Hmm, ok, this seems right (though it's not exactly what i had in mind).
> Or are you suggesting saving that data even if its incomplete in both
> directions? I seem to remember Jack saying that this last one was a bad
> idea.
Right, using partial traceroute results seems like a bad idea, since we'd
be lying about the last few hops if we claimed anything.
But what i was trying to say was: (please ignore me if you were already
thinking this) I was thinking that the ngon needs t be saved even if some
of the bidirectional paths are entirely missing. The Ngon still seems
useful as long as you have the primary path (A->B) and at least one other
path (A->X->..Y->B). Because it seems like you're trying to establish
something like, "Of 700 primary paths and 12,000 2-minute measurement
windows, we were able to find at least one path that was at least 20%
better in at least 8,000 of those windows." And discarding an Ngon because
it is missing an edge you wanted to measure seems like it's going to give
you an unrepresentative sample - biased results - because you're deciding
whether or not you include a sample based on the results from that sample.
Throwing away Ngons with missing legs would be like taking a survey of the
health of every person in each of 12,000 households, and then discarding
the results for households where a family member died of the plague while
you were there taking the survey. (very Monty Python-esque, i guess ;-)
Clearly if one did this, one would overestimate the general health of
households and people, because households with people dying of the plague
tend to be less healthy, and those were thrown out...
neal