Title:Diffusion Approximations for $\text{Cox}/G_t/\infty$ Queues In A Fast Oscillatory Random Environment

Abstract:We study infinite server queues driven by Cox processes in a fast oscillatory random environment. While exact performance analysis is difficult, we establish diffusion approximations to the (re-scaled) number-in-system process by proving functional central limit theorems (FCLTs) using a stochastic homogenization framework. This framework permits the establishment of quenched and annealed limits in a unified manner. At the quantitative level, we identity two parameter regimes, termed subcritical and supercritical indicating the relative dominance between the two underlying stochasticities driving our system: the randomness in the arrival intensity and that in the serivce times. We show that while quenched FCLTs can only be established in the subcritical regime, annealed FCLTs can be proved in both cases. Furthermore, the limiting diffusions in the annealed FCLTs display qualitatively different diffusivity properties in the two regimes, even though the stochastic primitives are identical. In particular, when the service time distribution is heavy-tailed, the diffusion is sub- and super-diffusive in the sub- and super-critical cases. The results illustrate intricate interactions between the underlying driving forces of our system.