Should transient analysis be taught?

  • Authors:
  • M. C. Court;J. L. Pittman;H. T. L. Pham

  • Affiliations:
  • University of Oklahoma, Norman, OK;University of Oklahoma, Norman, OK;University of Oklahoma, Norman, OK

  • Venue:
  • WSC '05 Proceedings of the 37th conference on Winter simulation
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

The authors present the results of experiments performed to identify the pitfalls of performing 'bad' transient analysis when estimating steady-state parameters via the method of independent replications. The intention was to demonstrate to students that failure to delete transient data may lead to confidence intervals that underestimate steady-state parameters. Two types of systems are analyzed: M/M/1/GD/∞/∞ systems and an M/M/s/GD/∞/∞ optimization problem. These systems are chosen since they are typically taught in an undergraduate stochastic operations research course where a closed-form solution of the steady-state parameter exists. Surprisingly, the results prove to support the opposite of our original intention---regardless of run length, ignoring transient analysis often leads to the same level of coverage at greater precision, or provides no gain in coverage to justify the effort of performing transient analysis. Thus, we now pose the question---should transient analysis be taught?