Control Problem for a Multi Server Loss System

摘要

A stationary break-in policy is discussed that controls a multi server losssystem. The policy allows the removal of a job from the system in favor of an arriving job. The decision epochs coincide with arrival epochs and the state space is comprised of service time information of jobs present at these epochs. It is shown that the proposed policy attains the maximal possible fraction of completed jobs in the long run given perfect information over the future of the underlying processes. A derivation of the optimal fraction for the system GI/G/1/0 is given.