[ntp:questions] server and peer lines correct?

Charles Swiger cswiger at mac.com
Mon Apr 7 23:55:55 UTC 2014


On Apr 7, 2014, at 4:39 PM, Harlan Stenn <stenn at ntp.org> wrote:
> Charles Swiger writes:
[ ... ]
>> However, the transition doesn't happen any faster if you only use one
>> server with iburst, or with all of the server lines.  The data I've
>> seen doesn't show any significant differences to the local clock
>> variables like frequency, sys_jitter, clk_jitter.
> That's true.
> And if the one machine you choose for iburst happens to be down, it
> takes a long time to sync up.
> This is why we recommend using iburst on all server lines - it quickly
> brings each server "into the fold".

Ah, I see.

Well, if clients were only pointing to a single S2 server, then yes,
the difference between it coming up quickly or not might matter a little.
Using default poll of 1024 seconds, about ~20% of the clients would hit the
S2 with a query during the window where it was not ready to serve time.

Even so, that shouldn't cause any drastic effects on overall timekeeping.

If precise timing matters, then the clients should be configured to use (eg)
four upstream sources; perhaps the three S2s and rotate in one of the S1s
per /24 DHCP subnet or whatever is being used to organize the clients.


More information about the questions mailing list