Adaptability Driven by Quality of Execution in High Level Virtual Machines for Shared Cloud Environments

Cloud infrastructures execute workloads from different tenants supported by a non-trivial virtualization stack, which includes high language virtual machines, operating system services and system-level virtual machines. As more and more applications target high level virtual machines (such as the Java VM), they are a relevant abstraction layer not properly explored to enhance resource usage, control, and effectiveness. We propose an economics-inspired model to balance relative resource savings (e.g., to prioritize tenants) and perceived performance degradation, resulting in a yield of applying a given management strategy. The model can be used to drive a resource scheduling algorithm aiming to determine where the reduction will be more economically effective, i.e., will contribute in lesser extent to performance degradation. We discuss how critical resources (heap size and CPU) can be allocated and transferred among high level virtual machines. Experimental evaluation shows that the application of our model, when choosing to take the appropriate resource allocation, results in a significant yield to the cloud provider while, in most cases, execution degradation is small.