Routing control of packet flow using neural network

We investigate a packet flow problem in a computer network within a framework of statistical physics. We propose a model for packet routing control by putting a neural network at each node. The energy function for the whole system is defined in order to express competition among queue length, distance to destination of a packet and firing rate of neurons at each node and so on. A dynamical equation for output of neurons is derived in order to make the energy decrease as a function of time, by using a mean field approximation. By means of computer simulations, we calculate an average number of fired nodes, an average number of packets arrived at their destinations, and so on. We find that the buffer size is essential in the proposed model for packet routing control.