[ntp:questions] wedge scattergram questions
David L. Mills
mills at udel.edu
Wed Jun 6 02:57:20 UTC 2007
You should read the book a little more carefully. Useful additional
information is in the briefings on the NTP project page linked from
Werner Siedenburg wrote:
> Hello! In Mr. Mills' book I saw some interesting "wedge scattergrams".
> I have some questions about them:
> 1. What is the exact use of these scattergrams?
> Mr. Mills writes that "the shape of the wedge reveals intimate details about
> the network path characteristics". How? The only thing I notice: The further
> left the
> apex of the scattergram is, the better (because of lower delay). What else
> can you see
> in them?
If you look at the scattergrams in Chapter 6, you may notice many subtle
things. In Figure 5.14 the scatters are concentrated on the upper limb
line. What does that tell you about the traffic characteristic on the
outbound and inbound paths? Even more dramtic is Figure 3.12.
Interessting, the evidence of a double apex in that figure suggests a
path change during the measurement. In Figure 6.10 the scatters are all
over the place. Can you think what might cause that? In Figure 6.8 there
are no apparent limb lines, but the apex is not clearly defined. Looks
like path has very small jitter, but the clocks show errors over time.
That would be frequency wander.
> 2. I understand that "under low traffic conditions, the points are
> concentrated about the apex of the wedge and begin to extend rightward along
> the limb lines as
> network traffic increases".
> But why is the most accurate offset measured at the lowest delay?
> The only reason I can think of is that the bigger the delay, the bigger the
> difference between both clocks. But I doubt this plays a role at "normal"
See the clock filter algorithm described in Section 3.5.
> 3. How can I create my own wedge scattergrams? I configured my ntpd to
> generate rawstats. Do you just calculate the offset 1/2*[(T2-T1)+(T3-T4)]
> and delay
> (T4-T1)-(T3-T2) for every line in the rawstats-file? and afterwards plot the
> whole thing where delay
> is the x-axis and offset the y-axis?
> Is there a script (e.g. MATLAB .m-file) that does this automatically?
> I suppose, a wedge scattergram only show delay/offset-combinations from ONE
> specific server.
The rawstats data produced by the filgen facility records the T1, T2, T3
and T4 timestamps. Use Equations 3.3 to calculate the scatter point.
> Thanks in advance,
> questions mailing list
> questions at lists.ntp.isc.org
More information about the questions