On Oct 13, 2007, at 9:50, Evandro Menezes wrote:

> And it seems that allowing for 100ms offset due to network traffic  
> is perhaps too much leeway. [...]

For now 100ms is "close enough".  For SNTP clients it's plenty  
accurate.  As you noticed, ntpd/NTP clients will figure out to pick  
another server.

> Finally, instead of an SNTP-way of probing the servers, why not run  
> NTP with a appropriate minpoll and noselect to evaluate the actual  
> offset better?

Mostly because ntpd doesn't let you remove and add servers "online",  
so it'd take longer to add a new server to the monitoring.   For the  
future distributed monitoring I'm planning to set it up that way though.


   - ask

-- 
http://develooper.com/ - http://askask.com/


_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to