Re: [time] new feature to test: only 1Mbit+ servers in the global pool

Tue, 07 Feb 2006 11:40:41 -0800


On Feb 7, 2006, at 12:39 PM, Karel Sandler wrote:
In <[EMAIL PROTECTED]> Ask Bjørn Hansen <[EMAIL PROTECTED]> writes:

I have thought of a few others things I'll try to improve the quality
of the zones typically used by themselves.  (And eventually if we
ever get "enough servers" the pool in general).  One of them is just
to make a "long term score" where you build the score much more
slowly and lose it much faster to filter out servers that are not
Very Stable.

A short look on the offset history page is largely not very pleasant. The stability of a server and its stability as seen by one monitoring server may be quite different. It is generally difficult to say where on the network all these glitches come up. Certain server can be good for some subset of global ISPs and not for another one. Of course frequent service breaks are something else.
--
Karel Sandler

I've noticed that I stay very consistent in my offset, yet I see giant offset spikes in the chart, and noticed at one point more than 10 days back that about 1/2 the servers in the pool got dinged for offset in one particular check. My curiosity is how a server with peers like those below is seen by the mon server with spiking offsets. Nearly all of the pool suffers this. Of course, giant is still < 100ms, usually topping out at ~50ms

     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
*130.207.244.240 .USNO.           1 u  505 1024  377    2.898   -0.230   0.152
+128.227.205.3   .USNO.           1 u  544 1024  377   13.190   -0.057   0.182
+129.6.15.28     .ACTS.           1 u  747 1024  377   21.894   -1.333   1.582


That said, I'd consider from the graphs that all the below servers are "unfit," yet still included in the pool.
I propose the following long-term exclusions:

Average score < 15
Std. dev. of offset > .1
Std. dev. of step > 1
                                                OFF STP SCR

Rutgers has network issues every weekeday daytime ;)
Note that Rutgers got to stay in by my logic.

Of course, this is a time-sensitive e-mail.... these values will change. But if you look at the graphs today, you can see why I think each of these are candidates for exclusion. I think my selected value of 1 for the Std Dev of Step is a poor one... perhaps .5? .5 would exclude everybody but Rutgers. Perhaps score & offset would be sufficient.

Thoughts?

-Jeff
SIG: HUP
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to