A Deep Reinforcement Learning Approach to Cloud Resource Optimization with Response Time Distributions

Date:2025-07-30 clicks:

Impact Factor:7.5

Affiliation of Author(s):福建理工大学计算机科学与数学学院

Journal:Expert Systems with Applications

Abstract:Cloud computing systems depend on elasticity to adapt resource allocation to fluctuating workload. Although traditional metrics effectively measure resource scaling, they fail to capture how these adjustments impact user-perceived service quality, which is a critical gap for providers and consumers alike. To bridge this gap, we introduce a novel performance metric that uses the probability distribution of task response times to complement the existing elasticity measures. This metric defines service quality as the likelihood that response times meet preset service-level objectives (SLOs) within a given timeframe. We developed a framework linking resource allocation, workload patterns, and this metric to optimize performance in various scenarios. We propose a decision-making algorithm to improve service quality without sacrificing cost efficiency. The experiments show that integrating this user-focused metric improves resource utilization by 23 % and reduces SLO violations by 31 % in the tested e-commerce workloads.

Indexed by:Journal paper

Discipline:Engineering

Volume:296

Issue:2026-01

Page Number:129081

Translation or Not:no

Date of Publication:2025-07-22

Included Journals:SCI

Links to published journals:https://www-sciencedirect-com-443.webvpn.fjut.edu.cn/science/article/abs/pii/S0957417425026983

First Author:Liwen Chen

Co-author:Chunmao Jiang,Qiaoping Zhong

Pre One:Federated learning with three-way decisions for privacy-preserving multicloud resource scheduling Next One:DF-RL: A Dynamic Fuzzy-Neuro Reinforcement Learning Framework for Cloud Resource Management