Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reproduction Research of FSA-Benchmark

Joshua Ludolf, Yesmin Reyna-Hernandez, Matthew Trevino

TL;DR

This work benchmarks ML-based fail-slow disk detection using the PERSEUS dataset, evaluating seven models (CSR, Multi-Pred, LSTM, PatchTST, Autoencoder, Isolation Forest, SVM) and a large language model across clusters A and B, with lookback windows and $ thresholds$ to assess precision, recall, and failure rates. The results show a spectrum of trade-offs: low-false-alarm detectors like Autoencoder and Isolation Forest excel in specificity, while high-sensitivity models like LSTM and SVM achieve earlier detection at the cost of higher failure rates. The study provides practical deployment guidance, highlighting environment-specific model selection, the impact of short versus long lookbacks, and the need for hybrid and real-time approaches. It also outlines future work on model integration, adaptive thresholds, and system-wide integration to improve predictive maintenance for large-scale storage systems, ensuring reliability amid growing data volumes.

Abstract

In the current landscape of big data, the reliability and performance of storage systems are essential to the success of various applications and services. as data volumes continue to grow exponentially, the complexity and scale of the storage infrastructures needed to manage this data also increase. a significant challenge faced by data centers and storage systems is the detection and management of fail-slow disks that experience a gradual decline in performance before ultimately failing. Unlike outright disk failures, fail-slow conditions can go undetected for prolonged periods, leading to considerable impacts on system performance and user experience.

Reproduction Research of FSA-Benchmark

TL;DR

This work benchmarks ML-based fail-slow disk detection using the PERSEUS dataset, evaluating seven models (CSR, Multi-Pred, LSTM, PatchTST, Autoencoder, Isolation Forest, SVM) and a large language model across clusters A and B, with lookback windows and to assess precision, recall, and failure rates. The results show a spectrum of trade-offs: low-false-alarm detectors like Autoencoder and Isolation Forest excel in specificity, while high-sensitivity models like LSTM and SVM achieve earlier detection at the cost of higher failure rates. The study provides practical deployment guidance, highlighting environment-specific model selection, the impact of short versus long lookbacks, and the need for hybrid and real-time approaches. It also outlines future work on model integration, adaptive thresholds, and system-wide integration to improve predictive maintenance for large-scale storage systems, ensuring reliability amid growing data volumes.

Abstract

In the current landscape of big data, the reliability and performance of storage systems are essential to the success of various applications and services. as data volumes continue to grow exponentially, the complexity and scale of the storage infrastructures needed to manage this data also increase. a significant challenge faced by data centers and storage systems is the detection and management of fail-slow disks that experience a gradual decline in performance before ultimately failing. Unlike outright disk failures, fail-slow conditions can go undetected for prolonged periods, leading to considerable impacts on system performance and user experience.
Paper Structure (34 sections, 10 figures)

This paper contains 34 sections, 10 figures.

Table of Contents

  1. Introduction
  2. Task description and data construction
  3. Machine Learning Algorithmns
  4. Original Research Machine Learning Models
  5. Cost-Sensitive Ranking Model
  6. Multi-Prediction Models
  7. LSTM Model
  8. PatchTST Model
  9. Large language model
  10. Novelty Machine Learning Models
  11. Autoencoder
  12. Isolation forest
  13. Suport Vector Machine (SVM)
  14. Performance Benchmarking
  15. Results
  16. ...and 19 more sections

Figures (10)

  • Figure 1: CSR Model Performance Heatmaps for Clusters A and B
  • Figure 2: Multi-Prediction Model Performance Heatmaps for Clusters A and B
  • Figure 3: LSTM Model Performance Heatmaps for Clusters A and B
  • Figure 4: PatchTST Model Performance Heatmaps for Clusters A and B
  • Figure 5: Autoencoder Model Performance Heatmaps for Cluster A
  • ...and 5 more figures