QoS-Driven Frequency Scaling for Energy Efficiency and Reliability of Static Web Servers in Software-Defined Data Centers

摘要

Conventional dynamic voltage and frequency scaling (DVFS) techniques use high CPU utilization as a predictor for user dissatisfaction, to which they react by increasing CPU frequency. However, QoS requirements are not linearly related to CPU utilization since they can vary from one business scenario to another. In this paper, we propose QoS-driven frequency scaling for energy-saving and reliability enhancement of static web servers in software-defined data centers (QoS-FS), an adaptive QoS requirement aware dynamic CPU frequency scaling technique. QoS-FS is implemented on a Linux-based static web server. Compared with the default Linux CPU frequency controller, QoS-FS reduces the measured system-wide power consumption of the static web server by about 5% while meeting user's QoS requirements. Besides, under some heavy workload, QoS-FS is 17% more reliable than the default Linux CPU frequency controller in the static web server.