Abstract

The file accesses of existing in-memory file systems have additional costs for traversing the software stacks of the kernel, such as VFS. To avoid such costs, the existing file systems generally enable user-space file accesses using the memorymapped file (mmap) techniques. The mmap approaches, however, are add-ons of the file systems in the kernel level that have large overhead for mapping the files into the user space. In this paper, we propose the design of a genuine user-space in-memory file system. A user-space in-memory file system is designed and implemented. The software routines of the file operations in the proposed file system are re-designed to enable user-space file accesses and be compatible with the POSIX interfaces. A file can be mapped into the user space in constant time regardless of the file size. The file data of the implemented file system can be accessed in the user space with consistency. Our design achieves high performance utilizing the contiguous virtual address space of the user process and the address translation hardware. Extensive experiments are conducted. The experimental results show that the proposed file system surpasses any of the existing file systems. To the authors' knowledge, our design is the first genuine userspace file system in the literature.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call