This paper establishes bounds on the effect of a perturbation of a distributional parameter, on performance measures of a DES that depend on n stochastic variables, such as the transient waiting time in a queueing network. The key observation is that, under quite general conditions, the Lipschitz constant of a probability measure can be computed in a simple way. Also, parameter-free perturbation analysis will be considered, that is, assessing the effect of replacing a given probability measure µ by another one, say ¿ .
|Title of host publication
|Proceedings 8th International Workshop on Discrete Event Systems, WODES'06, Ann Arbor MI, USA, July 10-12, 2006
|S. Lafortune, F. Lin, D. Tilbury
|Place of Publication
|Institute of Electrical and Electronics Engineers
|Published - 2006