Perturbation analysis for loss volume and buffer workload in tandem two-class stochastic fluid models for communication networks

This paper applies infinitesimal perturbation analysis (IPA) to packet loss and buffer workload relative performance metrics in tandem networks of two-class stochastic fluid models (SFM). It considers these performance metrics at a given SFM node processing a single-flow class, as function of the controlled traffic stream's threshold parameter at an upstream SFM node processing two class of incoming traffic. It further supplements the existing results concerning applying IPA methods to tandem SFMs. The derived IPA gradient estimators are simple and fast to compute, and have an appealing property that are unbiased and nonparametric in the sense that they can be evaluated directly online from measurements of real-life traffic processes as well as offline from simulation experiments, without any knowledge of underlying stochastic characteristic of the traffic and service processes. These properties hold the promise of utilizing these IPA gradient estimators to network control and optimization.