Abstract
A repairable-item provisioning system with two levels of repair is presented. Under the assumption that the machine time-to-failure and the repair times are exponentially distributed, a new algorithm is developed to compute the long-run average number of machines operating. Using the new algorithm we determine the optimal number of machines and repair channels at the two repair centres to minimize cost and meet a service-level constraint. The algorithm, which is based on Little's result in queueing theory and the theory of regenerative processes, is extremely efficient in terms of computer storage and execution time.