Skip to content

Latest commit

 

History

History
20 lines (15 loc) · 2.28 KB

2404.12135.md

File metadata and controls

20 lines (15 loc) · 2.28 KB
Raw

Background

  • Background Introducing mABC as a groundbreaking framework designed to revolutionize root cause analysis in cloud-native technology environments, particularly micro-services architectures. These architectures pose significant stability and efficiency challenges due to decomposing large applications into numerous independent service nodes focusing on specific business functions.

  • Existing Work Existing works like MicroScope, TraceAnomaly, and MEPFL struggle with cyclic dependencies and the need for high fault coverage through supervised data. While LLMs like GPT have shown significant abilities in RCA, they have limitations when it comes to cross-node failure analysis in complex micro-service architectures.

Core Contributions

  • Introduced mABC architecture

    • Challenge 1: Precision of RCA and formulation of solutions The proposed mABC framework, through the collaboration of seven specialized agents based on the intrinsic software knowledge of LLMs, ensures precision in RCA. To mitigate potential instability of LLMs and make the most of the transparent and egalitarian benefits of a decentralized structure, mABC employs a decision-making process inspired by blockchain governance principles.

    • Challenge 2: Efficient data collection and analysis With professional plugins and tools provided across service nodes, such as Data Detective for data collection and Dependency Explorer for dependency queries, mABC efficiently processes alert events and accurately identifies root causes in complex micro-services architectures.

Implementation and Deployment

Experimental results on public benchmark AIOps challenge datasets and a self-created train-ticket dataset showed that mABC outperforms previous strong baselines in accurately identifying root causes and creating effective solutions. The ablation study further emphasizes the importance of each component within mABC, underlining that Agent Workflow, multi-agent collaboration, and blockchain-inspired voting are key to optimal performance.

Summary

mABC is an innovative framework that leverages LLMs and multi-agent cooperation, facilitated by blockchain-inspired decision-making processes, aimed at root cause analysis (RCA) in micro-services architectures within cloud-native technologies.