[ntp:questions] NTP will fail once a week

Richard B. Gilbert rgilbert88 at comcast.net
Wed Feb 16 19:33:47 UTC 2005


Alain wrote:

>
>
> Brad Knowles escreveu:
>
>>>  How should I test if ntpd is seeing any server?. This would be even
>>>  better than testing IP changes.
>>
>>
>>     Look at the scoreboard output of "ntpq -p".  It should be pretty 
>> obvious if you've been online for a while and yet all your configured 
>> servers are still in the ".INIT." state.
>>
> What I am really asking is: what information do I have to test? Here 
> is what I get:
> [root at servnet1 root]# ntpq -p
>      remote           refid      st t when poll reach   delay   offset 
>  jitter
> ============================================================================== 
>
> +ntp1.pucpr.br   tick.ucla.edu    2 -  212 1024  377  195.895  -16.291 
>  0.258
> *paulo.hiway.com Tick.UH.EDU      2 -  981 1024  377  242.452  -17.407 
>  7.428
>  LOCAL(0)        LOCAL(0)        10 l   10   64  377    0.000    0.000 
>   0.015
>  200.137.65.100  bb3.pop-mg.rnp.  3 -  247 1024  377  2516.67  1146.10 
> 1237.74
>  titan.cais.rnp. 174.117.23.255   2 -  994 1024  373  2026.89  900.025 
> 880.936
>  cherokee.pop-pr ntp1.pop-df.rnp  3 -  229 1024  377  212.999    5.415 
> 1284.30
> -ns.pads.ufrj.br ns2.pads.ufrj.b  2 -  995 1024  177  1102.97  216.381 
> 210.572
> -cs.columbia.edu clepsydra.dec.c  2 -  231 1024  377  338.108  -11.103 
> 21.075
> +hydrogen.cert.u 207.126.98.204.  2 -  991 1024  377  381.415  -20.140 
> 23.029
>
> Shoud I check for one of - + * in first column? For reach if all last 
> 3 are 0? Or use some combination?
>
> thanks,
> Alain

You appear to have seven reachable network servers.   The round trip 
delays range from a minimum of 195 milliseconds to two and a half 
seconds!!!   Not good!!!!

Select servers near to you (low round trip delay).   The error is 
bounded by the round trip delay; the maximum error in transmitting the 
server's time to your system cannot be more than +/- one half the round 
trip delay.   With a round trip delay of 2.5 seconds (2516.67 
milliseconds) you might just as well use a cheap wrist watch as go to 
all the trouble of running NTP.   If you are in New York, servers in Los 
Angeles, Brazil, or Tokyo are not likely to be very good choices; 
instead consider servers in New York, New Jersey, Pennsylvania and 
Massachusetts.

The * in the first column indicates the currently selected 
synchronization source.   The + in the first column designates 
additional servers that are available as an "advisory committee".  An 
"x" in the first column is a designation of "insanity".   A - in the 
first column designates servers not considered hopelessly insane but 
still not good enough to have their opinions considered at the moment.



More information about the questions mailing list