Autor*innen mit den meisten Dokumenten
Auflistung nach:
Neueste Veröffentlichungen
- TextdokumentFull-Scale File System Acceleration on GPU(Tagungsband des FG-BS Frühjahrstreffens 2024, 2024) Maucher, Peter; Kittner, Lennard; Rath,Nico; Lucka,Gregor; Werling,Lukas; Khalil,Yussuf; Gröninger,Thorsten; Bellosa,FrankModern 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 pre- liminary 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.
- TextdokumentImprovements in Crash Consistency Testing for Persistent Memory File Systems(Tagungsband des FG-BS Frühjahrstreffens 2024, 2024) Werling, Lukas; Oder, Thomas-Christian; Wäldele, Lucas; Ritz, Daniel; Bellosa, FrankAchieving crash consistency in persistent memory (PM) applications is difficult due to the need for fine-granular cache flushes and memory fences. File systems are no exception. Previous works have found numerous bugs in PM file systems, but crash consistency testing as part of file system development is still rare. With this paper, we make crash consistency testing easier by improving Vinter, a crash consistency testing tool for file systems based on virtual machines. We introduce support for cross-media file systems with NVMe and PM that are not covered by other testing tools. To speed up testing, we add an alternative algorithm for simulating crashes that focuses on logic bugs rather than misuse of PM primitives.
- TextdokumentManagarm: A Fully Asynchronous Operating System(Tagungsband des FG-BS Frühjahrstreffens 2024, 2024) van der Grinten, Alexander; Słomiński, Kacper; Custers, GeertIn this paper, we give an overview of the system architecture of Managarm, a free and open source operating system that is based on a microkernel. The goal of Managarm is to build a general-purpose OS on top of unprivileged drivers and servers that run in user space, while still providing extensive source-level compatibility with existing POSIX and Linux applications. To minimize context switches even when driving modern hardware that supports high degrees of concurrency, Managarm exclusively relies an asynchronous IPC mechanism that enables the submission of an arbitrary number of independent IPC requests before performing a context switch. Additionally, since existing POSIX and Linux applications are not always designed around asynchronicity, we provide a user level emulation of POSIX and Linux APIs. Our emulation is sufficient to run various Linux application, including modern desktop environments.