A Hybrid Random Early Detection Algorithm for Improving End-to-End Congestion Control in TCP/IP Networks

The successful operation of the present Internet depends mainly upon TCP/IP which employs end-to-end congestion control mechanisms built in the end hosts. In order to further enhance this paradigm of end-to-end control the Random Early Detection algorithm (RED) has been proposed, which starts to mark or drop packets at the onset of congestion. The paper addresses issues related to the choice of queue length indication parameters for packet marking/dropping decisions in RED-type algorithms under varying traffic conditions. Two modifications to RED are proposed: (i) use of both instantaneous queue size and its Exponential Weighted Moving Average (EWMA) for packet marking/dropping and (ii) reducing the effect of the EWMA queue size value when the queue size is less than $min_{th}$ for a certain number of consecutive packet arrivals. The newly developed Hybrid RED algorithm can effectively improve the performance of TCP/IP based networks while working in a control loop formed by either dropping or marking of packets during congestion epochs. New guidelines are developed for better marking/dropping of packets to achieve a faster response of RED-type algorithms. The hybrid RED algorithm has been tested using ns-2 simulations, which show better utilization of network bandwidth and a lower packet loss rate.