Optimization of capacity allocation among users and services in integrated networks

Access to an integrated services digital network is considered as a two-level problem in which users allocate circuit- and packet-switched traffic types within an assigned portion of a total available bandwidth. In a model suitable for a drop/insert integrated network, the users' bandwidth shares (capacities) are periodically reallocated by a centralized agent (in order to minimize a global cost function) on the basis of the observed overall system's state at the time of intervention. The gradient of the cost with respect to the local capacities is explicitly computed and used in applying a gradient projection algorithm to the central optimization problem.<<ETX>>