On Fri, Aug 07, 2009 at 11:37:58AM -0700, Ask Bjjjrn Hansen wrote:
> I want [..] to have two "scores".
>
> One short term (like the current one); another score will be long term
> and use much fancier statistics and be much harder on servers with in-
> consistent results (bad time or downtime). ("Downtime" will be
> normalized per country or similar).
Hum, sounds as though it would be fun maintaining state on the longer
term value's calculation. Also, how would you combine the two values?
Also, what are actually good measures? You seem to punish downtime
quite highly, but I'm not sure if this is actually so bad. Maybe two
classes of server could exist, more "rough and ready" ones that you
can be reasonably sure they're going to be alive and another set of
ones where the client is more tolerant to uptime, but variance is more
important. For people who want to sync their time once off, it doesn't
matter much if time isn't so good (i.e. on the order of a second). For
people who run a proper daemon then they're going to be tolerant of a
server or two being down for a day, but would prefer those that are up
to be giving good time.
--
Sam http://samason.me.uk/
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
