A learning-automata-based controller for client/server systems

Polling policies have been introduced to simplify the accessing process in client/server systems by a centralized control access scheme. This paper considers a client/server model which employs a polling policy as its access strategy. We propose a learning-automata-based approach for polling in order to improve the throughput-delay performance of the system. Each client has an associated queue and the server performs selective polling such that the next client to be served is identi8ed by a learning automaton. The learning automaton updates each client’s choice probability according to the feedback information. Under the considered approach, a client’s choice probability asymptotically tends to be proportional to the probability that this client is ready. Simulation results have shown that the proposed polling policy is bene8cial in comparison to the conventional round-robin polling when operating under bursty tra;c conditions. The bene8ts are signi8cant for the delay reduction in the considered client/server system. c © 2003 Elsevier B.V. All rights reserved.