Design of Intelligent Controllers to Maximize Efficiency and Guarantee Security, Privacy, and Fairness in Heterogeneous Computing and Sensing Systems

今天的技术用户通常拥有多台计算机并携带多个移动设备（例如，笔记本电脑、智能手表、电话和平板电脑），但其中许多设备在大多数时间都处于空闲状态。尽管硬件技术取得了进步，但用户拥有的计算和感知能力无法满足他们想要运行的某些应用程序的要求（例如，人脸检测、语音识别和虚拟个人评估等认知应用程序）。这种差异促使许多用户要么购买更多设备，要么从基础设施提供商那里租用更多容量，以应对他们的高峰需求。虽然这些方法在实践中是有效的，但它们通常会带来额外的成本和严重的效率低下（例如，能源和电力）。在这个研究项目中，我们采用了不同的方法。我们建议开发工具和方法，使用户能够共享他们的计算和传感设备，以形成一个更有能力的共享系统。此外，我们建议开发管理共享系统的方法，以提高其性能和效率。该发现基金将资助一项前沿研究项目，旨在推动共享系统中智能决策的极限，同时解决提供安全、隐私和公平保证方面的主要挑战。该项目将培训至少10名HQP，为他们提供先进的工程理论方面，以及工业规模基础设施和工作量的实践和动手经验。虽然概述的研究重点是共享异构系统，但许多其他应用可以从提议的计划中受益，包括多处理器技术、无线和传感器网络、自组织应急移动系统和高性能数据中心。根据世邦魏理仕最近的一份报告，2017年北美数据中心投资额达到200亿美元。数据中心消耗大量能源，排放的二氧化碳与航空业一样多。根据自然资源保护委员会（NRDC）的数据，95%的数据中心电力消耗效率低下。这主要是因为服务器的利用率很低。本研究项目开发的方法和工具可直接用于提高服务器利用率和提高数据中心的能源效率。这样一来，该计划不仅可以节省数百万美元的能源成本，还可以保护环境免受进一步的二氧化碳污染。加拿大在高效大规模计算方面的创新和HQP培训将有助于保持加拿大在信息技术方面的世界领先地位。

Today's technology users often own multiple computers and carry multiple mobile devices (e.g., laptops, smart watches, phones, and tablets), but many of these devices are idle most of the time. Despite advances in hardware technologies, the computing and sensing capacity that users own cannot meet the requirements of some of the applications they want to run (e.g., cognitive applications such as face detection, speech recognition, and virtual personal assessment). This discrepancy has pushed many users to either buy more devices or rent more capacity from infrastructure providers to cope with their peak demand. Although these approaches have worked in practice, they usually impose extra costs and major inefficiencies (e.g., energy and power). In this research program, we pursue a different approach. We propose to develop tools and methods that enable users to share their computing and sensing devices to form a more capable, shared system. Furthermore, we propose to develop methods to manage the shared system and improve its performance and efficiency. This Discovery grant will fund a cutting-edge research program designed to push the limits of intelligent decision making in shared systems, while addressing major challenges in providing security, privacy, and fairness assurances. The program will train at least ten HQP, providing them with advanced theoretical aspects of engineering, and practical and hands-on experience on industry scale infrastructures and workloads. While the outlined research focuses on shared heterogeneous systems, many other applications could benefit from the proposed program, including multiprocessor technologies, wireless and sensor networks, ad-hoc emergency mobile systems, and high-performance data centres. According to a recent CBRE report, the data centre investment volume reached $20 billion in North America during 2017. Data centres consume a significant amount of energy and emit the same amount of CO2 as the airline industry. According to the Natural Resources Defence Council (NRDC), 95% of data centre electricity is consumed inefficiently. This is mainly because of servers' low utilization. The methods and tools developed in this research program could be directly used to increase utilization of servers and improve data centres' energy efficiency. In doing so, the program not only saves potentially millions of dollars in energy costs, but also protects the environment from further CO2 pollution. Canadian innovation and HQP training in efficient large-scale computing will contribute to maintaining Canada's role as a world leader in information technologies.