On Thursday 20 April 2006 00:43, Ian Oliver wrote:
> In article <200604180905.16410.manningc2@actrix.gen.nz>, Charles Manning
>
> wrote:
> > From a quick read, this software would appear toprovide a block driver
> > that then needs a further file system to provide files etc.
>
> Out of interest, can we use the lower layers of YAFFS but put our own
> filing system (probably an implementation of fat32) on top of it?
>
> I've seen this diagram - http://www.aleph1.co.uk/node/128
> but it's not clear if the layering of yaffs is such that there is a
> traditional block device at some point when using YAFFS direct.

YAFFS does not use any internal block device. The whole idea behind a flash 
file system is that it works directly with flash and does not implement a 
block device. 

Block device + regular fs are more fragile and slower than an FFS. In a 
non-scientific study, a device was run with YAFFS and with a block driver + 
FAT using the same low level NAND access functions + hw. YAFFS performed the 
canned tests approx 50% faster than the FAT implementation. Since YAFFS and 
FAT will have similar read speeds, and reading took up a lot of the test 
time, that would mean YAFFS was about 2x the speed of FAT in this case.

What yaffs does provide internally is block and page ("chunk") management. It 
would be possible to fit a block device on top of that.

-- Charles