Scholarlink

<p>Cloud-native architectures, characterized by loosely coupled microservices, enable the modular execution of specific business functions, significantly accelerating software development and reducing update complexity. However, the execution of a single application often involves hundreds of interdependent microservice nodes, where complex invocation topologies and dynamic resource demands make efficient resource allocation a critical challenge. To address this issue, we propose Convolutional Neural Network-based Prediction and Resource Adaptation Model (CPRAM), a novel framework that integrates deep learning and reinforcement learning to optimize resource allocation in cloud-native environments. The key innovations of CPRAM are as follows: (1) We propose A&#xa0;CNN-based performance prediction model by analyzing resource usage and response times, even with sparse or incomplete data. (2) We allocate CPU and memory via a Deep Deterministic Policy Gradient (DDPG) based mechanism so as to maximize global resource utilization. We evaluate CPRAM using DeathStarBench, a widely adopted cloud-native microservices benchmark, and deploy our experiments on Docker Swarm. Extensive experimental results demonstrate that CPRAM outperforms state-of-the-art baselines, achieving significant resource savings of at least 22.8% in CPU and 21.91% in memory utilization while consistently meeting end-to-end QoS targets.</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

Performance prediction and resource adaptive adjustment for cloud-native microservices

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Yu Peng,
  • Jia Hao,
  • Yang Chen

摘要

Cloud-native architectures, characterized by loosely coupled microservices, enable the modular execution of specific business functions, significantly accelerating software development and reducing update complexity. However, the execution of a single application often involves hundreds of interdependent microservice nodes, where complex invocation topologies and dynamic resource demands make efficient resource allocation a critical challenge. To address this issue, we propose Convolutional Neural Network-based Prediction and Resource Adaptation Model (CPRAM), a novel framework that integrates deep learning and reinforcement learning to optimize resource allocation in cloud-native environments. The key innovations of CPRAM are as follows: (1) We propose A CNN-based performance prediction model by analyzing resource usage and response times, even with sparse or incomplete data. (2) We allocate CPU and memory via a Deep Deterministic Policy Gradient (DDPG) based mechanism so as to maximize global resource utilization. We evaluate CPRAM using DeathStarBench, a widely adopted cloud-native microservices benchmark, and deploy our experiments on Docker Swarm. Extensive experimental results demonstrate that CPRAM outperforms state-of-the-art baselines, achieving significant resource savings of at least 22.8% in CPU and 21.91% in memory utilization while consistently meeting end-to-end QoS targets.