面向穿戴应用的大小核架构低功耗策略研究
发布时间：2018-11-16 20:56
【摘要】：随着技术的革新和穿戴设备的发展,目前智能穿戴设备逐渐被广泛应用在军事和医疗等各个领域,智能穿戴设备不断引领着新的潮流,改变着人们的生活,但是其所处理的应用场景却复杂多变。面对不断膨胀的应用需求,消费者要求穿戴设备在不断减少体积和质量的同时,进一步保证系统的性能和有效的降低功耗,延长使用时间和待机时间。然而性能和功耗本身就是不可调和的矛盾,体积减少、性能提升的同时,必然导致系统的功耗提升。因此功耗问题已经严重制约穿戴设备的进一步发展。以ARM big.LITTLE架构为代表的性能异构多核处理器由多个性能和功耗不同的处理器组成,通过不同性能和功耗的处理器对不同应用进行处理,可以有效的降低功耗。针对big.LITTLE架构处理器合理的进程调度和电源管理,对系统资源按需分配,可同时兼顾高性能和低功耗。目前成熟的调度算法或者动态电源框架,都是针对SMP等系统进行设计和优化,不适应本文使用的ARM big.LITTLE性能异构多核架构,更不适应穿戴设备复杂多变的应用场景。通过研究分析现有调度算法的不足,结合穿戴设备特殊的应用场景,提出了动态阈值的HMPDB负载均衡调度算法。该算法根据系统的总体负载调整系统的进程迁移阈值,在保证性能和公平性的同时实现负载均衡,不仅可以有效的降低功耗,更能适应穿戴设备极端复杂的应用场景。另一方面,传统的调度器与动态电源管理策略虽然都已资源分配、功耗降低为目标,但是各有侧重,这些框架各自为政,势必相互影响,造成额外的性能损失和功耗增加,本文进一步改进调度器和动态电源管理系统,以HMPDB调度器为核心,实现了一个目标统一的HMPDB-EAS节能调度框架,通过调度器对CPU和应用程序的负载进行统计分析,协调CPUFreq调频子系统和CPUIdle子系统的运行,在保证CPU性能同时满足系统的负载需求和应用程序的性能需求的同时,降低穿戴设备的功耗,同时延长CPU处于的休眠模式的时间,进一步降低穿戴设备的功耗。经过实验表明,本文设计的动态节能调度框架HMPDB-EAS更加适应穿戴设备的应用场景,通过调度器和动态电源管理框架的协调配合,能够在保证系统性能的同时有效的降低系统的功耗。
[Abstract]:With the innovation of technology and the development of wearable devices, smart wearable devices are widely used in military, medical and other fields. Intelligent wearable devices are leading new trends and changing people's lives. However, the application scenarios are complex and changeable. In the face of the ever-expanding application demand, consumers require wearable devices to reduce the volume and quality of the system while further ensuring the performance of the system and effectively reducing power consumption, prolonging the service time and standby time. However, the performance and power consumption itself is irreconcilable contradiction, the volume reduction, the performance enhancement at the same time, inevitably leads to the system power consumption enhancement. Therefore, the problem of power consumption has seriously restricted the further development of wearable devices. The performance heterogeneous multicore processor represented by ARM big.LITTLE architecture is composed of multiple processors with different performance and different power consumption. Different applications can be processed by different performance and power consumption processors, which can effectively reduce power consumption. According to the reasonable process scheduling and power management of the big.LITTLE architecture processor, the system resource can be allocated according to the demand, and the high performance and low power consumption can be taken into account at the same time. At present, the mature scheduling algorithms or dynamic power supply frameworks are designed and optimized for SMP and other systems, which are not suitable for the heterogeneous multi-core architecture of ARM big.LITTLE performance used in this paper, and not suitable for the complex and changeable application scenarios of wearable devices. By studying and analyzing the shortcomings of the existing scheduling algorithms and combining the special application scenarios of wearable devices, a dynamic threshold HMPDB load balancing scheduling algorithm is proposed. The algorithm adjusts the process migration threshold according to the overall load of the system and realizes load balancing while ensuring performance and fairness. It can not only effectively reduce power consumption but also adapt to the extremely complex application scenarios of wearable devices. On the other hand, although the traditional scheduler and the dynamic power management strategy have been allocated the resources, the power consumption is reduced as the goal, but each has its own emphasis, these frameworks are each other, which will inevitably affect each other, resulting in additional performance loss and increased power consumption. In this paper, the scheduler and the dynamic power management system are further improved. With HMPDB scheduler as the core, a unified HMPDB-EAS energy saving scheduling framework is implemented, and the load of CPU and application is statistically analyzed through the scheduler. The CPUFreq FM subsystem and the CPUIdle subsystem are coordinated to ensure that the CPU performance meets both the system load requirements and the application performance requirements, while reducing the power consumption of the wearable devices, while prolonging the time spent in the dormant mode of the CPU. Further reduce the power consumption of wearable devices. The experimental results show that the dynamic energy saving scheduling framework HMPDB-EAS is more suitable for wearable device application, and coordination of scheduler and dynamic power management framework. It can effectively reduce the power consumption of the system while ensuring the performance of the system.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP368.33
