Transient analysis of RED queues: A quantitative analysis of buffer-occupancy fluctuations and relevant time-scales

This paper presents an analytic model for the transient behavior of the buffer occupancy in Random Early Detection (RED) queues under temporarily constant input traffic. The model shows that depending on the parameter choice, two types of qualitatively different behavior can be observed: (1) attenuated oscillation and (2) exponential convergence. The conditions for the occurrence of this behavior are derived as well as quantitative parameters in particular relaxation times, that allow to quantify the time-scales on which that transient behavior is observed. The model is subsequently applied to investigate an example scenario based on RED queues in a DiffServ QoS concept in which traffic changes are caused by mobility events.