Energy-aware Scheduling For Workflow On DVFS-Enabled Heterogeneous Cluster

Energy consumption accounts for most cost of operation and maintenance in data centers.Cloud service providers realize that energy-saving technologies have many advantages: firstly,reducing carbon emissions is conducive to environmental protection;secondly,reducing energy consumption costs achieve more economic benefits.Therefore,how to reduce the energy consumption of cloud data centers has become an important issue in academia and industry.As an effective energy-saving technology,DVFS technology has been widely used in practice,by dynamically adjusting the chip's operating frequency and voltage to reduce execution energy consumption.However,improper frequency adjustment will also violate the QoS requirement of cloud workflow,such as deadlines,reliability and so on.There is a conflict between the QoS requirement of workflow and the negative impact brought by DVFS technology,which makes the energy-aware workflow scheduling on resources supporting DVFS more challenging.Considering the actual scenario,cloud service providers sign service level agreements with users to agree on quality of service.Especially,for aerospace computing,weather prediction and other scientific applications with high real-time requirements,if fault-tolerant mechanism is not considered,once the tasks fails to execute,it may cause disastrous consequences.This thesis studies the optimization of workflow scheduling to minimize the energy consumption of resources supporting DVFS.The objective of this problem is to minimize energy consumption while meeting the deadlines and reliability constraints of workflow.According to the characteristics and challenges of this problem,an ESRWH algorithm is designed.It is divided into five parts: task sequencing,slack time calculation,sub-deadline partition,heuristic resource allocation and slack time block optimization based on frequency regulation.In the task sequencing phase,three methods are designed to generate the task scheduling sequence to find the effective topological sequencing of workflow tasks,so as to meet the priority constraints.Then,the slack time calculation considers the assignment of task replicas,calculates the slack time of workflow,and sets sub-deadline for each task.In the resource allocation phase,two resource allocation strategies are designed to allocate physical machine for each task.In this phase,if the resources can not meet the reliability constraints,replicas are added to the task.In order to make full use of the remaining idle time blocks between tasks on the processor,DVFS technology is used to further regulate the processor voltage and clock frequency to reduce energy consumption.In order to verify the effectiveness of the proposed algorithm,this paper uses ANOVA to calibrate the relevant parameters of the algorithm and select the optimal combination of parameters.Then a large number of workflow examples are generated,and the proposed ESRWH algorithm is compared and analyzed with the benchmark algorithm in the relevant literature.The experimental results show that the proposed ESRWH algorithm is better than the related benchmark algorithm under different deadlines and reliability constraints.