Achim Gratz Stromeko at nexgo.de
Mon Nov 7 19:01:46 UTC 2016

Gary E. Miller writes:
> Hard to control the power dissipation on a PC serving the internet.

We weren't talking about a PC, or at least I was not.  The only subject
I'm discussing is a rasPi and more specifically the 2B, which has
multiple cores to play with.

>> 99% of all
>> PPS timestamps over the last day stay within ±10µs, 75% within ±1µs
>> and slightly less than half within ±500ns.
> Odd that I get much better numbers without all that work:

Read again what I wrote, please.  This data was for 86400 raw PPS
timestamps (which you are seemingly not recording), you are talking the
loopstats.  Not surprisingly, the loopstat data as filtered through NTP
and has a much tighter distribution.  So in fact I do get better numbers
than yours at least at the tail ends of the distribution.  Over the last
48 hours (the distribution is slightly skewed with the median around
-10ns and the average around 30ns):

 1%…99%	-600ns … +850ns = 1350ns (1200ns no skew)
 5%…95% -375ns … +550ns =  925ns ( 750ns no skew)
10%…90% -290ns … +400ns =  690ns ( 580ns no skew)
25%…75% -160ns … +170ns =  330ns

The skew seems to originate from the aging transient (especially the
initial part, see below), assuming it gets fully removed as indicated by
looking at the last few hours only it should converge to better than the
numbers shown in parenthesis.  Here's the last five days on that system,
including some configuration changes:

-------------- next part --------------
A non-text attachment was scrubbed...
Name: lstats.png
Type: image/png
Size: 63220 bytes
Desc: not available
URL: <http://lists.ntpsec.org/pipermail/devel/attachments/20161107/28529c6c/attachment.png>
-------------- next part --------------

The x axis is in hours, the ppm scale is 1.5ppm top to bottom on the
(unlabeled) y2 axis.  The first two days shown was an intermediate
configuration (case closed, a single core loaded).  Then comes a day of
running with two cores loaded, then the system getting wrapped up and
boxed and the load switched back to a single core.  The long-term
downward trend you see on the ppm is the aging transient of the crystal,
the temperature (not shown in the plot) has been stable over the last
two days.  A week out this will look like a linear trend and a month out
you'll not be able to see it on such a short timescale.  It will start
anew however if you do such a wild temperature swing on that crystal

+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Samples for the Waldorf Blofeld:

More information about the devel mailing list