Peter Maucher, Lennard Kittner, Nico Rath, Gregor Lucka, Lukas Werling, Yussuf Khalil, Thorsten Gröninger, Frank Bellosa

Tagungsband des FG-BS Frühjahrstreffens 2024, Bochum, Germany, March 14 - 15, 2024

Abstract
Modern HPC and AI Computing solutions regularly use GPUs as their main source of computational power. This creates a significant imbalance for storage operations for GPU applications, as every such storage operation has to be signalled to and handled by the CPU. In GPU4FS, we propose a radical solution to this imbalance: Move the file system implementation to the application, and run the complete file system on the GPU. This requires multiple changes to the complete file system stack, from the actual storage layout up to the file system interface. Additionally, this approach frees the CPU from file system management tasks, which allows for more meaningful usage of the CPU. In our preliminary implementation, we show that a fully-featured file system running on GPU with minimal CPU interaction is possible, and even bandwidth-competitive depending on the underlying storage medium.

@misc{misc,
author = "Maucher, Peter and Kittner, Lennard and Rath,Nico and Lucka,Gregor and Werling,Lukas and Khalil,Yussuf and Gröninger,Thorsten and Bellosa,Frank",
title = "Full-Scale File System Acceleration on GPU",
year = 2024,
doi = "10.18420/fgbs2024f-03",
howpublished = "Tagungsband des FG-BS Frühjahrstreffens 2024"
publisher = "Gesellschaft für Informatik e.V.",
}