CSR: Small: Green Farms: Towards a Stable Energy Optimization Architecture for Data Centers

点击翻译按钮获取中文摘要

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).This project develops techniques for increasing energy-efficiency of modern data centers with performance constraints. Both computing and cooling energy are considered. According to the U.S. Environmental Protection Agency, data centers in the United States incur an annual energy cost of approximately $4.5 billion, which is comparable to the total consumption of 5.8 million average US households. In the absence of intervention, data center consumption is expected to double in five years. Up to 80% of this projected energy expenditure is avoidable, which constitutes a prospective reduction in nationwide carbon dioxide emissions by up to 47 million metric tons (MMT) per year; an amount comparable to the annual carbon emissions footprint of all fuel combustion in a small nation. The observation motivating the energy-efficiency solutions developed in this project is that the proliferation of individual energy-saving mechanisms in server installations can lead to increasingly suboptimal overall energy management. The problem lies in performance composability or lack thereof; a challenge that arises because individual optimizations generally do not compose well when combined, leading to opportunities for improvement. A theory is developed to define preconditions of composability and a holistic distributed approach is investigated for coordinating energy-saving decisions across multiple components that span both the computing and cooling subsystems. A graduate course on cyber-physical systems is designed to cover the addressed challenges and solutions. Women and minority students are encouraged to benefit from these opportunities.