[time-nuts] TSIP, SCIP time sync
Hal Murray
hmurray at megapathdsl.net
Tue Mar 23 02:39:51 UTC 2010
holrum at hotmail.com said:
> The timing message ends 40-45 msec after the 1PPS pulse. Interesting that
> is it is not a fixed or random delay. It seems to be quantized to 40, 45,
> or occasionally 50 msec. The delay does not change randomly, but seems to
> settle on a particular value for extended periods. Heather now assumes it
> is 45 msec. Not sure how much of the message jitter is from the Tbolt or
> Windoze.
I'm running NTP on Linux. I'm seeing about 10 ms of jitter. That's after
all of NTP's filtering.
My stuff changed when I turn on the low-latency flag for that serial port. I
didn't say "got better". The graph looks cleaner. I think it removed some
of the high frequency jitter. The remainder reminds me of the hanging
bridges - very hard to filter out.
> Anyway it looks like Heather can keep your Windoze clock fairly accurate...
> probably within a Windoze timer tick (or two). One could probably do
> better by hooking the 1PPS signal to a modem control interrupt (Carrier
> Detect?) but that opens up a fresh can 'o nematodes.
That's the way to go if you want good timing. Beware, there is a big
rat-hole in making computers keep good time, and it's got "time-nuts" written
all over it.
DCD is the common pin for that purpose. Some OSes support other pins.
Pin 1 (DCD) on the TBolt isn't connected to anything. I added a jumper over
to the PPS connector which is conveniently nearby. More news when the
self-survey finishes. (Which may take a long time since my antenna setup is
poor.)
Pins 6 (DSR) and 9 (RI) have traces that are visible. One goes to an empty
resistor. The other goes to chip marked "232". I assume it's a level
converter. If so, it's an output pin. All the documentation I've found says
"Not used". Maybe I mistraced someting.
--
These are my opinions, not necessarily my employer's. I hate spam.
More information about the Time-nuts_lists.febo.com
mailing list