STUDY OF TRAFFIC CONTROL IN ATM NETWORKS THROUGH STOCHASTIC ESTIMATOR LEARNING ALGORITHM

Abstract

The statistical multiplexing of sources with diverge traffic characteristics in ATM networks leads to serious congestion . One of the most fundamental controlling mechanism is policing of these sources ,particularly those with bursty traffic. Because of the statistical nature of bursty traffic, policing of these sources is difficult and the known policing mechanisms such as Queuing Management ,Jumping Window and Leaky Bucket cannot control them effectively. In this dissertation, a new approach Stochastic Estimator Learning Algorithm based Leaky Bucket has been simulated, which learns the behaviour of the source effectively. Through simulation LB and LB-SELA approaches have been compared with respect to deviation from negotiated traffic parameters. The results confirm that LB-SELA has tighter and faster control of source, saves bandwidth and offers better guarantee of the QoS constraints. The effective control of source by LB-SELA results in fewer cell losses and smaller delays in an internodal node where many sources share the same buffer. 111