Best practices question

[Hal Murray]

> I do know that darned near  *any* activity I do on the command line of any of
> them tends to cause the  the graphs to become considerably more spikey.

Most likely, that's the temperature change on the crystal.  In some sense, if 
that is a significant problem, that means you have a reasonably good setup.

If you want another rathole to go down, measure the temperature of the crystal 
and plot NTP's drift vs temperature.

You want something that reads to a small fraction of a degree.  I don't know 
of a great setup to recommend.  The standard CPU or disk temperature reading 
is usually only good to 1C.  That's great for monitoring CPU/disk temperature, 
but not good enough for making great plots.

You don't need accuracy as long as it is consistent and reasonably linear.

--------

> I do  compiling of ntpsec, gpsd, etc on the separate ntpviz server, then copy
>  the 'built' environment across. Still gives a spike for the 'install'  part,
> but only a blip. 

Or do a compile or some artificial load to deliberately introduce a spike in 
the temperature and see what you can learn from the graphs.

--------

> So I can run NTPsec without GPSD? For some reason I thought it was  required
> to interpret the PPS from the GPS. I'll do some further reading.
> Interesting. 

Eric and Gary came from GPSD.  They think it is the solution to all problems.  
My background is ntp-classic.  I try to avoid GPSD.  It's different.  
Sometimes better, sometimes not, sometimes just different.

--------

I forgot to mention...

Another variable to play with is the polling interval - minpoll/maxpoll.  This 
isn't simple.  You want to see what it does in both the calm/stable "normal" 
conditions as well as how well it works when interesting things happen.



-- 
These are my opinions.  I hate spam.