Reinforcement Learning Based Load Balancing for Geographically Distributed Data centres

Mackie, Max; Sun, Hongjian; Jiang, Jing

Reinforcement Learning Based Load Balancing for Geographically Distributed Data centres

Mackie, Max; Sun, Hongjian; Jiang, Jing

Authors

Max Mackie

Professor Hongjian Sun hongjian.sun@durham.ac.uk
Professor

Jing Jiang

Abstract

This paper proposes a method of migrating workload among geo-distributed data centres that are equipped with on-site renewable energy sources (RES), such as solar and wind energy, to decarbonise data centres. It aims to optimise the performance of such a system by introducing a tunable Reinforcement Learning (RL) based load-balancing algorithm that implements a Neural Network to intelligently migrate workload. By migrating workload within the network of geo-distributed data centres, spatial variations in electricity price and intermittent RES can be capitalised upon to enhance data centres’ operations. The proposed algorithm is evaluated by running simulations using real-world data traces. It is found that the proposed algorithm is able to reduce costs by 6.1% whilst also increasing the utilisation of RES by 10.7%.

Citation

Mackie, M., Sun, H., & Jiang, J. (2021). Reinforcement Learning Based Load Balancing for Geographically Distributed Data centres.

Conference Name	IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe) 2021
Conference Location	Espoo, Finland
Start Date	Oct 18, 2021
End Date	Oct 21, 2021
Acceptance Date	Jul 20, 2021
Publication Date	2021-10
Deposit Date	Jul 20, 2021
Publicly Available Date	Oct 22, 2021
Publisher URL	https://easychair.org/smart-program/ISGT-Europe2021/index.html
Additional Information	Conference dates: 18-21 October 2021

Files

Accepted Conference Proceeding (316 Kb)
PDF

Copyright Statement
© 2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.