[ippm] Comments on draft-morton-ippm-delay-var-as-03.txt

General:
(a). It may be useful to incorporate a reference to ITU-T G.1050.  This is
an IP impairment model that is somewhat oriented toward modeling congestion
over limited bandwidth links, providing a useful tool for experimenting with
and testing PDV measurement techniques.  G.1050 is based on a time series
model that was originally developed as a potential PDV measurement
algorithm.

(b) ITU-T COM12-D98 "Analysis, measurement and modelling of jitter" (2003)
provided some useful background on the sources and measurement of jitter.
This also pointed out some significant characteristics of the PPDV
performance metrics that affect reported values with particular patterns of
delay (e.g. delay ramps, alternate pairs of early late packets......)

3.4 Service Level Comparison
IMHO - percentile based approaches are particularly suitable for SLA
agreements.  It is both easy to specify and easy to implement an SLA
measurement of the type "no more than x% of packets shall arrive later than
y milliseconds".  This is similar to (within the scope of?) Y.1540 however
reports the percentage of packets outside some given threshold rather than
the threshold that corresponds to some given percentile.

3.5 Separation of scheduling/ smoothing PDV from PDV introduced by the
network. 
Some delay variation may be introduced by packet scheduling for smoothing
purposes or the queuing of packets within sending systems.  Smoothing could
potentially be introduced by shaping devices in the network.  If packet
spacing/ size is uniform then this smoothing or shaping would not
necessarily result in PDV however for any system that uses irregular or
bursty packet transmission smoothing/ shaping is likely to result in PDV
that could in some cases significantly exceed the PDV due to network
congestion.  From the perspective of network performance monitoring it is
very useful to separate the two components of smoothing/ shaping PDV (which
would result in strong correlation between delay variation and the bandwidth
of the test signal) and PDV due to other sources (which "should" be weakly
correlated or uncorrelated with the current bandwidth of the test signal).

Regards

Alan