Computational methods for the local control of loss-sensitive sources in integrated services networks

Procedures for solving a constrained control problem arising in the study of a network access point model are considered. The methods are formulated in the dual domain of the conventional linear programming solution technique. They trade numerical accuracy of substantially lower memory and time requirements than conventional approaches. The methods find the maximum of a one-dimensional piecewise-linear curve through function evaluation. In principle, they could also be used as line search procedures for multidimensional maximization problems arising from control structures with multiple constraints