TWO-LEVEL DECOMPOSITION METHOD FOR RESOURCE ALLOCATION IN TELECOMMUNICATION NETWORKS

In this paper, we consider a two-level problem of resource allocation in a telecommunication network divided into zones. At the upper level the network manager distributes homogeneous resource shares among zones in order to maximize the total network profit, which takes into account the inner zonal payments from users and the implementation costs. This means that each zonal income calculation at a given resource share requires solution of the inner resource allocation problem. As a result, we obtain a two-level convex optimization problem involving non smooth functions whose values are calculated algorithmically. Unlike the usual convex non smooth optimization methods we suggest this problem to be solved by a Lagrangean duality method which enables us to reduce the initial problem to a sequence of hierarchical one-dimensional problems. Besides, we suggest new ways to adjust the basic problem to networks with moving nodes. We present results of computational experiments which confirm the applicability of the new method. n this paper, we consider a two-level problem of resource allocation in a telecommunication network divided into zones. At the upper level the network manager distributes homogeneous resource shares among zones in order to maximize the total network profit, which takes into account the inner zonal payments from users and the implementation costs. This means that each zonal income calculation at a given resource share requires solution of the inner resource allocation problem. As a result, we obtain a twolevel convex optimization problem involving non smooth functions whose values are calculated algorithmically. Unlike the usual convex non smooth optimization methods we suggest this problem to be solved by a Lagrangean duality method which enables us to reduce the initial problem to a sequence of hierarchical onedimensional problems. Besides, we suggest new ways to adjust the basic problem to networks with moving nodes. We present results of computational experiments which confirm the applicability of the new method.