Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Performance Modeling of Data Storage Systems using Generative Models

Abdalaziz Rashid Al-Maeeni, Aziz Temirkhanov, Artem Ryzhikov, Mikhail Hushchyn

TL;DR

This study developed several models of a storage system using machine learning-based generative models to predict performance metrics such as IOPS and latency, offering a vendor-agnostic approach for simulating data storage system behavior.

Abstract

High-precision modeling of systems is one of the main areas of industrial data analysis. Models of systems, their digital twins, are used to predict their behavior under various conditions. We have developed several models of a storage system using machine learning-based generative models. The system consists of several components: hard disk drive (HDD) and solid-state drive (SSD) storage pools with different RAID schemes and cache. Each storage component is represented by a probabilistic model that describes the probability distribution of the component performance in terms of IOPS and latency, depending on their configuration and external data load parameters. The results of the experiments demonstrate the errors of 4-10 % for IOPS and 3-16 % for latency predictions depending on the components and models of the system. The predictions show up to 0.99 Pearson correlation with Little's law, which can be used for unsupervised reliability checks of the models. In addition, we present novel data sets that can be used for benchmarking regression algorithms, conditional generative models, and uncertainty estimation methods in machine learning.

Performance Modeling of Data Storage Systems using Generative Models

TL;DR

This study developed several models of a storage system using machine learning-based generative models to predict performance metrics such as IOPS and latency, offering a vendor-agnostic approach for simulating data storage system behavior.

Abstract

High-precision modeling of systems is one of the main areas of industrial data analysis. Models of systems, their digital twins, are used to predict their behavior under various conditions. We have developed several models of a storage system using machine learning-based generative models. The system consists of several components: hard disk drive (HDD) and solid-state drive (SSD) storage pools with different RAID schemes and cache. Each storage component is represented by a probabilistic model that describes the probability distribution of the component performance in terms of IOPS and latency, depending on their configuration and external data load parameters. The results of the experiments demonstrate the errors of 4-10 % for IOPS and 3-16 % for latency predictions depending on the components and models of the system. The predictions show up to 0.99 Pearson correlation with Little's law, which can be used for unsupervised reliability checks of the models. In addition, we present novel data sets that can be used for benchmarking regression algorithms, conditional generative models, and uncertainty estimation methods in machine learning.
Paper Structure (12 sections, 18 equations, 10 figures, 10 tables)

This paper contains 12 sections, 18 equations, 10 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Problem statement
  3. Data collection
  4. Models
  5. CatBoost model
  6. Normalizing flow
  7. kNN model
  8. Model selection
  9. Quality metrics
  10. Results
  11. Discussion
  12. Conclusion

Figures (10)

  • Figure 1: The CatBoost-based model for performance predictions of the storage pools and cache for the given values of data load and configuration (for the pools only) parameters.
  • Figure 2: The normalizing flow model for performance predictions of the storage pools and cache for the given values of data load and configuration (for the pools only) parameters
  • Figure 3: Means ($\mu$) and standard deviations ($\sigma$) of the measured IOPS and latencies for cache, SSD, and HDD pools under random and sequential data loads; each point corresponds to one data load and one component configuration
  • Figure 4: Example of real observations and predictions for read data loads on the cache; Each cloud corresponds to one data load.
  • Figure 5: Example of real observations and predictions for read data loads on the HDD pool; Each cloud corresponds to one data load.
  • ...and 5 more figures