Suppose that call reservation requests of K different types arrive randomly at a single capacitated link where each in turn must be accepted or rejected. Each request requires an amount of bandwidth and a random duration, both depending on its type. An accepted call generates an immediate fixed revenue followed by a variable revenue per unit time. We assume that each call's requested duration is known when it arrives and that the state of the link is constantly monitored. The problem is to find a call acceptance policy that maximizes the long-run average revenue per unit time generated by the accepted calls. We propose and investigate threshold-type control policies that use the known duration of arriving calls as well as the current link status. Two main contributions result. First are interpretable analytical results for the case of K=1 that can also be applied to the case of K>1 using complete partitioning scheme. Second, we illustrate how to apply stochastic optimization techniques to compute optimal thresholds in the general case. We also include the results of initial simulation experiments.
