Energy Efficient Scheduling of Scientific Workflows in Cloud Environment

Abstract

High end scientific applications in the form of workflows are being executed in cloud for various benefits. But with an increase in the processing capabilities of the cloud system, the energy consumption has also increased significantly. Thus energy efficient execution of these scientific workflows in cloud becomes essential. Existing research on energy efficient scheduling of scientific workflows in cloud mainly focus on reducing only the dynamic energy consumption of the compute nodes and uses DVFS technique. In this paper, we have proposed six different energy efficient scheduling approaches for a set of online scientific workflows in a cloud system considering both static and dynamic energy consumption of the compute nodes. These approaches are divided into two categories: non-splittable allocation of VMs on single host, and splittable allocation on multiple hosts. We have compared the performance of our proposed policies with state-of-art energy efficient scheduling policy, EnReal and found that the policies perform better than EnReal. All three scheduling policies with non-splittable VM allocation perform at par with EnReal in energy consumption but they do not require any migration of VMs. And all policies under splittable VM category perform significantly better than EnReal with an average energy reduction of 70%.