Analysis of evolutionary multi-objective algorithms for data center electrical systems

• Published in: Computing Vol. 107; no. 2; p. 65
• Main Authors: Monte Sousa, Francisco, Callou, Gustavo
• Format: Journal Article
• Language: English
• Published: Wien Springer Nature B.V 01.02.2025
• Subjects: Availability; Block diagrams; Computer centers; Data centers; Data storage; Demand analysis; Design optimization; Energy efficiency; Energy flow; Evolutionary algorithms; Genetic algorithms; Multiple objective analysis; Optimization; Particle swarm optimization; Petri nets; Sorting algorithms
• ISSN: 0010-485X; 1436-5057
• DOI: 10.1007/s00607-025-01416-z

Nowadays, the demand for data storage is increasing due to various factors, such as educational needs, social media, and streaming services. This drives the need for robust data center infrastructures characterized by high availability, low cost, and high energy efficiency. However, achieving the optimization of such objectives is often challenging due to their conflicting characteristics. To enhance the design of electrical data center architectures, this paper uses two optimization strategies based on evolutionary algorithms: the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and Multi-Objective Particle Swarm Optimization (MOPSO). The primary aim is to optimize conflicting metrics such as availability, cost, and energy efficiency of data center electrical infrastructures. This study proposes models using stochastic Petri nets (SPNs), reliability block diagrams (RBDs), and energy flow models (EFMs). A key objective of this research is to reduce the execution time required to obtain results by implementing an optimization strategy, as opposed to the brute force method that evaluates all possible outcomes. A case study is conducted to demonstrate the applicability of the proposed approach. The results indicate that a significant reduction in execution time can be achieved, producing results that are near-optimal within a shorter execution time demanded.