[ntp:questions] Re: Netra X1 time warp

Roman Maeder maeder+news at mathconsult.ch
Sun Jan 25 22:46:35 UTC 2004


David L. Mills wrote:

> I see nothing like that in the herd here; could the glitches be from the
> radio?

I assume there is not code in the refclock drivers that would allow that 
driver to mess with the system time by itself?
The radio is one of thouse "mouse clocks", configured as

	server 127.127.8.0 mode 14 prefer
	fudge  127.127.8.0 time1 0.25376 flag1 1 flag2 1

To make up for the crappy serial port of the Netra X1 I inserted a RS232 
amplifier into the cable, and it usually performs well. The time code does 
have only a two-digit year, but that doesn't seem to lead to any problems.

Also, there may be many more glitches that don't hit the logs due
> the popcorn spike filters and clock filter and step processing. The
> spikes would have to result in continue outside the 128-ms step
> threshold and last longer than the 900-s stepout threshold. Is this what
> you see?

There are no consistent errors in the refclock that would last that long. When 
an occasional wrong time code satisfies the parity checks, it will only lead 
to a single spike, easily handled by ntpd. Sometimes, when the wheather is 
bad, there is almost no good time code during several hours, and the asterisk 
in the "ntpq -p" output just switches to one of the other configured servers.


Added in print:

My other Netra X1 just produced a 50s glitch, which ntpd can handle, but it 
seems to attribute such inexplicable sudden offsets to a frequency error, and 
thus shakes quite a bit, exceeding the 128ms threshold a total of three times 
before settling again:

53029 30357.459 0.003431935 19.932398 0.001736596 0.002554 8
53029 30614.452 0.003291704 19.935537 0.001603221 0.002712 8
53029 31141.474 0.000000000 19.935537 0.001388430 0.002349 6
53029 31194.446 0.000000000 19.935537 0.001202416 0.002034 6
...
53029 31967.451 0.000000000 19.935537 0.000214013 0.000362 6
53029 32981.460 0.000000000 500.000000 0.001378185 25462.759145 4	time reset 
+50.008859 s
53029 33053.469 -0.030573173 499.883373 0.001510144 22051.396270 6
53029 33122.474 -0.030573173 499.766745 0.006928128 19097.069359 6
53029 33442.475 -0.035265806 499.632217 0.052716149 16538.547203 6
53029 34461.841 0.000000000 -205.942015 0.109682033 14327.146137 4	time reset 
-0.650857 s
53029 34533.468 -0.013358020 -205.992972 0.094987895 12407.672518 6
53029 34606.474 -0.013358020 -206.043929 0.082330665 10745.359603 6
53029 34671.468 -0.012895945 -206.093123 0.071650569 9305.754389 6
53029 34865.470 -0.011509047 -206.137027 0.065932046 8059.019702 6
53029 34928.472 -0.011045937 -206.178505 0.063682059 6979.315792 6
53029 35378.467 0.084992733 -205.854284 0.066296771 6044.264779 6
53029 36838.818 0.000000000 29.558552 0.049110071 5235.810094 4		time reset 
+0.351402 s
53029 36909.474 0.100178917 29.940704 0.042531369 4534.344555 6
...
53029 39301.473 -0.002078656 31.074376 0.019585416 1434.694962 6

This machine synchronizes to the aforementioned Netra and one external source, 
which performed normally during this time. The ripples then propagate to the 
broadcast clients of this machine.

I am beginning to hate it.

Roman




More information about the questions mailing list