OPTIMAL-DESIGN OF A FINITE-BUFFER POLLING NETWORK WITH MIXED SERVICE DISCIPLINE AND GENERAL SERVICE ORDER SEQUENCE

Abstract

Polling schemes have been widely used in LANs, MANs, and distributed systems. The authors design a polling network (system) in which each station is finite-buffered and is served according to a mixed service discipline and a general service order sequence. They first analyse the network by means of an imbedded Markov chain approach to obtain probability generating functions (PGFs) for the number of customers in stations at four observation points: customer-service beginning, customer-service ending, server's arrival, and server's departure. Two performance measures of mean waiting time and blocking probability are then derived. Finally, the authors search for an optimal pattern of the mixed service discipline and service order sequence for a polling network via a genetic algorithm. The results show that a polling network with an optimal pattern of mixed service discipline and service order sequence offers a great improvement in performance over a polling network with a unique service discipline and cyclic service order sequence; a polling network has different optimal patterns for different traffic intensities; and a near-optimal pattern could exist for all traffic loads.