Hauke Lampe wrote:
> On Fri, Jan 27, 2006 at 16:38:02 +0100, Benoit Panizzon wrote:
>
>   
>> I have two Stratum 0 Servers connected to a DCF77 Clock and use
>> several pool servers to sync with.
>>
>> During the last few days I noticed my DCF Clocks not being accepted
>> anymore because of the offset becoming quite large (20ms at the
>> moment) while all other NTP servers stay within a few ms...
>>     
>
> Does your server have an ADSL connection?
> The asymmetry distorts NTP's network delay calculation.
>
> I have seen the same behaviour with my timeserver at home which has an old
> DCF receiver connected to the serial port. When it's synced to the DCF
> signal, it reports about 30ms offset to external timeservers. Whenever it
> lost the DCF signal, it began to drift towards the external servers and back
> again when its (preferred) refclock came back.
>
> That's why I only use the DCF receiver and let the clock run free in times
> of bad reception (about 5% of the time, according to ntpq).
>   
A typical DSL asymmetry will yield between 1 and 5ms of offset depending
on the line speed (you can work it out using the size of an NTP packet
and the data rates).  If the line is heavily loaded then all bets are
off as traffic will skew the numbers wildly.    My experience is that
most of the net is asymmetrical to the order of 1-5ms.

ntpq> pe
     remote           refid      st t when poll reach   delay   offset 
jitter
==============================================================================
*GPS_NMEA(0)     .GPS.            0 l   11   16  377    0.000   -0.006  
0.008
-jeeves.localnet .GPS.            1 u   28   64  177    0.985   -0.037  
0.113
-clock.sjc.he.ne .CDMA.           1 u   61  256  377   18.835    1.820  
0.377
+ntp1.symmetrico .GPS.            1 u   36  256  377   25.501    1.570  
0.760
+clock.isc.org   .GPSi.           1 u   70  256  377   24.934    1.229  
0.734
-mewmie.mainecoo 63.192.96.10     2 u   46  256  377   30.652    2.330  
0.720

As you can see my two local stratum 0 clocks show about 1.5 ms offset to
external S1 refclocks - this is a combination of my DSL offset and my
ISP having an asymmetrical path to the other clocks (as is typical
traceroute to a host and from a host will take completely different paths).

If a refclock is 30ms off it's either on a very very skewed or over
loaded line or it's not got the right radio propagation delay factored in.

John

_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to