[ntp:questions] question regarding NTP configuration for clusters, and "cluster time" stability

Richard B. Gilbert rgilbert88 at comcast.net
Tue Oct 6 19:39:47 UTC 2009


Terje Mathisen wrote:
> rotordyn at yahoo.com wrote:
>> On Oct 6, 11:52 am, Terje Mathisen<Terje.Mathi... at tmsw.no>  wrote:
>>
>>> Having an external server fail but still reply (a "falseticker") is an
>>> expected issue, it is solved by having at least 4 servers.
>>
>> Right. But even if we require 4, there is the unlikely but possible
>> scenario of 3 going offline during the same period that the last one
>> is ill-behaved.
> 
> At his point you need more servers:
> 
> 4 allow for a single falseticker, then I believe you need 6 to handle 
> two errors and 7 or 8 for three simultaneous falsetickers.
> 
Five servers allows for two failures!  Seven allows for three.  If you 
have three or more simultaneous failures, you need more help than we can 
provide!

The general idea is to have enough "sane" servers to outvote the insane 
servers.

> You should know though that nobody has ever seen this happen, over the 
> entire history of the ntp protocol, unless all those servers have a 
> common reference: If you make sure you have several independent stratum 
> 1 or 2 servers, then you can very safely bet that it simply won't happen.
> 
> You _can_ have a backhoe cutting all your internet connections at the 
> same time, this is far more likely than multiple simultaneous errors all 
> going in the same direction (and within a few ms from each other).
> 
> Terje




More information about the questions mailing list