An efficient log-structured scheme for disk arrays

Citations

WEB OF SCIENCE

0
Citations

SCOPUS

0

초록

Disk contention is a significant performance issue in cloud storage where multiple applications share disk resources and perform I/O requests with different access patterns. Log-structured disk array designs can alleviate this I/O contention between applications by sequentializing the I/O requests. However, garbage collection (GC) can still affect the performance of disk arrays. In this paper, we propose a log-structured disk array management with an efficient disk layout and parallel GC schemes to alleviate the performance interference caused by GC. In our scheme, we first split a large space of disk array into multiple logical units (e.g., zones and segments) to reduce the I/O contention and enable efficient GC operations. Second, we provide an efficient parallel GC scheme that performs the data migration process using multiple logical units and disks in a parallel manner. We implement our scheme at the block I/O layer in Linux kernel 5.4.0. The experimental results show that our proposed scheme can outperform RAID-0 scheme and a log-structured disk array scheme by up to 43.9x and 5.86x, respectively. © 2022 ACM.

제목
An efficient log-structured scheme for disk arrays
저자
Zhu, Guangyu; Lee, Sang Jin; Son, Yongseok
DOI
10.1145/3477314.3507148
발행일
2022-04
유형
Proceedings Paper
저널명
Proceedings of the ACM Symposium on Applied Computing
권
37
페이지
1197 ~ 1204