Paper detail

Diffusion approximation for efficiency-driven queues: A space-time scaling approach

Motivated by call center practice, we propose a tractable model for $\mbox{GI}/\mbox{GI}/n+\mbox{GI}$ queues in the efficiency-driven (ED) regime. We use a one-dimensional diffusion process to approximate the virtual waiting time process that is scaled in both space and time, with the number of servers and the mean patience time as the respective scaling factors. Using this diffusion model, we obtain the steady-state distributions of virtual waiting time and queue length, which in turn yield simple formulas for performance measures such as the service level and the effective abandonment fraction. These formulas are generally accurate when the mean patience time is several times longer than the mean service time and the patience time distribution does not change rapidly around the mean virtual waiting time. For practical purposes, these formulas outperform existing results that rely on the exponential service time assumption. To justify the diffusion model, we formulate an asymptotic framework by considering a sequence of queues, in which both the number of servers and the mean patience time go to infinity. We prove that the space-time scaled virtual waiting time process converges in distribution to the one-dimensional diffusion process. A fundamental result for proving the diffusion limit is a functional central limit theorem (FCLT) for the superposition of renewal processes. We prove that the superposition of many independent, identically distributed stationary renewal processes, after being centered and scaled in space and time, converges in distribution to a Brownian motion. As a useful technical tool, this theorem characterizes the service completion process in heavy traffic, allowing us to greatly simplify the many-server analysis when service times follow a general distribution.