Multidimensional Resource Consumption Analysis of Co-Located VMs using PCA

摘要

One of the strategies employed to deal with resource inefficiency in data centres is dynamic virtual machine/container consolidation. The idea behind is, by populating physical servers with an optimal number of virtual machines, all the server's resources (CPU, memory, network bandwidth, etc.) can be utilised effectively. This approach requires (1) the free migration of virtual machine at runtime and (2) the identification of virtual machines which exhibit complementary features. Most existing or proposed approaches are based on elaborate and complex multi-variate optimisation and do not easily lend themselves to fast and intuitive solutions. In this paper, we investigate the scope and usefulness of dimensionality reduction techniques, ideas borrowed from unsupervised machine learning, to analyse the existence of contentious and complementary features in the resource consumption characteristics of co-located virtual machines. Initial results suggest that fast and tractable scheduling can be achieved using these techniques.