Virtual machine profiling for analyzing resource usage of applications

Abstract

From the cloud provider perspective, applications are usually black boxes hosted on Virtual Machines. Managing these black boxes without knowing anything about the features of the workload can generate inefficiencies in the performance. In fact, this information can be relevant to take deployment decisions which consist both in considering the interferences between applications with similar resources demands and predicting future peak demands avoiding performance degradation. Monitoring applications in cloud facilities and data centers is the only approach to manage and ensure the performance level of the hosted applications. This paper considers applications as black boxes and, using monitoring data analysis of the VMs on which applications are running, provides a methodology for building an application profile reflecting relevant behavioral features of a VM. This information is precious to lead deployment and adaptive decisions in data center management. The approach is validated on a real monitoring data set of an Italian data center.