Ergodicity of one-dimensional resource sharing systems. (Q1766051)

Resource sharing systems appear when several servers must complete jobs using some resources which must be shared with other servers. Such situations are common in many fields, and the applications to computer and communication networks have attracted much attention in the past years, both in deterministic and stochastic contexts. Within the latter there are some random quantities involved, and the long term behaviour of the system is studied. The authors consider a stochastic version of these systems. At each point of a graph \(G\) there is a server who can do different types of jobs. Jobs arrive at a server following Poisson processes whose rates depend on his state and the state of the rest of the servers. The times of completion of a job by a server are exponentially distributed with rates depending again on the type of the job and on the state of the server and the other servers sharing resources with him. All the random times are assumed to be independent. The case when \(G=Z\) is studied, that is, the servers are placed along an infinite line, and the model can be seen as an interacting particle system. The authors first get, by coupling techniques, an estimate of the invariant measures. Then, for processes having a reversible measure, they show the uniqueness of the invariant measure and conclude its ergodicity. As a consequence, they prove that every loss network on \(Z\) with calls of bounded length is ergodic.