CSR-DMSS, SM: Next-Generation Thermal-Aware, Energy-Efficient Resource Management for Data Centers

该项目解决了为异构高性能计算数据中心开发预测、自主热感知和节能任务调度算法的挑战。大规模数据中心中软件控制作业管理的节能问题尚未得到充分的探讨。现有的调度和电源管理算法本质上大多是被动的。本研究将探讨调度效率和生成时空任务调度（每个任务的开始时间及其计算节点分配或放置）的运行时复杂性之间的权衡，目的是开发可应用于现实世界的算法。本研究的新颖之处在于将数据中心视为一个网络物理实体，并以协调的方式主动配置和控制冷却系统和计算系统。以往的研究结果表明，除非根据计算负荷动态调整冷却，否则节能效果非常有限。本研究的结果将包括：A)研究调度算法的计算复杂性和效率之间的权衡，b)以统一方式产生任务、电源和冷却控制调度的主动算法，c)将包含计算设备的电源控制的软件架构，以及冷却系统的动态配置，d)来自实际数据中心的测试结果，显示所开发方案的适用性和实用性。结果将在IMAPCT实验室的网站（http://impact.asu.edu/）上公开，科学发现将包括在以网络物理为重点的多学科课程中。本研究的实际效益将在学术和企业数据中心得到证明。

This project addresses the challenges in developing predictive and autonomic thermal-aware and energy-efficient task scheduling algorithms for heterogeneous High-Performance Computing datacenters. Energy savings of software controlled job management in large-scale datacenters has not been adequately explored. Existing scheduling and power management algorithms are mostly reactive in nature. This research will investigate the trade-off between the schedule efficacy and the runtime complexity of producing a spatio-temporal task schedule (with start times of each task and its computing node assignment or placement), with an objective to develop algorithms that can be applied in real-world situations. Novelty of this research lies in viewing datacenters as a cyber-physical entity and proactively configuring and controlling both the cooling and computing systems in a coordinated manner. Previous research results show that energy savings are very limited unless the cooling is dynamically adjusted to the computing load. Outcomes of this research will include: a) study of the trade-off between computational complexity and efficiency of the scheduling algorithm, b) proactive algorithms that produce task, power and cooling control schedules in an unified manner, c) software architecture that will encompass the power control of the computing equipment, as well as dynamic configuration of the cooling systems, and d) test results from real-world data centers that show the applicability and practicality of the developed schemes. Results will be publicly available at IMAPCT Lab's website (http://impact.asu.edu/), and scientific findings will be included in multi-disciplinary courses with cyber-physical focus. Practical benefits of this research will be demonstrated in academic and corporate datacenters.