Flow control protocols for integrated networks with partially observed voice traffic

The design of flow control protocols for integrated networks with partially observed voice traffic on the data link level is investigated. A closed Markovian queueing network with two classes of users and a preemptive resume queueing discipline for modeling the integrated link is introduced. The class of admissible flow control policies analyzed maximizes the average data link throughput subject to an average system time delay constraint on a finite horizon. A separation principle between the flow control policy and an estimate of the state of the voice traffic is derived. In particular, it is shown that the optimum control law is bang-bang and the conditional mean estimate of the state of the voice traffic is a sufficient statistic for the optimal control strategy. Since the complexity of the analytical results prohibits finding the explicit flow control policy, a suboptimum and easily implementable adaptive window flow control mechanism is proposed. The window size changes dynamically according to the estimated state of the voice traffic at the destination node. The robustness of the estimator, the dynamics of the window size, and the effectiveness of the suboptimum scheme are verified by means of simulations.