Power Consumption Optimization for Deadline-Constrained Workflows in Cloud Data Center

Abstract

With the rapid expanding of the demand for shifting application workflows in both scientific and business fields to clouds to reduce the execution time and costs, cloud data centers are increasingly established and their volumes are becoming bigger. As a result, soaring power consumption of data centers and the related environmental costs have become inevitable concerns of cloud providers and government. We study the power consumption optimization problem in homogenous cloud data centers and propose a method consisting of two algorithms to schedule deadline-constrained workflows. Time utilization maximization scheduling (TUMS) is a heuristic list scheduling algorithm which is designed to find the minimum number of VM instances needed to finish a workflow in the given time. Then, through slack time reclamation, working time minimization (WTM) algorithm minimizes the working time of the VMs being used in the scheduling scheme. By comparing with some effective scheduling algorithms on randomly generated DAGs, the experimental results show that the TUMS can guarantee the deadline constraint with less VMs, in addition, with these two algorithms, our method can effectively reduce the power consumption.