(no subject)

Review: Random Early Detection Gateways for Congestion Avoidance

Reviewer: Di Niu

This paper proposes random early detection (RED) gateways for
congestion avoidance in packet-switched networks. Rather than
inferring congestion at end hosts, the paper maintains that the most
effective detection of congestion can occur in the gateway itself. In
RED, gateways detect congestion according to the average queue size.
They could notify connection of congestion by dropping packets
arriving at them or by setting a bit in packet headers. When the
average queue size exceeds a certain threshold, the gateway drops or
marks each arriving packet with a certain probability. In the
meantime, RED can allow bursts of packets in the queue which are
frequent in applications such as variable-bit streaming. It also
avoids the problem of global synchronization.

To let gateways detect congestion is a questionable method. Given
that end to end congestion avoidance through adjusting window sizes
based on binary or other simple feedback is somehow effective, it is
questionable whether gateway reports are necessary.

Despite of this fact, the algorithms in RED has at great experimental
values. By using a randomized scheme, a gateway could control the
flow of each connection according to the traffic of that connection
currently passing the gateway. A probabilistic model also helps RED
to avoid global synchronization. Another important feature RED is
that it could adapt to bursts of traffic, which is a big improvement
upon Drop Tail or Random Drop gateways.
Received on Wed Sep 27 2006 - 22:42:34 EDT