[ntp:questions] Re: NTP broadcastclient update times?

Richard B. Gilbert rgilbert88 at comcast.net
Fri Oct 15 21:22:59 UTC 2004


W. D. wrote:

>Harlan Stenn wrote:
>  
>
>>What is ...2.122?  
>>    
>>
>
>FreeBSD / NTP time server
>
>
>  
>
>>How many machines is it sync'd to?
>>    
>>
>
>See below.
>
>
>  
>
>>What OS?
>>    
>>
>
>FreeBSD 4.9
>
>
>  
>
>>How about an ntpq -p against 2.122?
>>    
>>
>
>ntpq -p
>
>     remote           refid      st t when poll reach   delay   offset 
>jitter
>==============================================================================
>-now.cis.okstate .PSC.            1 u  307 1024  377  226.292  -46.779  
>5.793
>+navobs1.wustl.e .USNO.           1 u  993 1024  377  196.782  -29.412  
>0.498
>*clock.xmission. .GPS.            1 u  914 1024  377  198.391  -31.001  
>2.665
>+bonehed.lcs.mit .CDMA.           1 u  853 1024  377  194.764  -29.863  
>0.985
>-clock.via.net   .GPS             1 u  303 1024  377  227.888  -47.328  
>7.590
>-otc1.psu.edu    .WWV.            1 u  874 1024  377  199.845  -48.614  
>3.173
>+timekeeper.isi. .GPS.            1 u   38 1024  377  231.352  -30.935
>528.528
>-ntp-cup.externa .GPS.            1 u  318 1024  377  233.925  -35.138  
>7.098
>-time-B.timefreq .ACTS.           1 u   41 1024  377  231.446  -53.467
>280.040
>-clock.isc.org   clepsydra.dec.c  2 u  307 1024  377  227.118  -34.082  
>7.550
> 192.168.2.255   .BCST.          16 u    -   64    0    0.000    0.000
>4000.00
>  
>
There are two very notable things about this output.

First, you have extremely high round trip delays to all the servers you 
are using.  This can be caused by network congestion or by very long 
network paths between your site and the servers.  In addition to 
physical distance you must also consider the number of routers the 
packets pass through; each one adds some delay in each direction.   
Given a private wire from the east coast of the US to the west cost, the 
speed of light delay should be less than thirty milliseconds so you are 
seeing something more than simple distance.

Finding servers closer to you in net space should substantially reduce 
the delay and also the offsets.  Assuming that the servers have the 
correct time in first place, the error/uncertainty is bounded by the delay.
With delays over 200ms, ntp is essentially guessing what time it is.

The second notable thing is that you are using numerous stratum 1 
servers and, unless you are serving several hundred clients, you should 
not be.  A few stratum two servers much closer (lower delay) should give 
you much better performance and accuracy.

If  your network situation is such that there are no available servers 
with delays of less than 25 milliseconds, you should probably consider a 
hardware reference clock such as a GPS receiver.




More information about the questions mailing list