A multi-objective optimization method for cloud data center resource management based on a repairable queueing system

Zhao Tao

doi:10.1016/j.ress.2026.113072

Return

A multi-objective optimization method for cloud data center resource management based on a repairable queueing system

2026-06-23

PRE

DOI：10.1016/j.ress.2026.113072

Original

Original request for help

Save

Abstract

En 中文

Cloud data centers (CDCs) must coordinate computing resources and recovery resources to sustain service performance while controlling energy consumption and operating cost. However, existing CDC resource management studies often overlook how failure-recovery dynamics alter effective service capacity or treat recovery capability as a fixed parameter in optimization. This paper addresses this limitation through an analytical modeling and multi-objective optimization framework based on a repairable queueing system. An RQS-based response time model is first developed by embedding processor failure-recovery dynamics into request processing, from which the mean response times of a single node and the entire CDC IT architecture are derived. Mean energy consumption and system cost models are then constructed under the same CDC configuration, and a constrained three-objective optimization model is formulated with configurable processor numbers and recovery-rate variables. NSGA-II is further combined with CCR- and TOPSIS-based methods to screen Pareto solutions and identify representative strategies under different operational priorities. Numerical results show that the proposed model matches the response time accuracy of the SPN-based and DES-based baselines while reducing state-space size and computation time, and further reveal the trade-offs among response time, energy consumption, and system cost under constant-workload, variable-workload, and DVFS-enabled scenarios.

Keywords:

Cloud data centers

Repairable queueing system

Response time

Energy consumption

System cost

Fault-tolerant operation

Multi-objective optimization