File Deletion and File Movement for GPU4Fs
- Typ:Bachelorarbeit
- Datum:24.10.2024
- Betreuung:
Prof. Dr. Frank Bellosa
Peter Maucher - Bearbeitung:Leoluca Tinzmann
- Links:PDF
-
Abstract
Persistent memory, like Intel Optane, is a step to narrow the performance gap between non-volatile Solid State Drives (SSDs) and volatile Dynamic Random Access Memory (DRAM). However, file systems backed by Intel Optane cause disproportionate CPU usage when compared to file systems backed by SSDs. GPU4FS intends to solve that problem by offloading all file system operations to the GPU.We extend the limited feature set of GPU4FS to include file deletions and moves. In our evaluation, we compare our implementation to the CPU file system ext4. We conclude that GPU file deletions can at least be four times faster than its CPU counterpart, depending on the number of files and structure of directories that contain the files. However, GPU file moves are slower than CPU file moves due to the transfer time of file paths from DRAM to VRAM.BibTex:
@bachethesis{File Deletion and File Movement for
GPU4Fs,
author = {Leoluca Tinzmann},
title = {File Deletion and File Movement for
GPU4Fs},
type = {Bachelor Thesis},
year = 2024,
month = oct# "24",
school = {Operating Systems Group, Karlsruhe Institute of Technology (KIT), Germany}
}