IPA-based Tuning of Queue Admission Control under Imperfect Information 1

The Infinitesimal Perturbation Analysis (IPA) is applied to the queue admission con-trol problem for a system modeled by tandem queues. The control takes the form of holding a customer in a controller buffer with a low holding cost until admitting that customer to the main queueing system with a high holding cost. An open-loop dynamic admission control policy simple to implement is proposed for a control of the queueing system. The basic idea of this policy is to ensure minimal interadmission time, 0, in order to prevent congestion in the main queueing system; 0 is left as a design parameter. A simple gradient estimate of the total steady state holding cost with respect to 0 is derived using Infinitesimal Perturbation Analysis techniques. A stochastic gradient-like algorithm based on this estimate is studied for optimizing 0, and convergence is proven under certain conditions. Key words: perturbation analysis, stochastic gradient-like algorithm, open-loop ad-mission control, minimal interadmission tim