[ntp:questions] Re: NTP Pool rules vs. four-server advice
brian.utterback at sun.removeme.com
Mon Dec 19 16:30:12 UTC 2005
Michael Deutschmann wrote:
> I've noticed a conflict between the instructions on the NTP server Pool
> project page and other NTP advice.
> It is commonly advised here to configure ntpd either to slave to 1
> server, or to track 4 or more. The reason being that, when not slaving
> blindly to one server, NTP's algorithms require 3 good servers, and if you
> know all three will remain good, you may as well blindly track one.
This is a pretty simplistic view of why 4 servers are needed. One and
two server configurations each have their own problems, but either
might be what you need, depending on the failure modes you are
which to deal with.
Three servers is better than two, and may be adequate for many types
of situations. It has the unfortunate characteristic of becoming
a two server configuration if one server becomes unavailable. Since
it is almost certain that one will eventually become unavailable,
at least temporarily, this could be a significant drawback. Three
servers can protect from certain classes of falseticker situations
involving one falseticker, but some combinations may not be prevented.
It takes four servers to protect against one of them becoming a
falseticker for all such situations. After that, add one more server
to for each additional falseticker you wish to be protected from.
But even if you do not anticipate any falsetickers, network outages
and failures can amount to the same thing. The more independent
servers you have, the less likely that these types of things will
affect your system. But in general, it is difficult to get complete
independence, so you need to examine this carefully.
I find that with one or two servers, I often will get occasional
time steps. With three, it is very rare and with four or more,
I never get them, unless there is a hardware problem.
"Having them stolen may become our distribution model..."
Nicolas Negroponte on the Hundred Dollar Laptop.
Brian Utterback - OP/N1 RPE, Sun Microsystems, Inc.
More information about the questions