A Fine-grained Model for Adaptive On-demand Provisioning of CPU Shares in Data Centers

Data Centers usually host different third party applications, each of them possibly having different requirements in terms of QoS. To achieve them, sufficient resources, like CPU and memory, must be allocated to each application. However, workload fluctuations might arise, and so, resource demands will vary. Allocations based on worst/average case scenarios can lead to non-desirable results. A better approach is then to assign resources on demand. Also, due to the complexity and size of current and future systems, self-adaptive solutions are essential. In this paper, we then present Grains, a self-adaptive approach for resource management in Data Centers under varying workload.