Adaptive virtual resource clustering and monitoring through nonlinear dimensionality reduction

Network virtualization provides a promising way to overcome the ossification of current Internet. One important issue in network virtualization is the problem of real-time monitoring of resource usage information. In this paper we investigate a novel method for clustering virtual resources inspired by the nonlinear dimensionality reduction method. Then a clustering algorithm extending the k-means method with the isometric feature mapping (Isomap) is used to analyze the relationships of substrate nodes and links in different time slots. By replacing the classical Euclidean distance with the geodesic distance, we can preserve the intrinsic geometry of the high-dimensional data and discover the regularities and irregularities in the substrate network. Simulation results demonstrate that the proposed method can classify the real-time states of virtual resources and provide accurate VN mapping guidance and resource management.