[ntp:questions] Re: Dopey selection of server

Richard B. Gilbert rgilbert88 at comcast.net
Fri Jan 30 13:09:34 UTC 2004

I've experienced the same thing.  I was told that NTP was working as 
designed.  I guess I just don't understand the design!  Intuitively, it 
would seem best to select the server with the lowest delay and jitter 
but that's not what it does.  If cudns.cit.conrell.edu  consistently 
shows high delay and jitter, you might want to look for another server.

Server selection is, I think, a bit of an art.  Servers that should be 
superb are, in fact, terrible due to server or network overload; e.g. 
the various NIST servers.  I know of a stratum two server that drifts 
+/- 20ms over the course of the day and I know of others that are rock 
solid!  Unfortunately, the rock solid and apparently dead accurate 
servers are difficult to find.  And what might be rock solid and dead 
accurate for me could appear totally useless to you due to us reaching 
it by totally different network paths.  There are only about 100 stratum 
2 servers open to the public and another 100 or so open stratum 1 
servers.  Servers that are too distant are generally not very good ones 
to use; based on your choice of servers I'd guess that you are probably 
in or near New York state.
Servers in Florida, California, or Europe would have high delay and 
jitter for you but might be ideal for local users.

I'm trying to get a stratum 1 server working in my home to avoid some of 
these problems.  Unfortunately, my contract with Comcast appears to 
prohibit my making it publicly available. 

Christopher Burian wrote:

>Can anyone explain this:
>ntpq> pe
>     remote           refid      st t when poll reach   delay   offset 
>+cs.columbia.edu      2 u  725 1024  377   16.819   -7.688  
>+filbert.cc.colu     2 u  704 1024  377   22.523   -4.356  
>*cudns.cit.corne      2 u  656 1024  377   60.335   13.325 
>Instead of rejecting the one that is off the most from the others, it is 
>selecting that wayward server, despite it having the worst jitter as well.
>Is the selection based on a longer time frame, and I just happened to pick
>a bad time to look at how it's doing?

More information about the questions mailing list