A Starting Rule for Data Collection in Queueing Simulations.
Abstract
This paper proposes a rule for determining when to start collecting data in a queueing simulation. The rule is designed to reduce dependence between the empty (queue) and idle (servers) initial conditions and the collected sample record. The rule is an outgrowth of earlier work by Fishman and Moore (1978) and relies on a comparison between a priori information on the activity level (traffic intensity) and a corresponding sample estimate computed during the course of simulation. Experiments with simulations of the M/M/c queue with c = 1,2,4 and p = .7,.8,.9,.95 reveal that the rule reduces and in most cases removes the dependence on the empty and idle initial conditions. In particular, the rule begins data collection when the simulation is in a congested state or in the steady state. The rule is well behaved in that it has low probabilities of requiring long runs before data collection is started. Although our data suggests an association between the rule's performance and activity level, the performance is insensitive to variation in the number of servers. Since the rule is based upon the activity level, a parameter that frequently can be computed from the input parameters of the simulation, the rule is easily generalized to a wider class of queueing simulations.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1979
- Accession Number
- ADA073622
Entities
People
- George S. Fishman
- Veena G. Adlakha
Organizations
- University of North Carolina at Chapel Hill