Scholarlink

<p>Automatic failure diagnosis is critical for large-scale microservice systems. Most existing failure detection methods rely solely on single-modal data, such as logs, traces, or metrics. This study conducts an empirical analysis using actual failure scenarios to demonstrate that integrating multiple data sources (multimodal data) leads to a more precise diagnosis. However, effectively representing these data and handling unequal failures remain challenging. The proposed approach, MD-RFD-MS-IGNN, introduces Multimodal Data for Robust Failure Diagnosis of Microservice Systems using an Optimized Intention-Adaptive Graph Neural Network. First, the GAIA dataset is used to collect input data. To implement this, the input data is pre-processed using Adaptive Kernel Learning Kalman Filtering (AKLKF), which removes N/A (not applicable) values or empty rows from the collected data. Then, the pre-processed data undergoes feature extraction using Automated Tunable Q Wavelet Transform (ATQWT) to extract spatial features such as traces, logs, and metrics. Subsequently, the extracted data is fed into an Intention-Adaptive Graph Neural Network (IGNN) to efficiently classify failure types such as file missing, system stuck, process crash, and access refused. Generally, optimization algorithms that can be adapted to obtain optimal parameters for accurate failure classification are not incorporated within IGNN. To address this, Human Evolutionary Optimization (HEO) is employed to optimize IGNN, ensuring precise identification of microservice system failures. Next, the proposed MD-RFD-MS-IGNN approach is implemented, and performance metrics including F1-Score, Precision, and Recall are evaluated. When compared to existing techniques such as Automated Functional and Robustness Testing of Microservice Architectures (AFRT-MA-GCM), MTG_CD: Multi-Scale Learnable Transformation Graph for Fault Classification and Diagnosis in Microservices (MTG-FCDM-GCN), and Robust Failure Diagnosis of Microservice Systems through Multimodal Data (RFD-MS-MD-GNN), the MD-RFD-MS-IGNN approach achieves 17.30, 23.39, and 32.41% higher F1-Score, respectively.</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

Optimized intention-adaptive graph neural network for robust failure diagnosis of microservice system using multimodal data

  • 期刊论文

  • 发表日期：

    2025年8月30日

查看全文
  • N. Naveen Kumar,
  • S. Suresh,
  • S. Balamurugan,
  • P. Seshu Kumar,
  • R. Maruthamuthu,
  • P. P. Devi,
  • Jude Moses Anto Devakanth

摘要

Automatic failure diagnosis is critical for large-scale microservice systems. Most existing failure detection methods rely solely on single-modal data, such as logs, traces, or metrics. This study conducts an empirical analysis using actual failure scenarios to demonstrate that integrating multiple data sources (multimodal data) leads to a more precise diagnosis. However, effectively representing these data and handling unequal failures remain challenging. The proposed approach, MD-RFD-MS-IGNN, introduces Multimodal Data for Robust Failure Diagnosis of Microservice Systems using an Optimized Intention-Adaptive Graph Neural Network. First, the GAIA dataset is used to collect input data. To implement this, the input data is pre-processed using Adaptive Kernel Learning Kalman Filtering (AKLKF), which removes N/A (not applicable) values or empty rows from the collected data. Then, the pre-processed data undergoes feature extraction using Automated Tunable Q Wavelet Transform (ATQWT) to extract spatial features such as traces, logs, and metrics. Subsequently, the extracted data is fed into an Intention-Adaptive Graph Neural Network (IGNN) to efficiently classify failure types such as file missing, system stuck, process crash, and access refused. Generally, optimization algorithms that can be adapted to obtain optimal parameters for accurate failure classification are not incorporated within IGNN. To address this, Human Evolutionary Optimization (HEO) is employed to optimize IGNN, ensuring precise identification of microservice system failures. Next, the proposed MD-RFD-MS-IGNN approach is implemented, and performance metrics including F1-Score, Precision, and Recall are evaluated. When compared to existing techniques such as Automated Functional and Robustness Testing of Microservice Architectures (AFRT-MA-GCM), MTG_CD: Multi-Scale Learnable Transformation Graph for Fault Classification and Diagnosis in Microservices (MTG-FCDM-GCN), and Robust Failure Diagnosis of Microservice Systems through Multimodal Data (RFD-MS-MD-GNN), the MD-RFD-MS-IGNN approach achieves 17.30, 23.39, and 32.41% higher F1-Score, respectively.