Dr. An-I Andy Wang has a paper accepted by the 23rd USENIX Conference on File and Storage Technologies (USENIX FAST), the flagship conference on file systems and storage. The paper is titled “Silhouette: Leveraging Consistency Mechanisms to Detect Bugs in Persistent Memory-Based File Systems” and was coauthored by Dr. Ashvin Goel at the University of Toronto. Bing Jiao, a doctoral student in the Computer Science Department at FSU is the primary author of this paper.
Writing file system code for persistent memory is challenging because writes may not be propagated to persistent memory in the correct order, resulting in inconsistency after crashes. Theoretically, we want to examine crashed system images resulting from all possible persistence orderings of pending writes, but this search space grows exponentially as the number of pending writes increases. Existing systems use sampling to trim this search space by skipping analyses for classes of writes, resulting in missing opportunities to detect bugs.
We introduced the Silhouette framework, which leverages the observation that file systems tend to use a few common consistency mechanisms to persist data. Therefore, instead of exploring the persistence ordering constraints of individual writes, we explored the ordering constraints at a higher level of persistence phases in common consistency mechanisms. Once we have verified that these mechanisms are correct, we only need to examine the remaining writes that are not protected by these mechanisms.
Compared to the state of the art, our solution reduced the sampled search space by 100x, accelerated the bug-finding process by 10x, and found all bugs discovered by existing systems and 15 new bugs not previously reported.
The paper will be presented at the USENIX FAST Conference in Santa Clara in February 2025.