A two-layer traffic clustering framework for network virtualization

Network virtualization, through building multiple virtual networks on top of a shared substrate, provides a potential strategy for addressing the ossification of the Internet and encourages network innovation. Traffic clustering is the base and key technology in network virtu-alization. This paper presents a two-layer traffic clustering algorithm. In the first layer, different types of user traffic are clustered into LDR and HDR according to their time characteristics, in the second layer, a revised K-means algorithm is introduced to further classify the traffic in HDR set, the result is used to match the user specified virtual network construction parameters to traffic clustering parameters which will provide finer bandwidth management capability for SPs. Numerical results obtained from experimental show that the result of clustering is dynamically and universally adaptive to the different types of traffic. The proposed two-layer traffic clustering scheme contributes to the constructing of virtual networks.