"Toru Nishimura" <locore64@alkyltechnology.com> writes:

> I feel hesitating being a man of "howling for full moon," though,
> I've been skeptic about FTL/journalled flash filesys
> implementations;

May I join you in your howl?  I too am skeptic about FTL.

But my reason for asking that NetBSD design consider it is that the
commercial file systems that we use are organized around an FTL --
even though they have a file system that is optimized for NAND use.

> Then, what characterises should be pursued to invent the filesys?
> - well-defined ordered meta operation to modify filesys.
> - a distinct block reclaim strategy; blocks which are free'd away can
> be re-written only when all the blocks are exhausted.  "Never re-write
> in the place."
>
> We have good basis for them; the former is nothing other than
> softdep, and the latter is the characteristic (ah, goal) of BSDLFS.
> Bad block avoidance can be done in ease just FAT can do.
>

In a former life I participated in a project to adapt LFS to NAND for
a custom OS, without an FTL. It turns out to be a lot of work. I
recommend that a custom file system tuned to NAND characteristics be
designed and would like to participate in such a design.