With the rapid growth of demand for data center services, the energy and water use of data centers has become a critical concern in the contexts of energy use, climate change, and freshwater conservation. Therefore, understanding, quantifying, and optimizing the use of energy and water resources in data centers has become an important task for data center designers, engineers, operators, and policymakers, because it provides them with useful insights on resource-efficient data center design, sustainable operations, and effective policy formulation and technology incentives. However, current models developed by the energy analysis community fail to provide a reliable method to quantify data center energy and water usage and performance under different climatic conditions, system and technology implementations, and operating practices. These models also do not provide a robust approach to optimize the energy and water usage and performance in different aspects of data center operations, including data center workload management, information technology equipment operation, and infrastructure system operation. Addressing these challenges requires an accurate and generalizable computational modelling framework that can incorporate multi-dimensional and correlated spatio-temporal (e.g., climates, power grid mixes, computing tasks), technological (e.g., server configurations, cooling system types, economizer usage), and operational (e.g., workload levels, equipment efficiencies, environment setpoints) factors that affect data center energy and water use. Accordingly, a hybrid physics-based and data-driven modelling framework was developed in this research for the quantification and optimization of resource usage and performance in data centers.The developed modelling framework consists of four sub-models, which were elaborated in this dissertation for model demonstration. Section 4 (case study 1) describes a statistical and thermodynamics-based model for predictive analysis of data center power usage effectiveness (PUE), which was validated by publicly-available data from hyperscale data centers (including Google and Facebook). Section 5 (case study 2) describes a thermodynamics-based model for simultaneous simulations of PUE and water usage effectiveness (WUE) of data centers, validated by data from real data center operations, and was used to study the energy and water use performance of ten data center archetypes in 15 climate zones in the U.S. Section 6 (case study 3) describes an integrated decision-making framework for multi-objective optimization of carbon-, water-, and economic-intelligent data center workload scheduling, demonstrated with a case study data center in California. Section 7 discusses data-driven models for predictive analysis of server power use, power-performance, and server throughput in data centers, validated using the open-source SPECpower_ssj2008 database. In general, the modelling framework is validated for its accuracy and generalizability and can enable predictive decision makings at both the regional and facility levels. It can be a valuable tool to energy analysts who model data center energy use, and to data center designers, engineers, operators, and policymakers, which can enable robust quantitative decision-makings to reduce the resource consumption and optimize the resource use performance of data centers.

Creator

• Lei, Nuoa

DOI

• https://doi.org/10.21985/n2-b98v-n543

Subject

• Statistics
• Sustainability
• Energy

Language

• en

Alternate Identifier

• http://dissertations.umi.com/northwestern:15929
• etdadmin_upload_879204

Keyword

• Data Center
• Data-Driven Modeling
• Thermodynamics
• Information Technology
• Energy-Water Nexus
• Energy and Sustainability

Date created

• 2022-01-01

Resource type

• Dissertation

Rights statement

• In Copyright