Improving the performance of RED deployment on a class based queue with shared buffers

Abstract

We address the deployment of random early detection (RED) on class based queues (CBQs) with shared buffer by deriving novel RED algorithms for the complete sharing and sharing with minimum allocation schemes and comparing their performance with RED under complete partitioning that is equivalent to using a dedicated buffer for each CBQ. In our novel RED algorithms, packet drop decision is based on the estimated average lengths of both the effective subqueue and the whole queue. Simulation results ascertain and demonstrate the superiority of the proposed schemes in terms of lower packet drop ratio under congestion situations, with different RED parameters under diverse RED drop probabilities and weight factors.