Fast and Synchronous Crash Consistency with Metadata Write-Once File System

Learn about a novel file system architecture designed to achieve fast and synchronous crash consistency in this 15-minute conference presentation from OSDI '25. Discover how the metadata write-once file system (WOFS) model addresses the limitations of existing crash consistency approaches like journaling and log-structured file systems, which generate numerous small, random, and ordered metadata I/Os that fail to exploit persistent memory's I/O potential. Explore the core innovation of generating specific metadata for each file operation as a checksum-protected package that is written once with a single ordering point, then managed through a package translation layer to provide file abstractions without additional writes. Examine the array of techniques used to generate, organize, and recover from packages that enable WOFS to deliver practical, efficient, and reliable file system services. Review the implementation of WOLVES, a WOFS prototype developed in the Linux kernel, and analyze experimental results using benchmarks and applications that demonstrate the system's ability to recover from crashes, improve operation throughput, and potentially reach persistent memory I/O bandwidth limits.