Real-time Traffic Allocation Using Learning Automata

We present a new fixed structure, multi-action, multi-response learning automaton and use it to allocate arriving traffic at a multimedia network. For each source-destination pair, for each traffic type, a learning automaton allocates every new arriving call on one of the available routes from source to destination or rejects it. The state diagram of the learning automaton has a star shape. Each branch of the star is associated with a particular route. Depending on how much "good" the traffic performance is on a route, the automaton moves deeper in the corresponding branch. On the other hand, depending on how much "bad" it is, the automaton moves out of this branch. Finally, we provide several performance nietrics to characterize the traffic performance on a route as "good" or "bad".