Stochastic fluid models for control and optimization of systems with quality of service requirements

We use stochastic fluid models (SFM) for the control and optimization of communication networks in which detailed discrete event models become impractical. By analyzing a SFM for a threshold-based admission control problem on a single node, we derive gradient estimators which can subsequently be used on the actual system. It is shown that these gradient estimators are unbiased and independent of all traffic and service processes involved, including the traffic and service rates. Moreover, they enable one to develop simple optimization schemes that recover the optimal thresholds of the actual discrete event model