Optimal noncyclic server allocation in a polling model

One of the basic problems in high-speed telecommunication networks is how to share resources efficiently between traffic types, that may represent different users or different applications, with different requirements for quality of services. One of the most expensive resources to be shared in high speed networks is the available bandwidth, and a standard mechanism for sharing it between several users with different traffic types is to allocate the whole bandwidth to one of the traffic streams at a time, during time intervals that are predetermined. We consider two types of traffic that arrive according to a Poisson process, with possibly different parameters, to two queues. The size of a packet is exponentially distributed. The access of the queues to the communication channel is controlled by the polling mechanism, so that one packet at a time can be transmitted. The access policy is performed without any knowledge of the number of packets in the queues. In order to obtain efficient access policies, we assign holding costs for each type of packet. We wish to find policies minimizing the long run average costs. The system is modeled as a Markov decision chain with partial information and an algorithm is presented to find good policies. The decision epochs are chosen in two different ways, resulting in different types of policies