On 9/14/07, Tim Shoppa <[EMAIL PROTECTED]> wrote:
> One purpose of distributing the monitoring servers is so that poor
> network connectivity/asymmetric latency at one doesn't mess up
> the scores everywhere. Diversity of network paths and endpoints
> is the goal, not letting a bad single network path drag things down.
Obviusly it would be best to have the monitoring servers hosted at
major peering point to ensure netowork path diversity. But that isn't
really achievable with a volunteer effort, and the most important
peering points are now private (at least in the USA).
One goal would be to have most - or perhaps just many - NTP servers in
the pool monitor a randomized subset of the the others in the pool
with "nopeer", and then report the results centrally somehow.
If we had each server in the pool monitor 10 others (randomly chosen),
that would give us a hugely diverse monitoring data set to work with.
But it would add minimal additional load for any single node. The
central monitoring server(s) could then use ntpq to pick up
reachability and quality statistics from monitoring servers, and
coallate the results into quality scores.
--
RPM
=========================
All problems can be solved by diplomacy, but violence and treachery
are equally effective, and more fun.
-Anonymous
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
