MapReduce Multi-target For Embedded Heterogeneous Multi-core Power Consumption Management Research And Implementation

As people continue to pursue high-performance computing,in resourceconstrained embedded systems,power consumption is becoming one of the major constraint affecting the performance and improvement of multi-core processors.Research on power consumption management and reasonable allocation of system resource in heterogeneous multi-core environments has become a hot topic in the field of multi-core research in recent years,and it is also the main direction of current power consumption management research.Relative to the development of the hardware level,power consumption management software technology is relatively lagging behind,but the energy efficiency of heterogeneous multi-core systems is largely determined by the resource management software on the system architecture.At present,most software-level power consumption management research focuses on processor dynamic power consumption.However,as the integrated circuit manufacturing process enters the nano stage,the static power consumption is more and more important in total power consumption compared to dynamic power consumption,the static power consumption becomes a key point and difficulty of power consumption management.Aiming at the frequent changes of power consumption constraint in mobile computing systems and the inability of DVFS to effectively overcome the energy loss caused by static power consumption,this paper has proposed a MapReduce multi-target power consumption management method for embedded heterogeneous multi-core.According to real-time power consumption constraint,a core-adjustment strategy has been established to determine processor core resource,and combined with OS thread affinity,process migration,and CPU Hotplug to complete core turn-on,turn-off,and load management to achieve multi-target power consumption management.The research work of this thesis includes:(1)Establishing a heterogeneous system model and designing a static power management method based on this modelThe paper has studied and analyzed the development of static power consumption management at home and abroad,and studied the static power consumption management method for heterogeneous multi-core system accordding to the current research focus of power management.A heterogeneous multi-core processor system model has been established.Based on this model,a static power consumption management method with optimal performance or maximizing energy saving priority has been designed.(2)Studying the porting of Phoenix on the big.LITTLE heterogeneous multi-core platform.The paper has analyzed the shared memory based MapReduce programming model Phoenix,and port it to the big.LITTLE heterogeneous multi-core platform.on the big.LITTLE heterogeneous multi-core platform,analyzed the Phoenix programming model and port it to the big.LITTLE architecture.The parallel processing capability of heterogeneous systems is fully exploited through the Phoenix multi-threaded task scheduling mechanism.Utilizing the flexible configuration of heterogeneous cores takes full advantage of the big.LITTLE architecture with both high performance and low power consumption.(3)Proposing a multi-target power consumption adaptive control method based on big.LITTLE architectureBased on the previous research,a multi-target power consumption adaptive control method based on big.LITTLE architecture has been proposed.It is used to solve the problem that the power consumption constraints of mobile computing systems change frequently and DVFS can not effectively overcome the energy loss caused by static power consumption.The proposed MapReduce multi-target power consumption management technology for embedded heterogeneous multi-core is mainly aimed at the inability of DVFS to effectively overcome the energy loss caused by static power consumption.Utilizing the flexible configuration of heterogeneous cores,under the condition of real-time power consumption constraint,exploits the energy efficiency of the system fully with the goal of optimizing performance or maximizing energy savings.This method can not only meet the real-time power consumption constraints of the processor,but also it can sense external power consumption constraint changes in time,and can use new external energy to invest in new computing resources that can fully utilize system resources and improve system energy efficiency.