[e2e] Can feedback be generated more fast in ECN?

[Vishal Misra]

On Wed, 14 Feb 2001, Qiong Li wrote:

> calculation results are then used as references to mark packets in the
> backward path in the same router. For obvious reasons the queues in the
> simulation are more stable than that produced by normal RED+ECN. The
> strong presumption of this scheme is that the traffic of both directions
> pass through the same node, which may not always be true in the Internet.
> 

There is another problem, and that is the heterogeneity of flows, not only
TCP/non-TCP but also the variations of TCP. The presumption here is I
guess that you see congestion on the forward (data) path and then mark
acks on the reverse path. However, the presence of TCPs which implement
delayed acks results in a non-proportional or "unfair" marking, and that
brings up the issue of fairness. In a more homogenous setting this would
work.

-Vishal