[e2e] Explicit Transport Error Notification

[Black_David at emc.com]

> > > Explicit error notification is just one way to distinguish
> > > between losses due to transmission errors and losses due to
> > > congestion. An alternative way is to use explicit congestion
> > > notification, and assume that packet losses are due to other
> > > factors -- e.g., only shrink the window on ECN, and expend it
> > > on the ACK. There are obvious deployment issues, but these
> > > could be solvved if NASA is operating in any kind of
> > > controlled set-up. In any case, the ECN only option could be
> > > simulated in NS...
> > 
> > I presume making sure that network nodes never have to drop
> > packets due to queue overrun/resource shortage/etc. is among
> > the "obvious deployment issues" ... ECN currently requires that
> > drops be treated as congestion signals for this reason.
> 
> No, that specific problem ought to not be too bad, as long that a node
> that is experiencing queue overrun/resource shortage/etc. still ECN
> marks those packets that are actually going through. Remember that ECN
> is a statistical process; it can work even if some of the signal is
> lost.

One also has to beware of overshooting on signal strength.  If
every packet that is actually going through is marked, the result is
undesirable flow synchronization/oscillation.  As long as something
like RED is being used to do the marking even when drops occur,
excessive marking shouldn't be a problem, but panicing and marking
everything when drops start to occur may not be a good idea.

> In any case, looking at the NS simulation could be interesting.

Definitely.  --David