本文编号：2336628
[Abstract]:With the innovation of technology and the development of wearable devices, smart wearable devices are widely used in military, medical and other fields. Intelligent wearable devices are leading new trends and changing people's lives. However, the application scenarios are complex and changeable. In the face of the ever-expanding application demand, consumers require wearable devices to reduce the volume and quality of the system while further ensuring the performance of the system and effectively reducing power consumption, prolonging the service time and standby time. However, the performance and power consumption itself is irreconcilable contradiction, the volume reduction, the performance enhancement at the same time, inevitably leads to the system power consumption enhancement. Therefore, the problem of power consumption has seriously restricted the further development of wearable devices. The performance heterogeneous multicore processor represented by ARM big.LITTLE architecture is composed of multiple processors with different performance and different power consumption. Different applications can be processed by different performance and power consumption processors, which can effectively reduce power consumption. According to the reasonable process scheduling and power management of the big.LITTLE architecture processor, the system resource can be allocated according to the demand, and the high performance and low power consumption can be taken into account at the same time. At present, the mature scheduling algorithms or dynamic power supply frameworks are designed and optimized for SMP and other systems, which are not suitable for the heterogeneous multi-core architecture of ARM big.LITTLE performance used in this paper, and not suitable for the complex and changeable application scenarios of wearable devices. By studying and analyzing the shortcomings of the existing scheduling algorithms and combining the special application scenarios of wearable devices, a dynamic threshold HMPDB load balancing scheduling algorithm is proposed. The algorithm adjusts the process migration threshold according to the overall load of the system and realizes load balancing while ensuring performance and fairness. It can not only effectively reduce power consumption but also adapt to the extremely complex application scenarios of wearable devices. On the other hand, although the traditional scheduler and the dynamic power management strategy have been allocated the resources, the power consumption is reduced as the goal, but each has its own emphasis, these frameworks are each other, which will inevitably affect each other, resulting in additional performance loss and increased power consumption. In this paper, the scheduler and the dynamic power management system are further improved. With HMPDB scheduler as the core, a unified HMPDB-EAS energy saving scheduling framework is implemented, and the load of CPU and application is statistically analyzed through the scheduler. The CPUFreq FM subsystem and the CPUIdle subsystem are coordinated to ensure that the CPU performance meets both the system load requirements and the application performance requirements, while reducing the power consumption of the wearable devices, while prolonging the time spent in the dormant mode of the CPU. Further reduce the power consumption of wearable devices. The experimental results show that the dynamic energy saving scheduling framework HMPDB-EAS is more suitable for wearable device application, and coordination of scheduler and dynamic power management framework. It can effectively reduce the power consumption of the system while ensuring the performance of the system.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP368.33
【参考文献】
相关期刊论文 前6条
1 杨亚琪;栾钟治;杨海龙;杨姝;钱德沛;;异构多核下兼顾应用公平性和能耗的调度方法研究[J];计算机工程与科学;2016年05期
2 王益涵;王凯林;孙宪坤;张冬松;吴飞;;基于CPUfreq的DVFS节能技术的研究与实现[J];计算机测量与控制;2016年02期
3 刘念唐;翁宇;林雨;张文睿;韦志磊;邵X;;基于软件行为预测的动态电源管理方案[J];计算机工程;2015年06期
4 周亦敏;沈云龙;曹丽东;;基于异构多核平台H.264解码的DVFS算法[J];计算机工程;2013年11期
5 刘婷;王华军;王光辉;;基于Linux内核的CFS调度算法研究[J];电脑与电信;2010年04期
6 朱旭;杨斌;刘海涛;;完全公平调度算法分析[J];成都信息工程学院学报;2010年01期
相关硕士学位论文 前2条
1 刘海峰;基于Android嵌入式系统的低功耗优化[D];北京交通大学;2016年
2 高晓川;面向动态异构多核处理器的公平性任务调度研究[D];中国科学技术大学;2015年
，本文编号：2336628
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2336628.html
