S. Pierre, A NEW METHODOLOGY FOR GENERATING RULES IN THE TOPOLOGICAL DESIGN OF COMPUTER-NETWORKS, Engineering applications of artificial intelligence, 8(3), 1995, pp. 333-344
The topological design problem consists in finding network topologies
which satisfy a set of performance criteria at the lowest cost. This p
roblem inherently bears the risk of combinatorial explosion, and there
fore requires the use of heuristic methods. More specifically, one suc
h approach uses a knowledge-based system allowing the application of a
set of rules, called perturbation rules, to a starting topology, in o
rder to reduce the total cost of the links or to restore a delay const
raint. Acquiring these perturbation rules constitutes one of the most
important aspects of this approach, which essentially attempts to redu
ce the search space of candidate topologies. This paper focuses on the
development of a sound methodology for generating such rules. Experim
ental results confirm the adequacy and effectiveness of the rules indu
ced by this methodology in the context of dimensioning reliable packet
-switched computer networks.