ABR traffic management using minimal resource allocation (neural) networks

This paper presents an adaptive available bit rate (ABR) traffic management scheme in asynchronous transfer mode (ATM) networks using the newly develo ped minimal resource allocation network (MRAN). MRAN generates a minimal ra dial basis function (RBF) neural network by adding and pruning the hidden n eurons based on the input data and is well suited for on-line adaptive cont rol of time varying nonlinear systems. In this paper, the ATM traffic is mo deled using the network simulation package OPNET. The performance of MRAN-c ontroller is compared with the conventional ABR control scheme explicit rat e indication with congestion avoidance (ERICA) for different traffic scenar ios such as bursty and Variable Bit Rate (VBR) traffic. Results indicate th at MRAN-controller performs better than ERICA by keeping the queue length a nd delay to a minimum while maintaining a higher Link utilization and throu ghput. (C) 2002 Elsevier Science B.V. All rights reserved.