Stochastic scheduling games with Markov decision arrival processes

Abstract In Hordijk and Koole [1,2], a new type of arrival process, the Markov Decision Arrival Process (MDAP), was introduced, which can be used to model certain dependencies between arrival streams and the system at which the arrivals occur. This arrival process was used to solve control problems with several controllers having a common objective, where the output from one controlled node is fed into a second one, as in tandems of multi-server queues. In the case that objectives of the controllers are different, one may choose a min-max (worst case) approach where typically a controller tries to obtain the best performance under the worst possible (unknown) strategies of the other controllers. We use the MDAP to model such situations, or situations of control in an unknown environment. We apply this approach to several scheduling problems, including scheduling of customers and scheduling of servers. We consider different information patterns including delayed information. For all these models, we obtain several structural results of the optimal policies.