QMapper: Dynamic Power and Performance Management in Virtualized Web Servers Clusters

• Andre Felipe Monteiro CEFET/RJ
• Felipe Henriques CEFET/RJ

Resumo

This work presents QMapper, an approach for modeling and managing Virtual Machines (VMs) on virtualized web servers clusters. In addition to providing energy savings and maintenance of the QoS of the cluster's applications, our model has linear scalability and high adaptability, enhancing its use in scale-out data centers and cloud computing platforms, and defines a standard VM, known as VAS (Virtual Application Server). A VAS operates in a flexible manner adapting its performance and power consumption at runtime according to the workload. Concepts of VAS agile clone, co-allocation of VAS in the same core, and dynamic voltage and frequency scaling (DVFS) of the processor are used in the model in order to manage the trade-off between power and performance. The experiments evaluate the effectiveness of our model by means of power consumption reduction and QoS violation as compared to the Linux governors and a state-of-the-art approach based on optimization. The results show that our model conserves up to 51.8% of the power required by a web servers cluster designed for peak workload scenarios, with a negligible impact on the application's performance.

Palavras-chave: Power-aware computing, Resource Management, Virtualization, Web Servers Clusters