Convergence of Recursive Optimization Algorithms using IPA Derivative Estimates

We propose two estimation and updating schemes which use infinitesimal perturbation analysis based derivative estimates for the recursive optimization of queues. With the aid of extensions of convergence theorems from stochastic approximation, we prove convergence of the two proposed algorithms, when applied to an M/G/1 queue and to a multi-queue system. We also present simulation results illustrating our theorems for the case of an M/M/1 queue, and a three-queue system.