[ntp:questions] Re: Philosophical question about strata
brian.utterback at sun.removeme.com
Fri Oct 7 14:38:56 UTC 2005
Lee Sailer wrote:
> I had this thought about strata and I am hoping someone might have some
> ideas to contribute.
> Suppose you have two ntp clients that get their time from multiple
> strata 1 servers. They should be very accurate statum 2 servers as a
> Now later, you are configuring a new client, and for some reason you
> have a choice. You can use one of the stratum 1 servers, or the
> stratum 2 server. (I know, make another choice. This is a theoretical
> question, ok?)
> It seems to me that the right choice is the stratum 2 server. Am I
> However, if you configure this new client so that it uses one stratum 1
> the stratum 2, I think the algorithm will choose the stratum 1. Again,
> am I wrong?
> Basically, my thought is that it is possible for a stratum 2 source to
> be better than a stratum 1 source, and likewise for a stratum n source
> to be better than a stratum m source, where n > m.
> Comments welcome. No flames please.
> Flamed online since 1977
Well, assuming that the stratum 1 server really has a correctly
configured refclock on it, the stratum 2 server could not be more
accurate than the stratum 1 server. However, the error introduced
by each stratum is dependent on the network characteristics between
the client and the server, so it is certainly possible that the
stratum 3 client could get more accurate time from the stratum 2
server than the stratum 1. And if we take the thought experiment
even further, for any given client, it is possible that the most
accurate time available to it could be from a server at any stratum
including higher numbered stratums than its own. The stratum has
very little to do with the accuracy. NTP choses low numbered stratum
primarily as a mechanism to prevent loops.
Sung to the tune of "Who Put the Bomp" by Barry Mann:
"Who put the Java in the Java Desktop?
"Who put the Ultra in the Ultra 20?"...oh wait...Nobody.
Brian Utterback - OP/N1 RPE, Sun Microsystems, Inc.
More information about the questions