This paper introduces the design of RED (random early detection) gateways. It is used for congestion avoidence by marking or dropping packets according to the queue length. In the paper, the detailed algorithm for calculating the probability for marking each coming packet, the bahavior with the different queue size related to max and min threshold. Some simulation results are also given to show the performance of this kind of gateway.

The strength of this paper:
1. An novel idea is mentioned in this paper. It gives a possible solution to handle the congestion problem in the net by doing congestion aviodance.

2. Both the detailed algorithm and simulation results are given in this paper. So we can get a clear view of this method and we can tell the effect of this algortim from the experiment results.

the weakness or improvement of this paper or the algorithm:
1. In this paper, the connection in the gateway usually means the physical connection. So for each connection, when the queue length of a certain connection is above some threshold, every packet is marked or dropped, no matter what these packets are or where they come from. So I wonder if we can base the connection on TCP connections to decide the dropping probability, not only on the physical connection. This may cause overhead in the gateway for large volume of traffic. But different TCP connections may have different properties, if we can take some of these properties into account, we may make more presice decisions. For example, some TCP connections may be more important or more time limited, so we give more weight for them to decide if we drop them.

2. There are several parameters in this RED gateway. How to decide the value of them is a problem. In the paper some experiments are done to test the different performance effects. But no good way are given how to decide them in this paper. And if these values are static or dynamic, we can not tell. Intuitively, if these values can be decided dynamically with some algorithm, the network can be utilized better. But how to do that?

3. There is no way to justify the simulation. If the results are accurate? or if the simulation is correctly done? We can not tell from this paper. This is also a general question for many research papers. So maybe a better way is to use several simulation tools to do the simulation, if we can get similar results, we can be more confident about the result.

4. Still the RED is based on local information to decide whether to drop a packet. usually from one node to another, there are possiblly several gateways in the middle, it may be good if these gateways can have information exchange to have kind of global decisions. For example, the first gateway may notify the following gateway that certain connections have been satisfied so no further action are needed for them.

5. Where to put this kind of gateways?put them everywhere? and how to solve the security problem, for example, if some faked gateway drops packets on purpose, how can we know that? All these are questions needed to be answered.

Generally speaking, this is a good paper. It gives us a good view of the new and promising idea of congestion avoidance. Still some work need to be done for this paper.