Add checkpointing or logging to UBIFS
- Add checkpointing or logging to UBIFS
- Tim Bird
Currently, the UBI file system takes a long time to be ready for use, due to a badblock scan that takes place during initialization of UBI (the UBI attach operation). The scan takes a time that is linearly proportional to the size of flash. As flash sizes grow, the delay caused by this scan becomes a problem.
It would be beneficial to eliminate the scan, or make it more scalable, using some kind of checkpointing or logging.
- Mount times
- This proposal was suggested last year, but the project was not done
UBI logging feature submitted by Samsung
A developer with Samsung (in India) proposed UBIL - UBI logging - to solve this problem, in April 2010.
- Initial announcement: http://lists.infradead.org/pipermail/linux-mtd/2010-February/029256.html
- performance improvement: http://git.infradead.org/users/brijesh/ubil_results/blob/da4f975b5a2fc5cb2f9398bcb6b4329ba7e8c770:/nand_mount_time.pdf
- patches in April, 2010: http://lists.infradead.org/pipermail/linux-mtd/2010-April/029573.html
I seem to recall someone at the kernel summit (maybe Thomas) telling me that these patches improved the speed, but still had scalability proportional to the flash size.
A rough estimate would be 3 months of development and test work, followed by some work to mainline the changes upstream.
- Thomas Gleixner
- Artem Bityutskiy
- Free Electrons