Paper Review : Congestion Avoidance and Control
Reviewer : Seh Leng Lim
The paper aims to provide some practical algorithmic solutions to solve network congestion problems, that can be implemented easily in the transport protocol. This is useful in the light of the explosive growth of computer networks, and the consequent worsening of network congestion problems.
The main contribution of the paper is its elaboration of the round-trip-time variance estimation, exponential retransmit timer backoff, slow-start, more aggressive receiver ack policy and dynamic window sizing on congestion algorithms. These algorithms are based on the principle of ‘conservation of packets’ whereby if this principle were obeyed, congestion collapse would become the exception rather than the rule.
The key main idea behind the algorithms is that packet conservation fails when :
(a) The connection does not get to equilibrium, or
(b) A sender injects a new packet before an old packet has exited, or
(c) The equilibrium cannot be reached because of resource limits along the path
To ensure the connection gets to equilibrium, the slow start algorithm is proposed. The slow start algorithm manipulates the congestion window to gradually increase the amount of data-in-transit until equilibrium is reached.
To avoid problem (b), the round–trip-time variance estimation algorithm is proposed. This algorithm helps to eliminate spurious retransmissions by using a low pass filter of the following form :
R <- aR + (1-a)M
To further stabilize a network prone to random load shocks and congestive collapse, an exponential timer backoff algorithm is suggested, but not elaborated in this paper.
To avoid congestion and solve problem ©, an additive increase/multiplicative decrease policy is suggested but not elaborated. In fact, the paper states without justification that the best increase policy is to make small, constant changes to the window size.
I think that the paper has a significant contribution (rating of 4) to the study of how to avoid and control congestion in the computer networks. The paper does present convincing and impressive experimental results. However, it does not go into the specifics of the design of the algorithms, especially as to the choice of the parameters and equations. The reader has to take it that it is correct based on the experimental data. Also, I am not able to find any clue about the aggressive receiver ack policy which the paper states that it will present briefly. I am not sure if the paper is summarized from some longer paper and written in a rush. Somehow, it just lets you have a feel about the work of the authors, but is not enough for you to be able to work the details out.
The authors have stated that this work needs to be continued into the gateway with the gateway ‘congestion detection’ algorithm as the next step. This is because congestion avoidance algorithms at the transport endpoints cannot help to insure fair sharing of network capacity, although it can help to ensure that the network capacity is not exceeded. Only at the gateways at the convergence of network paths with sufficient control information can network sharing and fair allocation be achieved.
Researchers and builders who build Internet and distributed applications will have a better feel of the network congestion problem from this paper.