WOJ: Enabling Write-Once Full-data Journaling in SSDs by using weak-hashing-based deduplication

摘要

Journaling is a commonly used technique in file systems to provide data reliability for applications. Full-data journaling, which stores all file system (data and metadata) updates in a journal before they are applied to their home locations, provides the strongest data reliability, reduces application developers' efforts on application-level crash consistency, and helps to remove most crash-consistency vulnerabilities. However, file system users usually hesitate to use it as it doubles the write volume to the disk, leading to compromised performance. While fast SSDs have the potential to make full-data journaling affordable, its doubled writes threaten the devices' durability, which is their Achilles heel.While data deduplication technique can be used to remove the second writes to the home locations, it can be too expensive to be a practical solution due to its high computation and space overheads as it has to compute and cache collision-resistant hashing values (fingerprints). The issue is especially serious for SSDs, which are becoming increasingly large and fast, but less tolerant of additional overhead in the I/O stack. Leveraging the fact that with data journal mode all writes to the home locations in a file system are preceded by corresponding writes to the journal, we propose Write-Once data Journaling (WOJ), which uses a weak-hashing-based deduplication dedicated for removing the second writes in data journaling. WOJ can reduce regular deduplication's computation and space overheads significantly without compromising the correctness. To further reduce metadata persistency cost, WOJ is integrated with SSD's FTL within the device. Experiment results show that the ext3 file system with data journaling on WOJ-enabled SSDs can deliver up to 2.7X higher throughput than that with regular deduplication, while both remove about half of the writes to the SSD. (C) 2018 Elsevier B.V. All rights reserved.