NAND Framework in HEAD.

[Don Lewis]

On 17 May, Grzegorz Bernacki wrote:

> NAND FS adopts log-structured approach and some parts of its internal 
> design are derived from the new implementation of the log-structured 
> file system (NILFS), with some concepts rooting in the original (now 
> legacy) BSD log-structured file system (LFS).
> 
> The NAND FS has the following major features:
>    - Hard links
>    - Symbolic links
>    - Case-sensitive, case-preserving
>    - Snapshots
>      ? No limit on the number of snapshots (only volume-limited)
>      ? Mountable as read-only file systems
>      ? Simultaneously mountable (there can be a writable mount concurrently
>        mixed with a number of read-only snapshots)
>    - Redundant super block
>    - Metadata
>      ? POSIX file permissions
>      ? Creation timestamps
>      ? Last content modification timestamps
>      ? Last metadata change timestamps
>      ? Checksum / ECC

Any thoughts on how well NAND FS might work on SSDs as compared to
something like UFS, which isn't aware of the properties of the
underlying storage?  I would think that avoiding random block overwrites
would help performance and device lifetime.

[Cc: trimmed]