Polling system optimization through dynamic routing policies

Polling systems have been used as a central model for modeling various communications networks, like the token ring network, some high speed local-area networks and others. In this paper we focus on pseudo cyclic algorithms which can prioritize the various stations by dynamically changing their service order but maintain fairness by visiting each station exactly once in a cycle. We study the waiting time performance of systems operated under cycle-time guided algorithms in semi-dynamic polling models. In fully symmetric systems we bound by stochastic dominance the waiting times of all pseudo-cyclic policies. This dominance is quantified by the analysis of a fluid approximation model and simulations of the general polling model.