On Sun, Aug 24, 2008 at 3:53 PM, Ask Bjørn Hansen <[EMAIL PROTECTED]> wrote:
> Well - if we do our job right with the monitoring system etc, then the
> only pre-requisite really is that you're willing and planning to be in
> "in it for the long haul" (until more of the ntp implementations learn
> to update the server list more frequently).

Indeed, I was going to reply and mention that, having a server in the
South American pool, I tend to see things differently -- while the
US and European (at least) pools tend to be quite well-served by
the pool, I'm one of a mere 17 servers for all of South America. At
a modest 3 Mbps setting, I had something like 4,000 active clients
when I last checked the stats... So I've always seen the pool as
something that we want to encourage anyone and everyone who
understands NTP to contribute to. The monitoring system
enforces tight standards, so that there's really no reason an
admin needs to be an NTP whiz.

I wasn't going to bother the list with the observation, but since
Ask mentioned it... I delisted one of my servers in June, in
preparation for taking it offline. It had been handling close to
4,000 active clients at the time. I kept ntpd running on the
machine right through the end, and the plug was finally pulled
on Friday night, when Wayne's scripts were still reporting
about 1,000 active clients. (Which could be loosely interpreted
as meaning that about 25% of clients go at least 60 days without
resetting.)

Has there been discussion in the past about making ntpd honor
DNS TTLs? It seems to me that it should be a fairly easy
change to make to the code, and that it would eliminate
problems like a "bad" server being pulled from the pool
but still being used by those who got that IP in the past, but
it could probably introduce new problems, like a low
TTL meaning that servers could theoretically be 'aged
out' via TTL before the maximum polling interview is ever
reached, resulting in a net increase in the load on NTP
servers... (And, for cases where the "IP" is an array from
pool.ntp.org, it may mean that all the servers would get
replaced at once... But even a cheap hack like a "Don't replace
more than 50% of the servers in any one iteration" setup
would still be an improvement over ignoring TTLs entirely.)
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to