Artificial Intelligence and Machine Learning Approach to Measuring Energy Consumption in Data Centre Facilities

PhD Thesis

Jayathilake, S. 2019. Artificial Intelligence and Machine Learning Approach to Measuring Energy Consumption in Data Centre Facilities. PhD Thesis University of East London School of Architecture, Computing and Engineering
AuthorsJayathilake, S.
TypePhD Thesis

Data centres are at the heart of the modern digital world; However, at the same time it accounts for 10% of the world electricity supply. To improve energy efficiency, measuring energy consumption is an important step. However, it is a challenging task, especially in small to medium-sized data centres. Due to the setup of such facilities, it is not always feasible to measure the energy consumption (e.g. due to being positioned in mixed use buildings).
This research project addresses this problem by providing the tools and models to help estimate the energy consumption of data centres, with particular emphasis on smaller facilities. The work made two main novel contributions. First, energy models along with a web-based user-friendly tool were developed. The tool is capable of calculating energy consumption of each DC equipment type to then approximate the overall energy consumption of the facility. This tool is available for DC managers and operators. It uses publicly available benchmark data as input for the calculations.
One of the limitations of the first set of models is their reliance on pre-existing data for specific hardware. However, there are many ways hardware can be configured, meaning benchmark data was not available to all types of servers. As such, the second research contribution was the design of new machine learning algorithms capable of predicting energy consumption of servers based on a small number of features within 12% error rate. An open source software tool, MALEP, was also developed based on the machine learning algorithms to automate the prediction of the energy consumption of any
servers, irrespective of the presence of benchmark data. The software is made available open source under the GNU General Public Licence and downloadable from GitHub.
Although this work focused on servers which account for the largest part of energy consumption in data centres, in future, we hope to extend this work to create such models for storage and networking equipment.

PublisherUniversity of East London
Digital Object Identifier (DOI)
File Access Level
Publication dates
OnlineOct 2019
Publication process dates
Deposited05 Feb 2020
Permalink -

Download files

License: CC BY-NC-ND 4.0
File access level: Anyone

  • 422
    total views
  • 980
    total downloads
  • 8
    views this month
  • 6
    downloads this month

Export as