Energy Saving and Scavenging in Stand-alone and Large Scale Distributed Systems.

Energy management;Embedded system;Data center;Electrical Engineering;Engineering;Electrical Engineering

He, XuejingMars, Jason ;

University of Michigan

关键词: Energy management;    Embedded system;    Data center;    Electrical Engineering;    Engineering;    Electrical Engineering;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/116746/xjhe_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This thesis focuses on energy management techniques for distributed systems such as hand-held mobile devices, sensor nodes, and data center servers. One of the major design problems in multiple application domains is the mismatch between workloads and resources. Sub-optimal assignment of workloads to resources can cause underloaded or overloaded resources, resulting in performance degradation or energy waste. This work specifically focuses on the heterogeneity in system hardware components and workloads. It includes energy management solutions for unregulated or batteryless embedded systems; and data center servers with heterogeneous workloads, machines, and processor wear states. This thesis describes four major contributions: (1) This thesis describes a battery test and energy delivery system design process to maintain battery life in embedded systems without voltage regulators. (2) In battery-less sensor nodes, this thesis demonstrates a routing protocol to maintain reliable transmission through the sensor network. (3) This thesis has characterized typical workloads and developed two models to capture the heterogeneity of data center tasks and machines: a task performance model and a machine resource utilization model. These models allow users to predict task finish time on individual machines. It then integrates these two models into a task scheduler based on the Hadoop framework for MapReduce tasks, and uses this scheduler for server energy minimization using task concentration. (4) In addition to saving server energy consumption, this thesis describes a method of reducing data center cooling energy by maintaining optimal server processor temperature setpoints through a task assignment algorithm. This algorithm considers the reliability impact of processor wear states. It records processor wear states through automatic timing slack tests on a cluster of machines with varying core temperatures, voltages, and frequencies. These optimal temperature setpoints are used in a task scheduling algorithm that saves both server and cooling energy.