Bridging the gap between theory and implementation in cognitive networks: developing reasoning in today's networks

Communication networks are becoming increasingly complex and dynamic. The networking paradigm commonly employed, on the other hand, has not changed over the years, and, as a result, performs poorly in today's environments. Only very recently, a new paradigm named cognitive networking has been devised with the objective to make networks more intelligent, thereby overcoming traditional limitations and potentially achieving better performance. According to such vision, networks should be able to monitor themselves, reason upon the environment, act towards the achievement of specific goals and learn from experience. Thus far, several cognitive network architectures have been conceived and proposed in the literature, but, despite researchers seem to agree on the need for a holistic approach, their architectures pursue such a global vision only in part, as they do not consider networks nor network nodes in their entirety. In the present work, we analyze the aspects to be tackled in order to enable this holistic view and propose to base reasoning on both intra- and inter-node interactions, with the ultimate aim to devise a complete cognitive network architecture. After a thorough analysis of advantages and drawbacks of generic reasoning framework, we select the most apt to form the basis on which to build the cognitive network we envision. We first formalize its application in network environments, by determining the steps to follow in the process to equip traditional network with cognitive capabilities. Then, we shift the focus from the design side to the implementation side, by identifying the problems that could be faced when realizing such a network, and by proposing a set of optional refinements that could be taken into account to further improve the performance in some specific situations. Finally, we tackle the problem of reducing the time needed for the cognitive process to reason. Validation through simulations shows that explicitly considering cross-layer intra- and inter-node interactions when reasoning has a twofold effect. First, it leads to better performance levels than those that can be achieved by today's non-intelligent networks, and second, it helps to better understand existent causal relationships between variables in a network.