A New Paradigm for On-line Management of Communication Networks with Multiplicative Feedback Control

We describe the use of Stochastic Flow Models (SFMs) for control and optimization (rather than performance analysis) of computer networks. After reviewing earlier work applying Infinitesimal Perturbation Analysis (IPA) to SFMs without feedback or with additive feedback, we consider systems operating with a multiplicative feedback control mechanism. Using IPA, we derive gradient estimators for loss and throughput related performance metrics with respect to a feedback gain parameter and show their unbiasedness. We also illustrate the use of these estimators in network control by combining them with standard gradient–based stochastic approximation schemes and providing several simulation-based examples.