Effectiveness and limits of response surface methodology in application to discrete and stochastic simulation of manufacturing plants