Is Ordering of Disk Updates Required to Maintain File-System Crash-Consistency?

On reboot after a crash, the file system should be consistent: e.g., previously correct files should not now contain garbage. In early file systems, getting to a consistent state involved a full scan after reboot. This was very slow, and impractical for large systems. Modern file systems improve upon this by writing updates to disk in a specific order : e.g, metadata before commit blocks. This allows them to get to a consistent state without a scan. However, ordering updates results in certain problems: 1. The file system write order may not be the most efficient order for writing blocks to disk. This reduces performance. 2. The file system has to be very careful about the order; this increases complexity, potentially leading to more bugs and lower reliability. 3. For disks with write caches, commands such as cache flushes are required to ensure correct ordering. If such commands are not properly implemented, consistency is compromised [1]. 4. In virtualized stacks, even if one of the many layers between the file system and the disk does not enforce ordering, consistency is lost. The question then arises: can crash-consistency be maintained without ordering updates? Recent work introduced the No-Order File System (NoFS) [2], the first file system to provide strong consistency despite not ordering updates. NoFS uses a novel technique called Backpointer-Based Consistency (BBC) that establishes consistency via mutual agreement between file-system objects. NoFS performs as well as a comparable journaling file system (ext3) for most workloads, and increases throughput by 20-70% for metadata-intensive workloads. BODYNot only is it possible to maintain file-system crash-consistency without order-ing updates, but doing so may actually increase performance. REFERENCES[1] Rajimwale, A., Chidambaram, V., Ramamurthi, D., Arpaci-Dusseau, A. C., andArpaci-Dusseau, R. H. Coerced cache eviction and discreet mode journaling: Dealing withmisbehaving disks. In Dependable Systems & Networks (DSN), 2011 IEEE/IFIP 41stInternational Conference on (Hong Kong, China, June 2011), IEEE, pp. 518–529.[2] Vijay Chidambaram, Tushar Sharma, Andrea C. Arpaci-Dusseau, Remzi H.Arpaci-Dusseau. Consistency Without Ordering. In Proceedings of the 10th Conference onFile and Storage Technologies (FAST ’12) (San Jose, California, February 2012). Volume 2 of Tiny Transactions on Computer ScienceThis content is released under the Creative Commons Attribution-NonCommercial ShareAlike License. Permission tomake digital or hard copies of all or part of this work is granted without fee provided that copies are not made ordistributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page.CC BY-NC-SA 3.0: http://creativecommons.org/licenses/by-nc-sa/3.0/.