Scholarlink

<p>Microservice Architectures demand continuous runtime adaptability to cope with fluctuating workloads, performance bottlenecks, and evolving quality of service requirements. However, existing solutions often focus on design-time optimizations, lacking real-time adaptation capabilities grounded in empirically validated principles (coupling, cohesion, and granularity). Based on triadic evaluation via quantitative experimentation and statistical analysis of empirical practitioner insights, this study introduces AMOF (Adaptive Microservice Optimization Framework), a runtime engine that combines a Random Forest Classifier (RFC) for rapid adaptation triggers with a Deep Q-Network (DQN) for learning long-term strategies under a Markov Decision Process (MDP) formulation. The framework formalizes coupling, cohesion, and granularity metrics as adaptation criteria and quantifies rewards to guide decisions. Empirical evaluation on Google’s Online Boutique, a realistic benchmark validates the approach’s ability to improve system-level objectives under various runtime scenarios. The results indicate the superiority of AMOF over static, HPA, and rule-based adaptation techniques 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

Toward intelligent runtime adaptation in microservices: Empirical trade-off analysis and a hybrid AI framework

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Syed Fakhar Abbas,
  • Musharif Ahmed,
  • Asim Bakhshi,
  • Sadaf Manzoor

摘要

Microservice Architectures demand continuous runtime adaptability to cope with fluctuating workloads, performance bottlenecks, and evolving quality of service requirements. However, existing solutions often focus on design-time optimizations, lacking real-time adaptation capabilities grounded in empirically validated principles (coupling, cohesion, and granularity). Based on triadic evaluation via quantitative experimentation and statistical analysis of empirical practitioner insights, this study introduces AMOF (Adaptive Microservice Optimization Framework), a runtime engine that combines a Random Forest Classifier (RFC) for rapid adaptation triggers with a Deep Q-Network (DQN) for learning long-term strategies under a Markov Decision Process (MDP) formulation. The framework formalizes coupling, cohesion, and granularity metrics as adaptation criteria and quantifies rewards to guide decisions. Empirical evaluation on Google’s Online Boutique, a realistic benchmark validates the approach’s ability to improve system-level objectives under various runtime scenarios. The results indicate the superiority of AMOF over static, HPA, and rule-based adaptation techniques in dynamic environments.