Markovian network processes: Congestion-dependent routing and processing (Q751071)

The author generalizes the usual queueing network in the following sense. In the usual Markovian (queueing) network, discrete units move about between a set of nodes which process these units. It is typically assumed that nodes are independent and the routes of the units are independent. Here, it is assumed that nodes and routings may be dependent. The author introduces a general network process representing the number of units at each node and derives its equilibrium distribution. It turns out that this distribution is a product of functions of vectors such that the arguments of the function satisfy an interchangeability property. The paper gives a detailed presentation of several interesting results, with helpful examples. Advantages of the approach used include a general characterization of blocking (of transitions) without the need for reversibility, and conditions under which a moving unit sees a time average for the units which do not move.