Autonomous Management in Virtual-Machine-Based Resource Providers

Nowadays, virtualization makes possible to consolidate the workloads of applications by sharing the resources of a physical node. The deployment of virtual machines (VMs) leads to a reduction in infrastructure costs, however, the CPU resources need an online management strategy to meet the particular requirements of each application. In this paper, we present and evaluate a component to provide autonomous management of VMs in a multi-core physical node. Our solution adjusts the CPU resources needed by each VM according to the application demand. We achieve this with (1) applying a policy that allows fulfilling the agreed application-level QoS, (2) a priority weight-based mechanism to spread the unused resources between the contending VMs, and (3) a learning mechanism to dynamically reach and meet service level objectives expressed in transactions per second. Our experiments show that the implemented approach meets service level objectives (SLOs).