Simulating GI/G/k Queues in Heavy Traffic

This paper proposes a new consistent estimator for the expected waiting time in queue in the simulation of a multiserver queueing system. This estimator is called the “error estimator” because it estimates only the error involved in the use of Kingman's heavy traffic approximation. Not only does this estimator retain all the advantages of the regenerative method, but it also reduces the variance of the estimator substantially when the traffic becomes heavy.