The RED Algorithm – Averaged Queue Weight Modeling for Non Linear Traffic

Abstract: This thesis presents an approach in developing a congestion avoidance algorithm in computer networks for router-based communication. In internet communications, the requests generated are transferred through network nodes called routers which routes information, from one node to another, depending upon the request made. As the traffic load increases in the communication links, the routers must be designed to provide fair traffic flows, from a source to a destination, for all network nodes. During heavy traffic conditions the routers may get congested and the traffic flow through such a router degrades rapidly resulting in heavy packet dropping. Furthermore, this fact may also lead to a complete traffic collapse. To achieve fair transportation through such routers, the routers must be designed and provisioned with advanced congestion avoidance algorithms in order to achieve good performance. Congestion avoidance algorithms were used in the past, e.g. Active Queue Management (AQM), or the Drop Tail (DT) algorithm. To reduce the congestion effect, a new congestion avoidance algorithm named Random Early Detection (RED) was suggested. In RED, the calculation of packet dropping probabilities uses a queue weight factor. When compared to the previous RED approach, a fixed value is assigned leading to constant congestion reduction and if the network is varied randomly this fixed value may result in over congestion. To avoid this, an average queue weight parameter is developed. This algorithm is evaluated on router architecture for its practical feasibility and this mechanism is evaluated for various quality metrics such as throughput, network overhead, congestion level, transportation delay, etc. Due to its high cost in implementation on the network side, the proposed RED algorithm is evaluated through simulations and the obtained results are used to illustrate the performance of RED-DT by using MATLAB version 7.4.