Silhouette: Leveraging Consistency Mechanisms to Detect Bugs in Persistent Memory-Based File Systems

This conference talk from USENIX FAST '25 explores "Silhouette," a bug-detection framework specifically designed for persistent memory (PM)-based file systems. Discover how researchers from Florida State University and University of Toronto address the challenges of crash-consistency bugs in PM software, which requires careful flushing and ordering of stores from CPU caches. Learn about Silhouette's innovative approach that combines static instrumentation with data-type-based dynamic analysis to verify correct implementation of standard crash-consistency mechanisms like journaling and replication. The presentation demonstrates how Silhouette significantly reduces the exploration space for bug detection, finds existing bugs 10 times faster than current tools, and identifies several previously unknown bugs in various PM file systems.