Phase transitions and departure statistics of critically loaded queues: oscillating cumulants and generalized BRAVO

Queueing theory is used for modeling biological processes, traffic flows and many more real-life situations. Beyond that, queues describe systems out of equilibrium and can thus be considered as minimal models of non-equilibrium statistical mechanics. We demonstrate that non-equilibrium phase transitions of queues in the steady state are accompanied by a nontrivial flow of departing customers. Our analytical results show that the cumulants of the departure statistics deviate strongly from Poissonian values and oscillate in the vicinity of phase transitions, i.e., if a critical load is approached. The load-dependent oscillations of the cumulants generalize the BRAVO effect (Balancing Reduces Asymptotic Variance of Outputs) in queues and may occur in other boundary-driven non-equilibrium systems.