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

["Jeff W. Boote" <boote@internet2.edu>]

Augusto Ciuffoletti wrote:
> My point about the memo is that losses should be addressed separately, as in
> the current version. I also found appropriate the presence of the "Loss
> Patterns (Statistical Properties)" sub-section.

As you point out, losses need to be addressed separately and losses are
addressed separately in the document.

However, I believe the question is really: Can you reasonably discuss
delay by ignoring the portion of the tail that is beyond the loss
threshold?

As long as the loss threshold value is available as an additionally
parameter of the data, it is trivial for a data consumer to use it to
filter the results in the way that has been described. (If some
consumers find this representation useful.) However, the converse is not
true. If the larger than loss threshold delays are not included as part
of the distribution, it is impossible to reconstruct a representation of
the complete distribution.

jeff