Data Center Energy Inefficiency Root Cause and Sensitivity Analysis: The case of Ethio-Telecom Legehar Data Center
| dc.contributor.advisor | Dereje, Hailemariam (PhD) | |
| dc.contributor.author | Zerihun, Tesfaye | |
| dc.date.accessioned | 2022-02-03T08:34:39Z | |
| dc.date.accessioned | 2023-11-04T15:13:18Z | |
| dc.date.available | 2022-02-03T08:34:39Z | |
| dc.date.available | 2023-11-04T15:13:18Z | |
| dc.date.issued | 2021-11 | |
| dc.description.abstract | Data centers are the cornerstone of today's information age. As more people use information technology (IT), the volume of data processed and stored in the data centers rises. As a result of this growth, data center energy consumption has grown for both actual works (data processing and storage) and supporting infrastructure. A data center subsystem is classified as mission-critical and support infrastructure. The mission-critical parts are IT equipment (e.g. servers, routers, switches, and storage systems), whereas support infrastructure parts include mechanical and electrical components such as backup power supplies, Uninterrupted Power Supplies (UPS), Power Distribution Units (PDU), and cooling systems. The Power Usage Effectiveness (PUE) metric is the global standard for data center energy efficiency, and it is defined as the ratio of total power delivered to the data center to actual IT equipment energy usage. The current overall average PUE value for the Ethio-Telecom Legehar data center is 2.34, indicating a considerable gap between the power supplied to the data center and the actual energy consumption of IT equipment. The PUE was calculated using data collected for 37-week (nine months) period from the Ethio-Telecom power and environmental monitoring system (NetEco). This study addresses the problem of energy inefficiency by finding the fundamental cause of the problem using a combination of machine learning and Global Sensitivity Analysis (GSA) techniques. First, a machine learning technique was used to identify important features using the Random Forest Regression (RFR). Second, the Sobol-GSA technique is used to quantify the impact level of the selected features on PUE. Sobol-GSA consists of two scenarios: - main effect (first-order) indices, individual variables' contributions with PUE, and total order indices interactions between variables, i.e., the sum of first indices and higher indices. It was discovered that UPS efficiency and cooling systems are major factors in the energy efficiency problem. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/29886 | |
| dc.language.iso | en_US | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | Data center | en_US |
| dc.subject | PUE | en_US |
| dc.subject | Sobol-GSA | en_US |
| dc.subject | Random Forest Regression | en_US |
| dc.subject | NetEco | en_US |
| dc.title | Data Center Energy Inefficiency Root Cause and Sensitivity Analysis: The case of Ethio-Telecom Legehar Data Center | en_US |
| dc.type | Thesis | en_US |