Transaction tagging in highly congested queueing simulations

The purpose of this paper is to illustrate how a very simple queueing model can be used to gain insight into a computer memory management strategy that is important for a large class of discrete-event simulation models. To this end, an elementary queueing model is used to demonstrate that it can be advantageous to run transaction-based simulations with a relatively few tagged transactions that collect data. The remaining transactions merely congest the system. Conceptually the tagged transactions flow through the simulation acting similar to radioactive trace elements inserted into a biological system. The queueing model analyzed in this paper provides insight into some trade-offs in simulation data collection. We show that, while resulting in a longer computer run, an optimal tagging interval greater than one will minimize the probability of prematurely aborting the run. Finally, we propose a heuristic procedure to estimate the optimal tagging interval. We illustrate this with an actual simulation study of a steel production facility.This research was partially supported by a grant to Cornell University by the Bethlehem Steel Corporation