Scholarlink

<p>Cloud computing environments must dynamically manage resources to ensure optimal performance and meet Quality of Service (QoS) and Service Level Agreement (SLA) requirements. One of the major challenges is the early detection of overload conditions that can lead to service degradation and task rejection. Traditional reactive methods and static thresholds often fail to effectively anticipate these events. While queuing theory provides an analytical foundation for modeling system behavior, machine learning offers data-driven adaptability. In this study, our primary objective is to predict resource overload in cloud computing systems. To this end, we propose a hybrid model that integrates an M/M/m/K queuing system with a machine learning classifier. The framework leverages queuing-theoretic metrics computed over the observation window and combines them with system resource statistics. We utilize the Google Cluster Trace dataset as a real-world benchmark and systematically extract all relevant queue-derived features. To identify the most effective machine learning classifier within the hybrid architecture, we conduct an extensive evaluation involving several classifiers. The results show that Random Forest and Decision Tree outperform the other methods in terms of accuracy, precision, recall, and F1-score. Based on these results, we select Random Forest for the remainder of our study, where it is used for more detailed performance assessment and statistical evaluation. We perform statistical analyses (ANOVA and Welch’s t-tests) to confirm the robustness and reliability of the hybrid model under varying data volumes. Furthermore, we conduct a comparative study between the proposed hybrid framework and purely AI-based models, evaluating each with the same set of classifiers. The comparison demonstrates that the hybrid approach achieves superior performance, particularly in scenarios characterized by workload variability and uncertainty. Feature importance analysis confirms the significant contribution of queueing-theoretic metrics to predictive performance. Overall, the proposed hybrid model proves effective, interpretable, and adaptable for scalable cloud resource management, enabling proactive detection of overload conditions and improved QoS in dynamic environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Scalable overload prediction in cloud computing using a hybrid queuing-theoretic and machine learning framework

  • 期刊论文

  • 发表日期：

    2025年11月18日

查看全文
  • Oumaima Ghandour,
  • Said El Kafhali,
  • Iman El Mir

摘要

Cloud computing environments must dynamically manage resources to ensure optimal performance and meet Quality of Service (QoS) and Service Level Agreement (SLA) requirements. One of the major challenges is the early detection of overload conditions that can lead to service degradation and task rejection. Traditional reactive methods and static thresholds often fail to effectively anticipate these events. While queuing theory provides an analytical foundation for modeling system behavior, machine learning offers data-driven adaptability. In this study, our primary objective is to predict resource overload in cloud computing systems. To this end, we propose a hybrid model that integrates an M/M/m/K queuing system with a machine learning classifier. The framework leverages queuing-theoretic metrics computed over the observation window and combines them with system resource statistics. We utilize the Google Cluster Trace dataset as a real-world benchmark and systematically extract all relevant queue-derived features. To identify the most effective machine learning classifier within the hybrid architecture, we conduct an extensive evaluation involving several classifiers. The results show that Random Forest and Decision Tree outperform the other methods in terms of accuracy, precision, recall, and F1-score. Based on these results, we select Random Forest for the remainder of our study, where it is used for more detailed performance assessment and statistical evaluation. We perform statistical analyses (ANOVA and Welch’s t-tests) to confirm the robustness and reliability of the hybrid model under varying data volumes. Furthermore, we conduct a comparative study between the proposed hybrid framework and purely AI-based models, evaluating each with the same set of classifiers. The comparison demonstrates that the hybrid approach achieves superior performance, particularly in scenarios characterized by workload variability and uncertainty. Feature importance analysis confirms the significant contribution of queueing-theoretic metrics to predictive performance. Overall, the proposed hybrid model proves effective, interpretable, and adaptable for scalable cloud resource management, enabling proactive detection of overload conditions and improved QoS in dynamic environments.