Summary

The paper proposes a binary feedback scheme congestion avoidance scheme
using a connectionless network protocol at the network layer. The paper
clearly differentiates between a congestion control algorithm and a
congestion avoidance algorithm.

The main idea of the paper is to use a congestion bit in the network
header at the end points to detect congestion. The congestion bit is set
by intermediate routes along the path and it will be communicated to the
sender by transport acknowledgment.

The feedback control system has two sets of policies for controlling
traffic over the network: Router Policy and User Policy.
1)Router Policy
    a.Congestion Detection
    b.Feedback Filter
2)User Policy
    a.Decision Frequency
    b.Use of Received Information
    c.Decision function

The paper also emphasizes the importance of filtering at routers and end
points due to noise and other side effects. This introduces a notion of
feedback filtering and signal filtering. The later parts of the paper show
how fairness is achieved by hitting knee of the curve. They also study
performance of the congestion avoidance algorithm under heavy loads and
transient conditions.

Overall, ideas presented in the paper is very convincing. At the same
time, they compare the performance of their ideas against other ideas to
further support their argument. The problem is that it requires changes to
the transport protocol. In the next generation networks, where alternative
paths gain more importance, I think we need a better scheme to detect
congestion because it binary scheme just tells if any of the router along
the path is congested. For e.g. the sender/receiver has no way of
inferring how many routers along the path were congested with binary
feedback scheme.
Received on Wed Sep 27 2006 - 00:02:14 EDT