Transient-fault-aware optimal configuration for cloud service providers: a composite queuing modeling approach

Journal of Cloud Computing

卷:15

Research

Siyi Chen¹  Wenhao Xu¹  Junyan Li¹  Haiyang Kuang¹  Huixian Huang¹ 

[1] Industrial Electronics Xiangtan University, 411105, Hunan, China

关键词: Cloud computing;    Multi-server system;    Server transient fault;    Server state switching;    Queuing theory;    Profit maximization;    Service completion rate;   

DOI  :  10.1186/s13677-026-00948-9

 received in 2025-10-21, accepted in 2026-06-14,  发布年份 2026

来源: Springer

【 摘 要 】

Effective capacity planning in cloud computing must balance profit maximization against the dual uncertainties of internal server faults and external traffic bursts. According to the supply-demand relationship between cloud service providers (CSPs) and customers, satisfying the interests of both is crucial for the sustained operation of the cloud computing system. Generally, due to the dynamic uncertainty of the cloud environment, server failures sometimes occur, resulting in the deterioration of system performance. Unfortunately, such a situation is currently not fully considered in cloud resource configuration. In this paper, our objective is to achieve the optimal configuration of resources in the cloud computing system in the presence of server transient faults, with the aim of maximizing profit and the service completion rate. By introducing queuing theory, we establish a multi-server queuing model to depict the service process of customer requests. Then, by considering the interactions among several server states and following the same methodology, the handling process of transient faults can also be defined as a specific queuing model. Combined with the analysis of the two models, the probability density function (pdf) of the generalized waiting time of customers is initially designed to investigate the quality of service. On this basis, the models of profit and service completion rate are constructed. Taking into account the potential for problem dimensionality scaling, two heuristic algorithms are employed to obtain the optimal solutions for such models, thereby simultaneously satisfying the requirements of both cloud service providers (CSPs) and customers. Furthermore, sensitivity analysis reveals that for both typical and high-failure cloud environments, the system favors a light-asset, elastic recovery strategy (characterized by minimal static redundancy) over heavy static buffering. Finally, trace-driven simulations using Google Cluster Data quantify the economic limits of static provisioning, demonstrating that handling extreme traffic bursts (\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$CV \gg 1$$\end{document}) purely via static capacity incurs a 60% resource wastage. These results position our model as a foundational tool for Base Capacity Planning, identifying the precise economic boundaries where dynamic scaling mechanisms must intervene for both fault tolerance and traffic bursts.

【 授权许可】

CC BY   
© The Author(s) 2026

【 图 表 】

Fig. 2

Fig. 1 Gene expression subtypes in myeloma cells from CASSIOPEIA RNA-seq cohort.

Fig. 4

Fig. 3

Algorithm 1

Fig. 5

Fig. 4

Fig. 10

Fig. 5

5

Fig. 1

Fig. 1

Fig. 3

Fig. 5

Fig. 5

Fig. 9

Fig. 1

Fig. 8

Algorithm 3