(no subject)

Summary of the paper

This paper presents Random Early Detection gateways for congestion
avoidance.

The main contribution of this paper is to introduce a new algorithm
in the gateways to do congestion avoidance by keeping the average
queue size low. The RED gateways are designed to accompany a
transport-layer congestion control protocol and it has no bias
against bursty traffic can avoids the global synchronization of many
connections decreasing their window at the same time.

In the introduction, paper describes the new algorithm generally. The
paper proposes a different congestion avoidance mechanism at the
gateway with somewhat different methods for detecting congestion and
for choosing which connections to notify of this congestion. RED
gateways can be useful in gateways with a range of packet-scheduling
and packet-dropping algorithm. The RED congestion control mechanisms
monitor the average queue size for each output queue, and using
randomization, choose connections to notify of the congestion. It can
work with current transport protocols.

In the second part, authors critics some previous research work,
especially the DECbit, because they are similar in the face.
According to the analysis in the paper, there are two main
differences. The first one concerns the method of computing the
average queue size. The problem of bias which happened in DECbit does
not rise in RED. The second difference concerns the method for
choosing connections to notify of congestion. DECbit networks can
exhibit a bias against bursty traffic because of the design flaw.
However, this is avoided in RED gateways.

Then comes the design principle including some design goals. The main
goal is to provide congestion avoidance by controlling the average
queue size. The goal of the gateway is to detect incipient congestion
that has persisted for a "long time", and decide which connections to
notify of congestion at the gateway. Another goal of the scheme is to
avoid a bias against bursty traffic, and decide which connections to
notify of congestion is to avoid the global synchronization that
results from notifying all connections to reduce their windows at the
same time.

The algorithm is to control the queue size and mark the packets for
congestion. If the average queue size is above the maximum size,
instead of dropping the packets, it will mark every packet. If
average queue is less than minimum size, no packets are marked. If
it's between the minimum and the maximum threshold, each packet is
marked with a certain probability which is calculated by the
algorithm. So, in the following parts, paper discusses how to
calculate the average queue length, how to calculate the packet-
marking probability.

Paper gives a evaluation of RED gateways from several aspects. The
result is that the RED can avoid congestion, have no global
synchronization, be simple, be maximizing global power, be fairness,
be appropriate for a wide range of environments and so on. Finally,
paper gives the testing data and analysis.

For the implementation, paper specifically indicate the bursty
traffic problem and how to identify misbehaving users. They could be
solved in the scheme.

Points in favour or against

The paper is generally well written, with fine and clear
presentation. Obviously, it's an improvement of the paper work "A
Binary Feedback Scheme for Congestion Avoidance in Computer
Networks". I think the whole scheme is not very new. It also realize
the congestion avoidance by controlling the queue size. The main
difference is that the RED uses the randomized method to mark the
packets instead of just dropping the packets when congestion happens.
This method could solve many existing problems, especially the bursty
problem and global synchronization problem. I think this scheme is
weak in the fairness, because according to the paper. RED gateway
does not discriminate against particular connections or classes of
connections. Moreover although it could identify connections using a
large share of the total bandwidth using the marking-probability
method, it could realize the fairness only cooperated by the
congestion avoidance algorithm in transport-layer. In another word,
RED only could indicate the misbehaving users, but the control and
avoidance work should be done in the transport-layer. I think this
scheme should consider this problem and refer to the research work of
"Analysis and Simulation of a Fair Queueing Algorithm".
Received on Wed Sep 27 2006 - 19:24:45 EDT