数据中心以可接受的成本,承载着超大规模的互联网应用.数据中心的能源消耗直接影响着数据中心的一次性建造成本和长期维护成本,是数据中心总体持有成本的重要组成部分.现代的数据中心普遍采用动态电压频率调节(dynamic voltage frequenc...数据中心以可接受的成本,承载着超大规模的互联网应用.数据中心的能源消耗直接影响着数据中心的一次性建造成本和长期维护成本,是数据中心总体持有成本的重要组成部分.现代的数据中心普遍采用动态电压频率调节(dynamic voltage frequency scaling,简称DVFS)来提升单节点的能耗表现.但是,DVFS这一类机制同时影响到应用的能源消耗和性能,而这一问题尚未被深入探索.专注于DVFS机制对应用程序性能的影响,提出了一个分析模型用来量化地刻画应用程序的性能与处理器频率之间的关系,可以预测程序在任意频率下的性能.具体来说,依据执行时访问内存子系统资源的不同,把程序的指令分为两部分——片上指令和片外指令,并分别独立建模.片上指令是指仅需访问片上资源就可以完成执行的指令,其执行时间与处理器频率呈线性关系;片外指令是指需要访问主存的指令,其执行时间与处理器频率无关.通过上述划分和对每一部分执行时间的分别建模,可以获得应用程序的执行时间与处理器频率之间的量化模型.使用两个不同的平台和SPEC 2006中的所有标准程序验证该模型,平均误差不超过1.34%.展开更多
The increase in computing capacity caused a rapid and sudden increase in the Operational Expenses (OPEX) of data centers. OPEX reduction is a big concern and a key target in modern data centers. In this study, the sca...The increase in computing capacity caused a rapid and sudden increase in the Operational Expenses (OPEX) of data centers. OPEX reduction is a big concern and a key target in modern data centers. In this study, the scalability of the Dynamic Voltage and Frequency Scaling (DVFS) power management technique is studied under multiple different workloads. The environment of this study is a 3-Tier data center. We conducted multiple experiments to find the impact of using DVFS on energy reduction under two scheduling techniques, namely: Round Robin and Green. We observed that the amount of energy reduction varies according to data center load. When the data center load increases, the energy reduction decreases. Experiments using Green scheduler showed around 83% decrease in power consumption when DVFS is enabled and DC is lightly loaded. In case the DC is fully loaded, in which case the servers’ CPUs are constantly busy with no idle time, the effect of DVFS decreases and stabilizes to less than 10%. Experiments using Round Robin scheduler showed less energy saving by DVFS, specifically, around 25% in light DC load and less than 5% in heavy DC load. In order to find the effect of task weight on energy consumption, a set of experiments were conducted through applying thin and fat tasks. A thin task has much less instructions compared to fat tasks. We observed, through the simulation, that the difference in power reduction between both types of tasks when using DVFS is less than 1%.展开更多
Dynamic Voltage Frequency Scaling (DVFS) techniques are used to improve energy efficiency of GPUs. Literature survey and thorough analysis of various schemes on DVFS techniques during the last decade are presented in ...Dynamic Voltage Frequency Scaling (DVFS) techniques are used to improve energy efficiency of GPUs. Literature survey and thorough analysis of various schemes on DVFS techniques during the last decade are presented in this paper. Detailed analysis of the schemes is included with respect to comparison of various DVFS techniques over the years. To endow with knowledge of various power management techniques that utilize DVFS during the last decade is the main objective of this paper. During the study, we find that DVFS not only work solely but also in coordination with other power optimization techniques like load balancing and task mapping where performance and energy efficiency are affected by varying the platform and benchmark. Thorough analysis of various schemes on DVFS techniques is presented in this paper such that further research in the field of DVFS can be enhanced.展开更多
研究芯片功耗中动态功耗部分,针对传统动态节能技术动态电压与频率调节(dynamic voltage and frequency scaling,DVFS)技术未能考虑预测CPU未来阶段行为的不足,提出BP-DVFS节能策略。为了提高下一阶段CPU利用率的预测准确性,更准确地对...研究芯片功耗中动态功耗部分,针对传统动态节能技术动态电压与频率调节(dynamic voltage and frequency scaling,DVFS)技术未能考虑预测CPU未来阶段行为的不足,提出BP-DVFS节能策略。为了提高下一阶段CPU利用率的预测准确性,更准确地对CPU进行动态调频进而降低其运行功耗。构建了一种FPU-CPU(forward predict utilization CPU)模型。模型假设下一时间段CPU利用率与CPU运行资源有关的事件特征量存在非线性函数关系,从处理器运行时环境出发提取出与CPU资源紧密相关的5个特征量进行度量,采用BP神经网络进行拟合训练。用训练后得到的神经网络预测CPU下一阶段的利用率,进行CPU处理不同类型任务程序的功耗仿真实验。并在相同实验条件下与常用的3种CPU调频策略实验结果进行对比。实验结果表明,在CPU处理不同类型任务程序时,采用BP-DVFS策略进行调频的CPU功耗都低于其他3种策略进行调频的CPU功耗。通过实验验证,本文提出的方法提高了预测CPU利用率的准确度,降低了CPU运行时功耗。同时验证了假设的合理性与有效性以及此方法实现CPU低功耗运行是有效的。展开更多
