NAND Write operations

Rob Landley rob at landley.net
Sat Apr 17 06:29:54 UTC 2010


On Monday 12 April 2010 17:06:19 Charles Manning wrote:
> You'd probably do far better to ask questions like this on the linux-mtd
> list or similar.
>
> I'm the author of the yaffs file system, so I can answer some of your
> questions though.
>
> Pretty much all flash file systems or flash management layers such as UBI
> perform at least some degree of wear levelling. This means that there is
> some sort of logical to physical, or other, policy so that the writes don't
> happen in the same place. That means those writes to a certain file will
> end up being spread across many different blocks, thus meaning that writes
> to one particular file won't wear out one part of flash.
>
> And yes, the endurance depends on the type of flash. Some are of the order
> of 10^6, some are only 10^3.
>
> Worst I've seen is 10^2, but that was for a "bootloader" section and not
> designated for file usage.

Apparently it's not just the chip, but also depends on how the flash is wired 
up.  (At a co-worker's previous company they had horrible flash reliability 
problems until an engineer went in and redid the wiring on the board.  
Grounded it properly or something, I dunno.  Got it back up to 10^6 without 
changing what chip the design was using.)

My takeaway from it was "flash is hard ot get right electrically, and you may 
not notice until later".

Rob
-- 
Latency is more important than throughput. It's that simple. - Linus Torvalds


More information about the busybox mailing list