Bounds for rewards of systems with client/server interaction
Susanna Donatelli, Serge Haddad, Patrice Moreaux · Base Institutionnelle de Recherche de l'université Paris-Dauphine (BIRD) (University Paris-Dauphine) · 1999
This paper presents a new method to compute bounds of performance parameters of Markov chains exhibiting a partition of the state space with some family of subsets visited in a sequential order. We use this structure to compute bounds of steady state reward rates on these subsets without computing the global steady state probabilities of the whole chain. The method presented is based on a combination of an aggregation procedure on the subsets and a strong stochastic ordering on the resulting aggregated space.