A new method for network routing:a preliminary report

The author considers the problem of packet routing in a dynamic communication network as a problem of autonomous agent navigation in an uncertain environment, and proposes a novel method for network routing. The basic idea is that, instead of using an estimated constant, one uses a probability distribution to model the delay of link in a computer network, and computes routing tables based on such a probabilistic model of the network. Since the new model is no longer just a "snapshot" of the network at a particular time, but represents a probabilistic "pattern" which treats the changes of link delays as normal rather than as an exception, it is reasonable to expect that the routing tables based on the new model can be very robust against the dynamic changes of link delays. A computer network with such probabilistic information can be represented by a U-graph and the network routing problem is quite similar to mobile agent navigation in uncertain environments.<
>