Optimal adaptive server allocation in a network (Q1088580)

The paper deals with a queueing network with N nodes. The service times are independent and have the distribution function \(G_ i(t)\) in node i (1\(\leq i\leq N)\). The arrivals are assumed to obey a Poisson process and each arrival is assigned to node i with probability \(p_ i\). Upon service completion in node i, a customer is sent to node j with probability \(r_{ij}\), the customer leaves the network with the residual probability \(r_{i0}=1-\sum_{j}r_{ij}\). The system is supported by a single server who is allocated to one of the nodes in a non pre-emptive way. The matrix \(R=(r_{ij})\) is assumed to have the property that \(R^ n\to 0\) as \(n\to \infty\) so that each customer leaves the network eventually. It is shown that the certainty control system that assigns the server optimally assuming that the sample means of the service times are the correct means is optimal.