Learning Automata Models for Adaptive Flow Control in Packet-Switching Networks

Abstract

Performance and Stability results for three adaptive isarithmic flow control systems for packet-switching networks are described. The performance models, which are based on the BCMP theory for closed networks of queues, are exact under stationary conditions. The control architectures considered include one centralized and two decentralized schemes. The decentralized architectures include single-chain and multiple-chain cases. The controllers are modeled by L _R−I learning automata. Four types of network feedback responses were considered. These are loop permit delay, loop population, loop power and path delay, where a loop includes the controller, the source queue, the network path to the message destination node and the path back to the controller. The model has been verified by Monte Carlo event simulation, thus demonstrating the feasibility of the proposed control systems and the accuracy of the analytic performance model. The various control architectures and algorithms are compared in regard to their power performance, transient response and stability characteristics. Several areas for further research are then identified.