Towards Energy Efficient Scheduling for Online Tasks in Cloud Data Centers Based on DVFS

Abstract

Energy efficient task scheduling is an important issue in cloud data centers. Dynamic Voltage Frequency Scaling (DVFS), which can make the processors work at suitable frequency, is an effective method to achieve power saving since the frequency could be automatically adjusted dynamically. However, the existing DVFS-oriented performance model does not suit many applications' computing paradigm in the cloud data centers. Meanwhile, the existing DVFS-oriented power consumption model have a lack of accuracy, and this situation makes it inefficient to achieve power saving. In this paper, we conduct extensive experiments in a real cluster testbed, and propose new DVFS-oriented performance and power consumption models, while taking into account both the frequency and utilization of the processors. Based on the models, we present a Power-aware Threshold Unit (PTU) algorithm to schedule the online tasks dynamically in cloud data center. The PTU algorithm is based on the fact that data centers are organized by rack-sized unit. The basic idea is to make a tradeoff between the power consumption and set up time of serves under a designated granularity. To the best of our knowledge, we are the first to propose the new characterization models and problem. We carry out extensive real experiments on a cluster which consists of several multicore servers, and the results show that the new DVFS-oriented performance and power consumption models are accurate. The experiment results show that our PTU algorithm can achieve considerable energy saving.