Contents 1. Introduction
With the traditional approach to buffer management, named Tail Drop (TD), arriving packets will not be dropped until the queue space is exhausted [1]. When the queue is full and tail drop is in effect, packets are dropped until the congestion is eliminated and some space becomes available in the queue. Such a method can easily cause two problems: “Lock-Out” and “Full Queues” [1]. These problems can lead to high queueing delay. A widespread deployment of Active Queue Management (AQM) in routers, which starts dropping packets before the queue becomes full in order to notify incipient stages of congestion, has been recommended in the IETF publications [2] for overcoming these drawbacks of TD. AQM with multiple thresholds has been selected as an effective mechanism of buffer allocation for differentiated QoS of heterogeneous traffic. Two critical problems for buffer and queue management are when and how to drop packets arriving at a queueing system. In general, the former is mainly based on the queue length and the given threshold. The latter is based on dropping function used to drop packets. Both have significant impacts on the average delay, system throughput and probability of packet loss.
