Supervised learning based automatic adaptation of virtualized resource selection policy

Network virtualization techniques enable network service providers to implement and provide multiple logical networks on a common physical network infrastructure. It is essential to guarantee quality of services (QoS) by automatically selecting appropriate network and computer resources to virtual networks (consisting of edge, core networks and data-centers) requiring high level of QoS when the surrounding network environments are rapidly-varying. In this paper, we propose to apply a classification method of supervised learning to determining a policy of virtualized resource selection for virtual networks (VNs). When a new VN is constructed, virtualized resources are selected according to the resource selection policy decided by the learning method which derives a classification boundary by means of past records on combinations of service requirements, time-varying network environments and actually-allocated virtualized resources. Then, our proposed scheme automatically and adaptively updates the resource selection policy by autonomously executing relearning on the basis of urgent contingencies, performances of VNs, and so on. Our proposed scheme will make it easier to exclusively occupy virtualized resources guaranteeing QoS especially for VNs requiring high level of QoS by reducing the blocking probability in resources request for the construction of the VNs.