Design of new multi-domain network algorithms
Multi-Type ACO has different types of ants, all working together with their type, but competing with other types. This is again achieved by pheromones: ants are attracted to their own pheromone type and repelled by pheromone of another ant type. In this way it is possible to find disjoint paths with minimal combined cost simultaneously or we can place a pair of disjoint paths that are protected by a single backup tree. In the last application mentioned we show that it is possible to use the competition between types to guarantee disjoint paths and to use the collaboration between ants to make the backup-paths converge to a single backup-tree. The applicability of this Multi-Type technique within multi-domain networks will be evaluated against other heuristics, with attention to issues such as computational complexity and the performance indicators mentioned above. Next to the Multi-Type ACO technique, COMO will evaluate other related reinforcement learning techniques, like Continual Exploration. In this approach, the degree of exploration is determined in each node in function of entropy of the probability distribution used for selection of the next action or link. This technique allows spreading the load over the network in a very efficient way. Within this project, we will investigate whether this approach can be used for determination of the gateways and paths within each domain.