Simulating Stable Stochastic Systems, VI: Quantile Estimation
Journal of the ACM (JACM)
| Hi-index | 0.00 |
We model key-punching in a computer center as a queuing simulation with 2 servers (typists) and 3 priority classes (small, medium, large jobs). The 90% quantile of queuing time is estimated for different borderlines between the 3 job classes. Confidence intervals for the quantiles are based on the regenerative properties of the simulation, as derived by Iglehart (1974). They utilize the asymptotic normality of the estimated quantile, and a rather complicated expression for its variance. Numerical results are given for the quantiles (and averages) of the queuing times in each job class, for several borderlines between the 3 job classes. The effects of simulation runlength on the confidence intervals were also examined. The effects of varying job-class borderlines were tentatively modeled by a regression model.