CAREER: Maximizing Energy Efficiency with Statistical Performance and Skin Temperature Quality of Service Guarantee for Handheld Platforms

职业：通过手持平台的统计性能和表面温度服务质量保证最大限度地提高能源效率

• 批准号: 1652132
• 负责人: Stephanie Forrest
• 金额: $ 45万
• 依托单位: Arizona State University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-02-01 至 2022-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1652132&HistoricalAwards=false
• 关键词: CAREER; Maximizing; Energy; Efficiency; Statistical

Smartphones reached almost 4 billion world-wide subscriptions in 2015 and have become our daily companion providing both high capacity computing, as well as personalized computing. For smartphones, user satisfaction determines the success or failure for a particular platform. The top ranked factor of user satisfaction is performance, which can be experienced by users through computation performance and battery performance. Therefore, smartphone designs must achieve balance between performance, temperature and energy management in a coordinated manner to maximize user satisfaction. Performance as measured by mobile application execution time has long been assumed to be a deterministic quantity. In reality, execution times vary substantially, depending on the characteristics of data inputs and the varying states of the system. Furthermore, the device surface temperature is a unique constraint for handheld devices. It is tightly coupled with the location of the major heat-generating source within a smartphone. Thus, a smartphone's energy efficiency is intricately related to both performance quality and skin temperature management, making optimization for system energy efficiency a complex task for battery-powered platforms. With the prevalence of portable electronics, the research outcome has a profound impact on the advancement of the relevant research domains and on society. The research agenda is complemented by an education agenda focusing on the design of handheld platforms, such as smartphones and other high-performance wearable electronics. The educational agenda includes (1) new graduate and undergraduate curricula that incorporates processor and handheld temperature and energy management techniques, (2) enhancing computer architecture and mobile computing courses through lab activities on the proposed skin temperature management for mobile devices, (3) mentoring undergraduate and graduate students in research, and (4) attracting and retaining underrepresented groups of students in STEM fields.This research tackles the problem of performance, temperature and energy efficiency co-optimization from the vantage point of managing the hardware resources. It builds on the PI?s prior work in system and hardware architecture by proposing an optimizing user satisfaction (OPUS) framework for holistic management and coordination of performance quality, skin temperature, and energy efficiency for handhelds. Through accurate execution time models, OPUS dynamically adjusts the mobile platforms to meet the different quality of service goals. The research investigates the dynamic voltage-frequency scaling and temperature-aware computation acceleration algorithms, as well as emerging dynamic cooling mechanisms, suitable for handheld platforms. Optimizing for performance, energy efficiency, or temperature is not new, but optimizing devices in the context of smartphone user satisfaction gives rise to a set of opportunities that are non-existent in conventional computing platforms. The findings can serve as foundations for future user satisfaction optimization research for handheld platforms.

智能手机在2015年达到了近40亿的全球订阅量，并已成为我们的日常伴侣，提供高容量计算以及个性化计算。对于智能手机来说，用户满意度决定了特定平台的成功或失败。用户满意度的首要因素是性能，用户可以通过计算性能和电池性能来体验性能。因此，智能手机设计必须以协调的方式实现性能、温度和能源管理之间的平衡，以最大限度地提高用户满意度。由移动的应用程序执行时间衡量的性能长期以来一直被认为是一个确定性的量。实际上，执行时间会根据数据输入的特性和系统的不同状态而发生很大变化。此外，设备表面温度是手持设备的独特约束。它与智能手机内主要发热源的位置紧密相连。因此，智能手机的能效与性能质量和皮肤温度管理密切相关，使得系统能效优化成为电池供电平台的复杂任务。随着便携式电子产品的普及，其研究成果对相关研究领域的进步和社会产生了深远的影响。除了研究议程之外，还制定了一项教育议程，重点关注手持平台的设计，如智能手机和其他高性能可穿戴电子产品。教育议程包括（1）新的研究生和本科生课程，包括处理器和手持温度和能量管理技术，（2）通过关于移动的设备的拟议皮肤温度管理的实验室活动来增强计算机体系结构和移动的计算课程，（3）指导本科生和研究生进行研究，（4）吸引和留住STEM领域中代表性不足的学生群体。本研究从管理硬件资源的Vantage解决性能、温度和能效协同优化问题。它建立在PI的基础上？他之前在系统和硬件架构方面的工作，提出了一个优化用户满意度（OPUS）框架，用于整体管理和协调手持设备的性能质量、皮肤温度和能源效率。通过精确的执行时间模型，OPUS动态调整移动的平台，以满足不同的服务质量目标。该研究调查了动态电压频率缩放和温度感知计算加速算法，以及新兴的动态冷却机制，适用于手持平台。针对性能、能效或温度进行优化并不新鲜，但在智能手机用户满意度的背景下优化设备会带来一系列传统计算平台中不存在的机会。研究结果可作为未来手持平台用户满意度优化研究的基础。

项目成果

Understanding the Power of Evolutionary Computation for GPU Code Optimization

了解用于 GPU 代码优化的进化计算的力量

• DOI: 10.1109/iiswc55918.2022.00025
• 发表时间: 2022
• 期刊: IEEE International Symposium on Workload Characterization (IISWC
• 作者: Liou, Jhe-Yu;Awan, Muaaz;Hofmeyr, Steven;Forrest, Stephanie;Wu, Carole-Jean
• 通讯作者: Wu, Carole-Jean

Understanding the thermal challenges of high-performance mobile devices with a detailed platform temperature model

通过详细的平台温度模型了解高性能移动设备的热挑战

• DOI: 10.1109/iiswc.2017.8167768
• 发表时间: 2017
• 期刊: 2017 IEEE International Symposium on Workload Characterization (IISWC
• 作者: Yu, Ying-Ju;Wu, Carole-Jean
• 通讯作者: Wu, Carole-Jean

EdgeWise: Energy-efficient CNN Computation on Edge Devices under Stochastic Communication Delays

• DOI: 10.1145/3530908
• 发表时间: 2022-04
• 期刊: ACM Transactions on Embedded Computing Systems (TECS)
• 作者: Mehdi Ghasemi;Daler N. Rakhmatov;Carole-Jean Wu;S. Vrudhula

GEVO-ML: a proposal for optimizing ML code with evolutionary computation

• DOI: 10.1145/3377929.3398139
• 发表时间: 2020-07
• 期刊: Proceedings of the 2020 Genetic and Evolutionary Computation Conference Companion
• 作者: Jhe-Yu Liou;Xiaodong Wang;S. Forrest;Carole-Jean Wu

AutoFL: Enabling Heterogeneity-Aware Energy Efficient Federated Learning

• DOI: 10.1145/3466752.3480129
• 发表时间: 2021-07
• 期刊: MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture
• 作者: Young Geun Kim;Carole-Jean Wu