Answers are grounded in this paper and cite their sources.
Suppose one simulates the M/M/1 queue for finite amount of time t, replicating the simulation N times, each time with an initial condition of n in the queue. An estimate of the mean of the steady-state distribution is the average (across replications) number in the queue at time t. We show that for large enough t an initial condition of n = 0 (in contrast to the often-conjectured initial condition n = λ/μ, the steady-state mean) minimizes mean-square error of the estimate. An example illustrates that the “large enough” t need not be very large—one-quarter the relaxation time in the example studied. An approximate tradeoff between N and t when n = 0 is also developed.
No takes yet. Share an insight, caveat, or question.
Albert Madansky (1976) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: