[ntp:questions] Using two NTP Server: Bad?

David Lord snews at lordynet.org
Thu Apr 21 10:10:03 UTC 2011

Ben Rockwood wrote:
> I've read in the past that clients should always have an odd-number of
> NTP servers; 1 server or 3 servers but not 2.  If I recall the reason
> was that clients could become "confused" and needs a tie breaker.
> Question 1: I want to challenge this old assumption.  Is there truth to
> it? 

Yes and no answer.

I've found it handy to run two or more ntp servers
but if you can stand the downtime for upgrades using
a single server is no problem.

The assumption you refer to is regarding the number
of sources used by ntpd (server lines in ntp.conf).
For a single local server I'd suggest using at least
five sources and for two local servers select at least
another couple of sources to add to the total polled.
There are utilities in the ntpd distribution for
testing remote servers as to their source and the
distance from your site. I try to use the closest
but also try to avoid those that are using the same
reference when I test.


> In many environments there is a desire to have 2 NTP servers for
> redundancy purposes, but not 3 due to limited resources.
> Question 2: Furthermore, if you have 2 local NTP servers is it
> preferable to have them sync off of different sources to avoid a client
> syncing servers that are using the same reference clock?  ie: Is this bad?:
> $ ntpq -pn
>      remote           refid      st t when poll reach   delay   offset 
> jitter
> ==============================================================================
> *    3 u  756 1024  377    0.494    0.596  
> 0.352
> +    3 u  953 1024  377    0.135    0.144  
> 0.241
> Thank you.
> benr.

More information about the questions mailing list