Re: [nm-wg] treatment of lost packets when measuring delay

[stanislav shalunov <shalunov@internet2.edu>]

Loukik Kudarimoti <loukik.kudarimoti@dante.org.uk> writes:

> If ten packets were sent between times t1 and t2 and 1 was *lost*
> (referred to as infinite delay), we report that *1 packet between
> times t1 and t2 was lost* and 9 packets have an average (arithmetic
> mean ) value v1, min value v2, max value v3 and 95 %ile v4 (extensible
> to include other types of aggregations as well).

Is this, in fact, statistically meaningful?  If so, what is the
estimator and its significance?  If the estimator is Sigma X/|X| where
X = {x in delays | x < clip}, then I am not aware of any special
significance such an estimator has.  It would certainly not be correct
to refer to the parameter estimated thusly as ``mean'' (except under
assumptions that do not hold for packet delays).  Such an estimator
could be meaningful for certain distributions (e.g., sharply bimodal
ones), but packet delay does not appear to be bimodal.

> A full report ( with no aggregations ) can also be provided. Now in
> such a report, whether we show packet loss as infinite delay or report
> it as packet loss still needs to be discussed.

Lost packet is simply one that hasn't arrived within a timeout, right?
So, it's known that its delay is larger than a certain value.  (It's
not known, after all, that the NSA does not record all packets that
traverse the Internet and does not have a plan to re-inject all lost
packets in 2100.)  There are estimators that can work with that.
Average is not one of them...

Just my 0.086g of Ag,

-- 
Stanislav Shalunov		http://www.internet2.edu/~shalunov/