On-line resource sharing in communication networks using infinitesimal perturbation analysis of stochastic fluid models

In this paper we study the problem of resource allocation in communication networks, specifically bandwidth and buffer allocation. We adopt a stochastic fluid model (SFM) and derive sensitivity estimators of the performance measures of interest (average workload and packet loss volume) with respect of the control parameters (bandwidth share and buffer thresholds). These estimators are shown to be unbiased (with respect to the fluid-based measures), they require no knowledge of the underlying stochastic processes that drive the system (model free) and are very easy to implement.