ATM flow control design by using high-order Hopfield network dynamics

Considers the problem of scheduling stochastic flows arising in traffic control of ATM networks. The planning mechanism based on an aggregation approach in the state space of a dynamic model describing the flow control is decoupled into a set of flow streams on which a decentralized control is applied. The basic reason for using such an approach is to better track the stochastic flows. Using high-order Hopfield neural network dynamics, the learning control within each cluster of flow patterns is achieved through a Parzen window estimation by a suitable updating of the window parameters within the current control epoch. The results of a numerical experiment on a Markov multiflow pattern when using such a control approach are reported and the comments on the qualitative properties of the control mechanism are presented