No label defined (Q4945339)

A stochastic automata network (SAN) consists of a number of individual stochastic automata that operate in a (more or less) independent manner. Each individual automaton is represented by a number of states and rules and the state of the SAN at a certain time is given by the state of its constituent automata at that time. The use of SANs becomes increasingly important in performance modelling problems related to parallel and distributed computer systems. SAN formalism provides an appropriate alternative to stochastic Petri nets and Markov chain theories. The present paper is devoted to the introduction and analysis of SANs and of the related concept of stochastic process algebras, on one hand, and to numerical issues of computation time and memory requirements in computing stationary distributions by means of iterative methods, on the other hand. Sections 2-4 contain an informal presentation of SANs and tensor algebra. In section 5, a number of sufficient conditions that ensure the existence of product forms of SANs are given. Sections 6-10 investigate suitable numerical iterative methods to compute the SAN stationary distributions. Vector-descriptor multiplications (section 6) and generalized tensor products (section 7) are presented. Section 8 discusses the applicability of the multiplication theorems, section 9 examines the trade-off between memory and CPU-time SAN approach to gain performance, and section 10 deals with numerical solution methods. For the entire collection see [Zbl 0929.00017].