RE: [ns] goodput calculation

[Michael Welzl <michael@tk.uni-linz.ac.at>]

> > Anyway, having read that, I was convinced that I should stop fooling
> > around with bound variables and just use the trace-all source file.
> > What is wrong about it??? After all, you see every packet 
> that arrives
> > at a TCP sink. Isn't this goodput?
> 
> No, it isn't. (sigh.) It's just network capacity use. Program your
> TCP sender to send out every packet twice, and hey, you could
> double your 'goodput' -- but in reality, good put won't change at all.

But this is an upper layer problem. Like, if you do FEC (presumably on
top of UDP instead of TCP in real life - why would anybody want to do that
on top of TCP?), your "goodput" should be how much you transferred but
without the additional bits you needed for FEC. Is this what we're
talking about?

Then, if I'm measuring how much a regular TCP agent gets across, I'm
doing the right thing   :)

Cheers,
Michael