Back to Search Start Over

CoDiscard: A revenue model based cross-layer cooperative discarding mechanism for flash memory devices.

Authors :
Feng, Xiaoliu
Chen, Xianzhang
Li, Ruolan
Li, Jiali
Song, Chunlin
Liu, Duo
Tan, Yujuan
Qiao, Lei
Source :
Journal of Systems Architecture. Jul2022, Vol. 128, pN.PAG-N.PAG. 1p.
Publication Year :
2022

Abstract

Write amplification inside the flash memory device is a severe problem, which shortens the flash's lifespan and degrades its performance. The TRIM command is proposed to reduce write amplification by avoiding unnecessary relocation in flash memory devices. However, the significant overhead causes the TRIM command to be completely prohibited by common file systems when I/O is busy. This completely prohibited TRIM strategy not only affects the lifespan of the flash memory device but also degrades the performance due to the huge write amplification. In this paper, we propose CoDiscard, a revenue model based cross-layer cooperative discarding mechanism, to optimize the I/O performance and reduce write amplification under intensive workloads. The core idea of CoDiscard is to sieve the high performance–price-ratio TRIM commands to the flash memory device because they can reduce write amplification significantly, thus improving performance. To identify such TRIM commands, we propose a revenue model to predict the write amplification reduction and the corresponding overhead. Based on the revenue model, CoDiscard employs a cross-layer cooperative scheme to schedule the TRIM command. The CoDiscard scheduler in the file system is responsible for making decisions. The CoDiscard monitor in the flash memory device is used to monitor the changing of revenue model related information. We conduct the experiment under real workloads. The results show that the performance improves for 113% and 118.4%, and the write amplification reduces up to 12.12 × and 8.18 ×. [ABSTRACT FROM AUTHOR]

Details

Language :
English
ISSN :
13837621
Volume :
128
Database :
Academic Search Index
Journal :
Journal of Systems Architecture
Publication Type :
Academic Journal
Accession number :
157253194
Full Text :
https://doi.org/10.1016/j.sysarc.2022.102564