[sf-lug] About USB partitioning and formatting

[Sameer Verma]

On Thu, Sep 17, 2009 at 5:35 PM, Rick Moen <rick at linuxmafia.com> wrote:
> Quoting jim (jim at well.com):
>
>>    i am, of course, interested in how my surmises
>> are wrong-headed and what might be misinformation
>> or omissions from the linked page.
>
> The basic advice is good (in that they qualify why they're saying what
> they say, and tell you why most repartitioning of NAND flash devices is
> likely to be actively harmful.
>
> I covered the same topic back in 2003 in an article I wrote for _Linux
> Journal_, back when these drives were a relatively new problem (and some
> of us ran still systems with 2.4 kernels):
> http://www.linuxjournal.com/article/6867
>
> You'll note that I said I was avoiding re-doing the drive formatting
> to cut device fatigue, and because the provided FAT filesystem has
> the advantage of ubiquitous support, even though I referred to FAT as
> an "antique disk format".  Detailing why would take a while, but
> basically it depends entirely too much on repeated reads/writes to a
> few fixed data structures.  When the 65-byte (or whatever applies) blocks
> those key structures live on wear out, the entire NAND flash drive
> becomes landfill.
>
> One could argue that you'd be better off doing a one-time conversion to,
> say, JFFS2, which deliberately "levels" the NAND wear over the entire
> device.  As the wiki page implies, you'd want to plan your mkfs.jffs2
> approach with some attention to the device's physical characteristics.
>
> I haven't personally played with JFFS or other NAND-appropriate
> filesystems, because flash drives are cheap, and the lingua-franca
> aspects of FAT are too useful.
>
>
> _______________________________________________
> sf-lug mailing list
> sf-lug at linuxmafia.com
> http://linuxmafia.com/mailman/listinfo/sf-lug
>

Here's something interesting about flash "disks" and filesystems that
work efficiently with them. This is a post by David Woodhouse, who
wrote JFFS. http://www.advogato.org/person/dwmw2/diary/211.html

Evidently, bulk of the problem stems from how vendors make SSDs appear
to be just like spinning disks. So, in effect, if you formatted your
filesystem to use ext3, it would be ext3 -> vendor's translation layer
-> SSD, which then implies that the vendor's translation layer (which
is proprietary) is to be relied on and if something goes wrong, we
don't have any open way to fix it.

cheers,
Sameer
-- 
Dr. Sameer Verma, Ph.D.
Associate Professor, Information Systems
Director, Center for Business Solutions
San Francisco State University
http://verma.sfsu.edu/
http://cbs.sfsu.edu/
http://is.sfsu.edu/