Research On Energy-efficiency Scheduling Based On DAG In Heterogeneous Multiprocessor Platform

Comparing to a single processor system,multi-processor system which has been widely applied and studied in recent years has a stronger ability in parallel processing,higher computing density and lower frequency of calculation.In addition,the energy crisis has made the power of the computer system to be an important research topic.Task scheduling can significantly increase the performance of multiprocessor systems and reduce the power consumption of systems.Task scheduling studies commonly use directed acyclic graph,(DAG)as task models,describing tasks that have a context of constraints.How to improve computing performance through reasonable scheduling tasks,and minimizing the power consumption of heterogeneous multiprocessor systems are the focus of this thesis.Dynamic voltage and frequency scaling(DVFS)and voltage frequency island(VFI)are common multiprocessor system low-power technologies.This thesis uses DAG as the task model and studies the low-power scheduling algorithm based on the heterogeneous multiprocessor system with DVFS and VFI technologies.This research contents described as follows:This thesis presents a multi-processor fault tolerant and energy-efficiency scheduling algorithm based on DVFS.In heterogeneous multiprocessor scheduling based on DVFS,DVFS technology based on the different needs of the tasks for computing power and dynamically adjusts the voltage and frequency of the processor to achieve energy efficiency.In a soft real-time system,task scheduling needs to ensure that the number of faults which task missed its deadline is not higher than a certain number,this process is called fault tolerant scheduling.The algorithm that this thesis use can significantly reduce the time overhead of finding the best checkpoint setup for a task set and consider the differences in computational overhead of different processors at the same time.In the task allocation phase,assigning tasks to the processor that minimize the fault tolerant heterogeneous multiprocessor energy(FTHME),which can effectively reduce the energy consumption of the system.Simulation results show that the proposed algorithm effectively lower the system power consumption of fault tolerant scheduling on heterogeneous multiprocessor platform.This thesis presents a multi-processor performance and power consumption optimization algorithm based on dynamic VFI.In heterogeneous multiprocessor scheduling based on VFI,VFI technology divides processors into multiple voltage frequency islands,and each island has distintive voltage domain and frequency,all processors within the same island have the same voltage and frequency.Through assigning the task to the appropriate island by it's characteristics,the system can be optimized for performance and power consumption.The algorithm of this thesis not only enlarges the space of system power consumption in structure,but also assigns tasks according to energy consumption and finish time(EAFT).In the calculation of EAFT,the weight of the scheduling length and the power consumption are set to balance the performance and energy costs of the system,and to improve the performance of the system while reducing the system power consumption.Simulation experiments show that the proposed algorithm reduce the power consumption of the system comparing with the previous static island algorithm and the length of scheduling was greatly reduced,as well.The overall performance of the system are improved comparing with the previous dynamic island algorithm.