Anssi Johansson wrote:
Rob Janssen kirjoitti:
Anyway, there is an error in this monitoring. The monitored time always has a
negative offset.
I have never seen anything different, no matter how correctly tuned my time
server.
(it should now be within .1 ms most of the time, but the monitoring plot cloud
is always
between -10 and -3 ms)
I would say there is a systematic error of about 4 or 5 ms.
The reason for this is most probably asymmetric routing -- the packets TO your
server are routed through some other route than the packets FROM your server.
This is perfectly normal, nothing to worry about.
If you have a look at my graphs at http://www.pool.ntp.org/user/avij , you'll notice that most them don't look perfect either. I've long ago stopped caring too much about the graphs. Those servers still provide reasonably accurate time to the pool
users, despite the funny-looking graphs.
My point is that for any server that I tried I always saw a bias towards
negative error, so there
probably is some bias inside or close to the monitoring server. Some of your
servers confirm
this, but you have two that don't. Curious.
I do not worry about it too much, but I notice it is a frequently discussed
topic, so when there is
a simple way to fix it it would save time to everyone if it were fixed.
Rob
_______________________________________________
pool mailing list
[email protected]
http://lists.ntp.org/listinfo/pool