在嵌入式多模式视频编码系统中,动态电压频率调整(Dynamic Voltage and Frequency Scaling,DVFS)技术可在一定程序上节约系统能耗,然而持续降低电压和频率可能影响处理器接口资源的传输性能,甚至导致系统无法正常工作.针对该问题,提出...在嵌入式多模式视频编码系统中,动态电压频率调整(Dynamic Voltage and Frequency Scaling,DVFS)技术可在一定程序上节约系统能耗,然而持续降低电压和频率可能影响处理器接口资源的传输性能,甚至导致系统无法正常工作.针对该问题,提出了一种任务敏感的功耗控制方法.通过研究多模式视频编码任务量和处理器资源之间的关系,建立一个任务敏感的资源配置模型,基于该模型设计了一个自适应功耗控制器,在系统工作过程中根据编码任务量的不同动态调节处理器工作频率和工作核数.实验表明,在满足多模式实时视频编码功能和性能要求的基础上,该文提出的方法与传统DVFS技术相比,单帧视频编码的平均功耗节省了11.4%.展开更多
文摘数据中心以可接受的成本,承载着超大规模的互联网应用.数据中心的能源消耗直接影响着数据中心的一次性建造成本和长期维护成本,是数据中心总体持有成本的重要组成部分.现代的数据中心普遍采用动态电压频率调节(dynamic voltage frequency scaling,简称DVFS)来提升单节点的能耗表现.但是,DVFS这一类机制同时影响到应用的能源消耗和性能,而这一问题尚未被深入探索.专注于DVFS机制对应用程序性能的影响,提出了一个分析模型用来量化地刻画应用程序的性能与处理器频率之间的关系,可以预测程序在任意频率下的性能.具体来说,依据执行时访问内存子系统资源的不同,把程序的指令分为两部分——片上指令和片外指令,并分别独立建模.片上指令是指仅需访问片上资源就可以完成执行的指令,其执行时间与处理器频率呈线性关系;片外指令是指需要访问主存的指令,其执行时间与处理器频率无关.通过上述划分和对每一部分执行时间的分别建模,可以获得应用程序的执行时间与处理器频率之间的量化模型.使用两个不同的平台和SPEC 2006中的所有标准程序验证该模型,平均误差不超过1.34%.
文摘The increase in computing capacity caused a rapid and sudden increase in the Operational Expenses (OPEX) of data centers. OPEX reduction is a big concern and a key target in modern data centers. In this study, the scalability of the Dynamic Voltage and Frequency Scaling (DVFS) power management technique is studied under multiple different workloads. The environment of this study is a 3-Tier data center. We conducted multiple experiments to find the impact of using DVFS on energy reduction under two scheduling techniques, namely: Round Robin and Green. We observed that the amount of energy reduction varies according to data center load. When the data center load increases, the energy reduction decreases. Experiments using Green scheduler showed around 83% decrease in power consumption when DVFS is enabled and DC is lightly loaded. In case the DC is fully loaded, in which case the servers’ CPUs are constantly busy with no idle time, the effect of DVFS decreases and stabilizes to less than 10%. Experiments using Round Robin scheduler showed less energy saving by DVFS, specifically, around 25% in light DC load and less than 5% in heavy DC load. In order to find the effect of task weight on energy consumption, a set of experiments were conducted through applying thin and fat tasks. A thin task has much less instructions compared to fat tasks. We observed, through the simulation, that the difference in power reduction between both types of tasks when using DVFS is less than 1%.
文摘Dynamic Voltage Frequency Scaling (DVFS) techniques are used to improve energy efficiency of GPUs. Literature survey and thorough analysis of various schemes on DVFS techniques during the last decade are presented in this paper. Detailed analysis of the schemes is included with respect to comparison of various DVFS techniques over the years. To endow with knowledge of various power management techniques that utilize DVFS during the last decade is the main objective of this paper. During the study, we find that DVFS not only work solely but also in coordination with other power optimization techniques like load balancing and task mapping where performance and energy efficiency are affected by varying the platform and benchmark. Thorough analysis of various schemes on DVFS techniques is presented in this paper such that further research in the field of DVFS can be enhanced.
文摘研究芯片功耗中动态功耗部分,针对传统动态节能技术动态电压与频率调节(dynamic voltage and frequency scaling,DVFS)技术未能考虑预测CPU未来阶段行为的不足,提出BP-DVFS节能策略。为了提高下一阶段CPU利用率的预测准确性,更准确地对CPU进行动态调频进而降低其运行功耗。构建了一种FPU-CPU(forward predict utilization CPU)模型。模型假设下一时间段CPU利用率与CPU运行资源有关的事件特征量存在非线性函数关系,从处理器运行时环境出发提取出与CPU资源紧密相关的5个特征量进行度量,采用BP神经网络进行拟合训练。用训练后得到的神经网络预测CPU下一阶段的利用率,进行CPU处理不同类型任务程序的功耗仿真实验。并在相同实验条件下与常用的3种CPU调频策略实验结果进行对比。实验结果表明,在CPU处理不同类型任务程序时,采用BP-DVFS策略进行调频的CPU功耗都低于其他3种策略进行调频的CPU功耗。通过实验验证,本文提出的方法提高了预测CPU利用率的准确度,降低了CPU运行时功耗。同时验证了假设的合理性与有效性以及此方法实现CPU低功耗运行是有效的。
文摘在嵌入式多模式视频编码系统中,动态电压频率调整(Dynamic Voltage and Frequency Scaling,DVFS)技术可在一定程序上节约系统能耗,然而持续降低电压和频率可能影响处理器接口资源的传输性能,甚至导致系统无法正常工作.针对该问题,提出了一种任务敏感的功耗控制方法.通过研究多模式视频编码任务量和处理器资源之间的关系,建立一个任务敏感的资源配置模型,基于该模型设计了一个自适应功耗控制器,在系统工作过程中根据编码任务量的不同动态调节处理器工作频率和工作核数.实验表明,在满足多模式实时视频编码功能和性能要求的基础上,该文提出的方法与传统DVFS技术相比,单帧视频编码的平均功耗节省了11.4%.
