Capacity Allocation in a Service System: Parametric and Data-Driven Approaches

We study the capacity allocation problem for a service system that serves its customers with a deterministic service time under a service level requirement. The service level is measured by the probability of customers waiting longer than a pre-specified duration. We model the system as an M/D/1 or a G/D/1 queue and examine two approaches to determining the capacity: a parametric approach based on the effective bandwidth theory and a data-driven approach based on the sample average approximation. We conduct a numerical study to investigate the effectiveness of these two approaches, and find that the data-driven approach is more streamlined, accurate, and widely applicable.