Bayesian-based virtual network reconfiguration for dynamic optical networks

Network virtualization is a key technology for accommodating varying traffic demands on optical networks. Network operators construct a virtual network (VN) by slicing physical resources, and resolve traffic congestion by reconfiguring the VN in response to traffic changes. A typical approach is to configure an optimal VN using the traffic demand matrix. However, since it is difficult for this approach to reconfigure the VN following traffic changes, other methods must be explored. Previously we developed a method that observes only the service quality on the VN, i.e., link-level load information on the VN, and searches for a suitable VN by repeatedly making random changes to the current one. However, in practice this noise-induced method repeatedly reconfigures the VN, which leads to over-reconfiguration. In this paper, we propose a VN reconfiguration framework based on the Bayesian Attractor Model, which models the human behavior of making appropriate decisions by recognizing the surrounding situation. Our framework memorizes a set of VN candidates, each of which works well for a specific traffic situation, then retrieves a suitable VN candidate for the current traffic situation from this set. We use certain patterns of incoming and outgoing traffic at edge routers to characterize the traffic situation, as this information can be obtained more easily than the traffic demand matrix. By identifying the stored traffic situation that is closest to the current one and retrieving a suitable VN, our framework can reduce the over-reconfiguration. Evaluation of our method shows that it can identify the traffic situation by observing the amounts of incoming and outgoing traffic at edge routers; as a result it can decrease the number of VN reconfigurations needed to reach a VN suitable for the current traffic situation when compared to the noise-induced method.