呼損頻度を用いた電話ネットワークの適応的ルーチング

抄録

We consider a problem of routing, that is, a problem of choosing an optimal route for each call in telephone networks. For this problem we present an adaptive routing scheme based on observation of traffic information. A main adaptive routing scheme that has been presented is a scheme using learning automata. Under this scheme, a learning automaton is located at each station. For each call arriving at the station, the automaton makes probabilistic choice of a link from some set of links connecting with the next relay stations. This scheme asymptotically attains equalization of blocking probabilities among the links. This “load equalization” provides a desirable effect to decrease the blocking, but with a possibility that a call takes a quite roundabout route.
To avoid such a possibility and to reduce costs of connecting calls, for each call generating at each station our scheme makes deterministic choice of a route from some set of routes leading to the destination, based on observation of a blocking frequency of each route. It is shown that our scheme asymptotically attains equalization of the blocking frequencies among the routes, which is also one kind of load equalization.