Derivative estimation and optimization of loss probability in satellite packet networks

In this paper we consider a resource allocation problem for a satellite network. Since thecapacity of the system is finite and divided in finite portions, the resource allocation problem reveals to be a discrete stochastic programming problem, which is typically NP-Hard. We propose a new approachbased on the minimization over a discrete constraint set using an estimation of the gradient, obtained through a 'relaxed continuous extension' of the performance measure. The computation of the gradientestimation is based on the Infinitesimal Perturbation Analysis (IPA) technique, applied on a Stochastic Fluid Model (SFM) of the network. Neither closed forms of the performance measures, nor additionalfeedbacks concerning the state of the system and very mild assumptions on the probabilistic properties about the statistical processes involved in the problem are requested.