CRII: CPS: Building Highly-efficient and Low-power Edge Computing with Data-driven Learning and Control

物联网（IoT）被描述为小型物理设备的网络，嵌入了传感器、软件和其他技术，可以轻松地通过互联网与其他设备和系统交换数据。在智能家居自动化和智能城市的强劲需求和快速增长的推动下，无线网络、控制系统和自动化等传统技术与小型化和低功耗设备的融合促进了物联网的发展。可负担得起的可互操作物联网系统在日常生活中越来越普遍。这些物联网设备紧密合作，协调一系列任务，越来越多地用于家庭和办公室的可编程个性化加热、冷却和安全控制等活动。随着这些物联网设备变得越来越强大，这些设备可以单独或作为本地分布式网络组合执行更多计算要求高的任务，从而使计算更接近需要提高响应能力的位置，即在互联网的边缘。挑战在于确保物联网设备与边缘计算甚至云服务作为一个高效的有目的系统的高性能、及时性能和无缝集体运行。本项目研究各种边缘和物联网系统中系统资源利用率与能源效率之间的关系，以便更好地了解如何优化边缘计算系统的关键性能参数。本项目探索通过数据驱动的方法减轻边缘系统的低效率。具体而言，主要研究方向包括：(1)分析不同边缘系统的功耗低效率，并为运行时边缘和物联网应用开发数据驱动的能量感知框架；(2)裁剪边缘运行时框架，包括部分数据和控制平面，以揭示隐藏的依赖关系；(3)在高保真的现实测试场景中扩展和评估该框架和方法。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The Internet of Things (IoT) is described as networks of small physical devices, embedded with sensors, software, and other technologies, that easily exchange data with other devices and systems over the Internet. The convergence of traditional technologies from wireless networking, control systems, and automation with miniaturization and low-powered devices contributed to the development of IoT, spurred on by strong demand and rapid growth in smart home automation and smart cities. Affordable interoperable IoT systems are increasingly ubiquitous in daily life. These IoT devices, working closely together, orchestrate a range of tasks, increasingly used for such activities as programmable personalized control of heating, cooling, and security in homes and offices. As these IoT devices become more capable, more computationally demanding tasks can be performed by these devices singly or in combination as a local distributed network bringing computing closer to the location where needed to improve responsiveness, i.e., at the edges of the Internet. The challenge is to ensure the highly capable, timely performance, seamless collective operation of IoT devices with edge computing and even cloud services as an efficient purposeful system.This project studies the relationships between system resource utilization and energy efficiency in various edge and IoT systems in order to better understand how to optimize the key performance parameters of edge computing systems. This project explores mitigating the inefficiency in edge systems through a data-driven approach. Specifically, the primary research directions include: (1) analyzing the power inefficiency in different edge systems and develop a data-driven energy-aware framework for runtime edge and IoT applications, (2) tailoring the edge runtime framework including parts of data and control planes to reveal hidden dependencies, and (3) scaling and evaluating this framework and methodology in high-fidelity realistic test scenarios.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.