===================================================
From takhoa@rice.edu  Thu Feb 26 09:14:21 2004
===================================================

There have already been a lot of papers discussing the shortcoming of RED
and suggesting different improvements.  But this paper is the first paper
that invokes many subsequent research.  In hindsight, this paper offers an
idea that seems trivial enough.  However, it probably was not so intuitive
before it was invented.

The problem with this paper is that most of the design elements were
either brought up as a conceptual idea, or were the results of
simulations.  When conclusions are made using simulations, it is common
the case that the designers cannot cover all scenarios in their experiment
testbeds.  For example, RED fails to take into account for multiple flows
sharing the same bottle neck link as pointed out in the first paper.

This paper offers a new concept that proves to be very useful.  However,
its value does not extend beyond that.  Suggestions for the parameters
were based mostly on simulations using artificial trace files.  Thus, it
requires a lot more research to determine appropriate values for different
parameters for RED.

===================================================
From santa@rice.edu  Thu Feb 26 23:02:40 2004
===================================================

Random Early Detection Gateways for Congestion Avoidance

This paper presents Random Early Detection (RED) AQM technique for congetion
avoidance. Instead of the traditional drop-tail technique in routers, which
leads to bad utilization, large queuing delay and global synchronization, RED
senses congestion early. It uses queue lengths to predict impending congestion
and uses random probabilistic techniques to drop or mark(in presense of ECN)
packets to prevent congestion from setting in early.

I think this is a seminal paper which started the field of Active Queue
Management (AQM). There has been various more recent control theoritic
techniques proposed as well as improvements suggested to RED, to make AQM work
better. But, the fact remains that this simple technique of probabilistically
dropping packets even before congestion sets in, led the way. It is deployed in
all the current core internet routers. The evaluation of this paper though
leaves much to be desired with only 5 nodes sharing a bottleneck router.