[time] {0,1,2}.<zone>.pool.ntp.org; new monitoring system

wayne wayne
Wed Sep 8 02:23:54 UTC 2004


In <200409072034.26439 at fortytwo.ch> Adrian 'Dagurashibanipal' von Bidder <clockmaster at fortytwo.ch> writes:

> Monitoring system
> -----------------
>
> As announced (but, of course, a few days late), I finally retired the 
> monitoring system as it is now and replaced it by a much simpler one.  Now, 
> every server is just queried once per hour by a simple perl script (using 
> Net::NTP), and scores are assigned according to the offset measured to my 
> server:

I don't know if you are already planning on doing this, but....

If I recall correctly, the old monitoring system used data from
several sites from around the world to determine the quality of
service the various pool servers provide.  I think this is still an
important thing to do.  Right now, the monitoring system will be
biased toward sites in Europe due to lower network jitter and
latency.  (Does the Net::NTP code deal with the round trip times?)




>  +1             if offset < 0.5s
>  -2*$offset + 2 if 0.5s < offset < 1.5s
>  -2             if offset > 1.5s
>  -5             if server is unreachable (timeout 2s)

Personally, I think it would be best to cut these constants down by at
least a factor of 2, maybe even a factor of 10.  It really isn't that
hard to have an ntp server stay within 50ms of true UTC now a days.
I vaguely seem to recall that offsets of more than 250ms can cause ntp
servers to not work well.  It is assumed that such large offsets means
something is wrong and the local clock won't be updated.  (Yeah, I
should RTFM and check this, but I figure others might know better
anyway.)




Anyway, thanks again for all the work on the NTP pool project.  I
still think it is a real cool/useful idea.


-wayne



More information about the pool mailing list