of over 63ms (ie says ``This clock is not very good.'').
You can improve the reported precision to -4 (and thus reduce the base
dispersion to about 31ms) by setting the fudge
Even a busy and slow IP link can yield lower dispersions than this from polls of primary time servers on the Internet, which reinforces the idea that this clock should be used as a backup in case of problems with such an IP link, or in the unfortunate event of failure of more accurate sources such as GPS.
By default this clock reports itself to be at stratum 2 rather than
the usual stratum 0 for a refclock, because it is not really suited to
be used as other than a backup source. The stratum reported can be
changed with the
fudge directive to be whatever you like.
After careful monitoring of your clock, and appropriate choice of the
time1 fudge factor to remove systematic errors in the
clock's reported time, you might fudge the clock to stratum 1 to allow
a stratum-2 secondary server to sync to it.
The driver code arranges to resync the clock to MSF at intervals of a little less than an hour (deliberately avoiding the same time each hour to avoid any systematic problems with the signal or host). Whilst resyncing, the driver supplements the normal polls for time from the clock with polls for the reception signal quality reported by the clock. If the signal quality is too low (0--2 out of a range of 0--5), we chose not to trust the clock until the next resync (which we bring forward by about half an hour). If we don't catch the resync, and so don't know the signal quality, we do trust the clock (because this would generally be when the signal is very good and a resync happens quickly), but we still bring the next resync forward and reduce the reported precision (and thus increase reported dispersion).
If we force resyncs to MSF too often we will needlessly exhaust the
batteries the unit runs from. During clock resync this driver tries to
take enough time samples to avoid
ntpd losing sync in
case this clock is the current peer. By default the clock would only
resync to MSF about once per day, which would almost certainly not be
acceptable for NTP purposes.
The driver does not force an immediate resync of the clock to MSF
when it starts up to avoid excessive battery drain in case
ntpd is going to be repeatedly restarted for any reason,
and also to allow enough samples of the clock to be taken for
ntpd to sync immediately to this clock (and not remain
unsynchronised or to sync briefly to another configured peer, only to
hop back in a few poll times, causing unnecessary disturbance). This
behaviour should not cause problems because the driver will not accept
the timestamps from the clock if the status flag delivered with the
time code indicates that the last resync attempt was unsuccessful, so
the initial timestamps will be close to reality, even if with up to a
day's clock drift in the worst case (the clock by default resyncs to
MSF once per day).
The clock has a peculiar RS232 arrangement where the transmit lines are powered from the receive lines, presumably to minimise battery drain. This arrangement has two consequences:
send_slow()routine to queue up command characters to be sent once every two seconds.
Three commands are sent to the clock by this driver. Each command
consists of a single letter (of which only the bottom four bits are
significant), followed by a CR (ASCII 13). Each character sent to the
clock should be followed by a delay to allow the unit to echo the
character, and then by a further 10ms. Following the echo of the
command string, there may be a response (ie in the cae of the
o commands below), which in the case of
o command may be delayed by up to 1 second so as the
start bit of the first byte of the response can arrive on time.
The commands and their responses are:
011. The leap-year logic for computing day-in-year is only valid until 2099, and the clock will ignore stamps from the clock that claim BST is in effect in the first hour of each year. If the UK parliament decides to move us to +0100/+0200 time as opposed to the current +0000/+0100 time, it is not clear what effect that will have on the time broadcast by MSF, and therefore on this driver's usefulness.
ntp.confconfiguration file for this driver might be:
# hostname(n) means we expect (n) to be the stratum at which hostname runs. #------------------------------------------------------------------------------ # SYNCHRONISATION PARTNERS # ======================== # Our betters... server 127.127.27.0 # ARCRON MSF radio clock(1). # Fudge stratum and other features as required. # ADJUST time1 VALUE FOR YOUR HOST, CLOCK AND LOCATION! fudge 127.127.27.0 stratum 1 time1 0.016 flag3 1 flag4 1 peer 220.127.116.11 # tick(1--2). peer 18.104.22.168 # tock(3), boot/NFS server. # This shouldn't get swept away unless left untouched for a long time. driftfile /var/tmp/ntp.drift #------------------------------------------------------------------------------ # RESTRICTIONS # ============ # By default, don't trust and don't allow modifications. Ignore in fact. restrict default ignore notrust nomodify # Allow others in our subnet to check us out... restrict 22.214.171.124 mask 255.255.255.0 nomodify notrust # Trust our peers for time. Don't trust others in case they are insane. restrict 127.127.27.0 nomodify restrict 126.96.36.199 nomodify restrict 188.8.131.52 nomodify # Allow anything from the local host. restrict 127.0.0.1There are a few
#defines in the code that you might wish to play with:
ntpddropping sync from the clock where possible. You may wish to undefine this, especially if you have better sources of time or your reception is ropey. However, there are many checks built in even with this flag defined.
ntp_refclock.csince the latter seems to have a minor bug, at least in version 3-5.90. If this bug goes away this flag should be turned off to avoid duplication of code. (The bug, if that's what it is, causes the last raw offset to be used rather than the median offset.)
Without this defined (and without
ARCRON_MULTIPLE_SAMPLES below) a typical set of offsets
reported and used to drive the clock-filter algorithm is (oldest
filtoffset= -4.32 -34.82 -0.78 0.89 2.76 4.58 -3.92 -2.17
Look at that spike!
With this defined a typical set of offsets is:
filtoffset= -7.06 -7.06 -2.91 -2.91 -2.91 -1.27 -9.54 -6.70
with the repeated values being some evidence of outlyers being discarded.
-4, but you may wish to set it to
-5if you are more conservative, or to
-6if you have particularly good experience with the clock and you live on the edge. Note that the
flag3fudge value will improve the reported dispersion one notch if clock signal quality is known good. So maybe just leave this alone. B^)
ntpdovershoot on clock corrections and can hold onto bad samples longer than you would like. With this set to 4 and
NKEEPset to 3 this allows the occasional bad sample (in my experience less than 1 value in 10) to be dropped. (Note that there seems to be some sort of `beat' effect in the offset with a periodicity of about 7 samples as of this writing (1997/05/11) still under investigation; a filter of approximately this length should be able to almost completely suppress this effect.) Note that if the fudge-factor
flag3is set to 1, a larger
Each timecode is written to the
clockstats file with a
signal quality value appended (`0'--`5' as reported by the clock, or
`6' for unknown).
Each resync and result (plus gaining or losing MSF sync) is logged
to the system log at level
LOG_NOTICE; note that each
resync drains the unit's batteries, so the syslog entry seems justified.
Syslog entries are of the form:
May 10 10:15:24 oolong ntpd: ARCRON: unit 0: sending resync command
May 10 10:17:32 oolong ntpd: ARCRON: sync finished, signal quality 5: OK, will use clock
May 10 11:13:01 oolong ntpd: ARCRON: unit 0: sending resync command
May 10 11:14:06 oolong ntpd: ARCRON: sync finished, signal quality -1: UNKNOWN, will use clock anyway
May 10 11:41:49 oolong ntpd: ARCRON: unit 0: sending resync command
May 10 11:43:57 oolong ntpd: ARCRON: sync finished, signal quality 5: OK, will use clock
May 10 12:39:26 oolong ntpd: ARCRON: unit 0: sending resync command
May 10 12:41:34 oolong ntpd: ARCRON: sync finished, signal quality 3: OK, will use clock
flag1 0 | 1
flag2 0 | 1
flag3 0 | 1
flag4 0 | 1
Reference Clock Drivers
ARC Rugby MSF Receiver page