Soft Error-aware Energy-efficient Task Scheduling For Workflow Applications In Cloud Computing

In recent years,the scales of cloud data centers have been continuously expanding and the resulting huge energy consumption has gradually become the focus of attention in academic and industrial fields.Dynamic Voltage and Frequency Scaling(DVFS)has been widely used as a promising power management technique to reduce the energy consumed by workflow applications in cloud data centers.However,due to the ever increasing chip density,lowering processor frequencies improperly inevitably increases the occurrence probability of soft errors and affects the reliability of workflow execution.Therefore,in cloud computing environments,how to effectively achieve task schedules with low energy consumption and high reliability for soft error-aware workflow applications is becoming a major challenge to designers of workflow scheduling systems.To address the above challenge,this paper proposes a soft error-aware energyefficient task scheduling approach for workflows in DVFS-enabled cloud data centers.Under the given reliability and deadline constraints requested by tenants,our approach can generate energy-efficient task schedules for workflows by allocating tasks to appropriate virtual machines with specific operating frequencies.This thesis makes three main contributions as follows:1.Based on the checkpointing technique with rollback-recovery,we present a comprehensive modeling approach for soft error-aware workflow scheduling systems in cloud computing environments that support DVFS.After analyzing the impact of frequency variation and soft errors on execution time,reliability and energy consumption of workflow tasks,this paper accurately models the main components of workflow scheduling systems,including virtual machines that support DVFS,workflows,tasks with fault recovery and energy consumption.2.To optimize the energy consumption of workflow applications,we propose a soft error-aware energy-efficient task scheduling approach for workflows.This method comprehensively considers the impact of DVFS and soft errors on the completion time,reliability and energy consumption of workflow applications.Under the given reliability and deadline constraints,our approach can effectively reduce the energy consumption of workflows by allocating tasks to appropriate virtual machines with specific operating frequencies.3.Based on the modeling of workflow scheduling systems and energy-efficient task scheduling strategy for workflows,we design and implement a soft error-aware energy-efficient workflow scheduling tool.This tool can construct a cloud computing environment that supports DVFS,simulate the executions of soft erroraware workflows,and quickly generate and evaluate energy-efficient task schedules for workflow applications submitted by tenants.Comprehensive experimental results on various well-known scientific workflow benchmarks show the effectiveness of our proposed approach.Our approach can generate task schedules with low energy consumption and high reliability for soft error-aware workflow applications in cloud computing environments.Compared with state-of-theart methods,the generated schedules can achieve more energy savings while satisfying both reliability and deadline requirements.