Transaction tagging in highly congested queueing simulations (Q1116206)

In a simulation model, a lot of entities are usually concurrently active at a given time. The attributes of each entity (transaction) require a memory space which can be so large that the program is aborted. The entity congestion problem is avoided by using a transaction ``tagging'' technique. The tagged transactions have all their attributes while the untagged transactions carry only those attributes which are necessary for the logic of the model. The problem considered in this paper is to determine the optimal interval k between two consecutive tagged transactions that minimizes the probability of prematurely aborting the run. The authors suggest to compute the probability of running out of memory by modeling this process as a random walk (Markov chain). A heuristic procedure to estimate the optimal tagging interval is also given. An application illustrates these techniques.