Binary Feedback Scheme For Congestion Avoidance

Paper Review: A Binary Feedback Scheme for Congestion Avoidance in
Computer Networks

The paper proposes a congestion avoidance scheme for networks using
connectionless protocols at the network layer. The combination of
network growth and lack of network control can lead to significant
congestion loss, considered a ‘social’ problem. The authors
distinguish between congestion control, which attempts to reduce load
after entering a congested state, and congestion avoidance, which
attempts to push the network to the optimal operating point without
overloading the network.

The scheme dictates that a router that is congested sets a congestion
bit in the network layer header of a data packet that is flowing in
the forward direction. The congestion bit is then copied into the
transport layer header of the acknowledgement packet headed to the
source. The scheme includes a router policy and user policy which
outlines the actions to be taken on setting and acting on the
congestion bit. When facing congestion, there is a multiplicative
decrease in the window size and an additive increase during favourable
conditions, a method shown to converge to both fair and optimal
network usage.

The scheme has several goals:
- maintain the overall global window size close to the maximally
efficient value
- maintain fairness across multiple sources
- minimize oscillations in the window size
- minimize the time to achieve steady state

The paper shows through analysis and simulation the efficient
operation of their scheme. Tests were conducted for several
situations. Using random packet size distributions, the dynamic
behaviour of the window size was very close to optimal. Under
transient conditions, the window sizes converged quickly to the new
optimal point. When users started with arbitrarily large window
sizes, they were eventually brought down to the fair value. In
networks where the source is the bound, the window size was not
increased unnecessarily. Overall, the simulations showed that the
scheme was stable even in an overloaded network. Although simulations
show that the scheme should work in practice, some real-world tests
would further support the work.

The strongpoint of this work is that it seems to be an effective
congestion avoidance scheme which also provides fairness without the
use of extra packets. The downfall, however, is that it would require
modifying routers to be able to set the congestion bit. As such, this
seems inapplicable for use in the Internet as it is distributed,
wide-spread and without a central authority.

-- Nadeem Abji
Received on Tue Sep 26 2006 - 02:09:45 EDT