Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 2

Fr.

CRCE Operating Systems Lab 2023-2024

Fr. Conceicao Rodrigues College of Engineering


Department of Computer Engineering

Assignment 1

Student’s Roll No 9980 Student Name Aashish Mhaske


Date of 26/03/2024 SE Computer – Div B
Performance

Major Topics: AI-ML in Operating System

Title: A Machine Learning Framework to improve storage


system performance

Abstract: Storage systems and their operating system parts are built
to handle different types of tasks and changing workloads
Machine learning can help by recognizing patterns,
making predictions, and improving storage systems.
KML, is a machine learning framework for storage
systems. It is tested it by optimizing readahead values and
found that it improved performance significantly without
adding much extra work or memory usage. In simple
terms, KML can make storage systems smarter and work
better by learning from data and making clever decisions.
The research showcases the potential of integrating ML
frameworks like KML into operating systems to enhance
storage system performance.

Description: In computer systems, storage performance is crucial for


smooth operation. However, optimizing storage systems
manually can be challenging due to the complexity of
workloads and system configurations. To address this
challenge, we explore the use of Kernel Machine
Learning (KML) as a solution for storage system
optimization, focusing on the specific problem of
readahead optimization.
Readahead is a technique used in storage systems to
prefetch data based on expected access patterns,
improving input/output (I/O) performance. Traditional
storage systems rely on manual configuration and simple
heuristics to optimize readahead values, which prefetch
data to improve I/O performance. However, these
methods often fall short in adapting to dynamic
workloads and may not achieve optimal performance.
KML, an innovative Machine Learning framework
designed specifically for storage systems within the OS.
KML utilizes ML techniques to learn from data patterns
and make intelligent predictions, thereby optimizing
readahead values automatically.
Key Features of KML:
1) ML-based Optimization: KML uses ML algorithms
to analyze data patterns and dynamically adjust
readahead values for optimal performance.
2) Adaptability: Unlike traditional methods, KML can
adapt to changing workloads and optimize
readahead values in real-time.
3) Low Overhead: KML is designed to have minimal
overhead, ensuring efficient operation without
compromising system performance.
4) Flexibility: KML's architecture allows for easy
integration and extensibility, making it suitable for
various storage system configurations.

KML is implemented as a prototype and conducted


experiments to evaluate its performance, results
demonstrate significant improvements in throughput,
with performance enhancements of up to 2.3 times
compared to traditional methods. Moreover, KML's
memory footprint is small, and it introduces negligible
overhead, making it a practical and efficient solution for
storage system optimization.
References: https://www.ijtrd.com/papers/IJTRD22752.pdf
https://macpaw.tech/research/ai-in-os/
https://dl.acm.org/doi/pdf/10.1145/3465332.3470875

Knowledge of Presentation(4) QnA(2) Total (10)


Topic(4)

Signature of Faculty Date of Submission

You might also like