Using Machine Learning for Accelerating the Detection Time of Unreliable Failure Detectors

Aug 3, 2026

Upload by Skyler

This study evaluates machine learning-based failure detectors for identifying faulty nodes in distributed systems, using metrics such as detection time and average mistake rate. Random Forest and Support Vector Machine failure detectors show the strongest overall performance. Despite this, machine learning approaches generate higher error rates, especially when detection time increases. To improve the trade-off, three ensemble methods are proposed, including one combining top machine learning and analytical failure detectors. A precision distribution module enhances accuracy but reduces completeness, addressed via an analytical fallback whose effectiveness depends on threshold relationships.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages70

Reading Time2 h 56 min

AccessFree to download

Introduction

Background

Problem Statement

Motivation and Related Work

Contribution of This Work

Disposition

Methods

Crash Failures

Failure Detectors

Unreliable Failure Detectors

Eventually Perfect Failure Detectors

Heartbeat Failure Detector

Analytical Failure Detectors

Machine Learning

Binary Classifiers

Ensemble Methods

Data Points Generation

Parameter Module

Timed and Time-free Windows

Generating Data Points

Precision-Distribution Module

Data Leakage

Using Machine Learning for Accelerating the Detection Time of Unreliable Failure Detectors Master’s thesis in Computer science and engineering CHENGBO REN Department of Computer Science and Engineering CHALMERS UNIVERSITY OF TECHNOLOGY UNIVERSITY OF GOTHENBURG Gothenburg, Sweden 2024 Master’s thesis 2024 Using Machine Learning for Accelerating the Detection Time of Unreliable Failure Detectors CHENGBO REN Department of Computer Science and Engineering Chalmers University of Technology University of Gothenburg Gothenburg, Sweden 2024 Using Machine Learning for Accelerating the Detection Time of Unreliable Failure Detectors CHENGBO REN © CHENGBO REN, 2024 . Supervisor: Elad Michael Schiller, Department of Computer Science and Engineering Advisor: Hasan Mahmood, Nexer Group Examiner: Romaric Duvignau, Department of Computer Science and Engineering Master’s Thesis 2024 Department of Computer Science and Engineering Chalmers University of Technology and University of Gothenburg SE-412 96 Gothenburg Telephone +46 31 772 1000 Typeset in LATEX Gothenburg, Sweden 2024 Using Machine Learning for Accelerating the Detection Time of Unreliable Failure Detectors CHENGBO REN Department of Computer Science and Engineering Chalmers University of Technology and University of Gothenburg Abstract This study evaluates Machine Learning-based Failure Detectors for detecting faulty nodes in distributed systems, focusing on metrics like Detection Time and Average Mistake Rate. Random Forest and Support Vector Machine Failure Detectors emerge as top performers. However, Machine Learning-based Failure Detectors exhibit higher error rates, especially with longer detection times. To address this, three ensemble methods are proposed, with one integrating top Machine Learningbased Failure Detectors and analytical Failure Detectors for better balance. The Precision Distribution module improves accuracy but struggles with completeness. One analytical Failure Detector is used as a fallback to address completeness concerns, although its effectiveness depends on the relationship between the analytical Failure Detector and Precision Distribution module thresholds. Additionally, the Li-Marin Long Short-Term Memory Failure Detector shows reduced detection time but with increased computational overhead, posing challenges in fine-tuning overestimation levels. Keywords: Failure Detector, Machine Learning, Distributed Systems, Ensemble Methods. Acknowledgements On finishing the last word of my thesis report, I see a heart filled with gratitude beating in my chest. I first need to thank my supervisor, Elad Michael Schiller, who guided me through the whole project with great patience and wisdom on not only academic research but also how to be a strong and honest man. His passion and devotion to his work have set a standard for me to reach regardless of my career choice in the future. Iam so grateful to have him as my supervisor. I extend my sincere appreciation to my company advisor, Hasan Mahmood. Hasan’s practical insights, expert advice, and willingness to share his knowledge have played a pivotal role in enriching the relevance and applicability of my research within the corporate context. His support and encouragement have provided me with invaluable perspectives and guidance, enabling me to navigate the complexities of integrating academic theory with real-world practices effectively. To my beloved wife and daughters, I owe an immeasurable debt of gratitude for their unwavering love, encouragement, and support throughout my thesis journey. Their constant belief in my abilities, unwavering patience during challenging times, and endless encouragement have been the cornerstone of my academic achievements. Whether it was offering words of wisdom, lending a listening ear, or providing practical assistance, my family’s steadfast support has been the driving force behind my perseverance and success. I am prof

FAQs

Which failure detector models perform best in detection accuracy and detection time?

Random Forest and Support Vector Machine failure detectors emerge as the top performers based on the study’s metrics, especially for detection time and average mistake rate.

What main problem remains with machine learning-based failure detectors?

Machine learning-based detectors show higher error rates when detection times become longer, creating an accuracy–latency imbalance.

How does the proposed ensemble approach address the accuracy–completeness trade-off?

Three ensemble methods are proposed, including an integration of top machine learning and analytical failure detectors. A precision distribution module improves accuracy, while an analytical fallback is used to mitigate completeness issues, depending on threshold relationships.

Aug 3, 2026

Upload by Skyler