Woodhouse on flash storage

[John Watlington]

On Oct 5, 2009, at 11:56 AM, Peter Robinson wrote:
> On Mon, Oct 5, 2009 at 4:44 PM, Martin Langhoff
> <martin.langhoff at gmail.com> wrote:
>> On Mon, Oct 5, 2009 at 5:27 PM, Chris Ball <cjb at laptop.org> wrote:
>>> I don't think there's much we can use Dave's advice for given the
>>> choice of microSD.

The fact that Woodhouse is right doesn't change the
situation from a system engineering point of view.
The interfaces to raw NAND flash required to run JFFS2
or UBIFS are hard to find (and expensive) these days.

>>> If a microSD card that had a passthrough mode
>>> to access the raw flash existed, his argument would be that we
>>> should consider using that mode.
>>
>> Exactly. To expand: we are in the same pickle as the rest of the
>> industry. This article gives some good background reading to David's
>> words http://lwn.net/Articles/353411/ -- and if you've been following
>> her writing, Valerie knows a ton about FSs, including work on ZFS.

The entire industry is moving in the direction of eliminating interfaces
that talk directly to NAND Flash chips.   That is what we need to  
change.
The few vendors I've talked to about this, however, are reticient.
They've been burned with attempts to do this in the recent past, and
would rather use an external ($0.50 - $1.50) controller.
eMMC/eSD parts integrate the FTL controller onto the NAND die, providing
a more consistent interface and leveraging the fact that over 90% of the
NAND Flash market goes into SD cards.

> Ted Ts'o, who's one of the lead dev's on ext4, has had some very
> interesting articles about the various issues with SSDs, erase blocks
> and all of the previously mentioned issues. I believe also SD cards
> handle things slightly differently than sata ssds too.
>
> His main blog is here http://thunk.org/tytso/blog/
> But some of the interesting ones are:
> http://thunk.org/tytso/blog/2009/02/20/aligning-filesystems-to-an- 
> ssds-erase-block-size/
> http://thunk.org/tytso/blog/2009/02/22/should-filesystems-be- 
> optimized-for-ssds/
> http://thunk.org/tytso/blog/2009/03/01/ssds-journaling-and- 
> noatimerelatime/

> From memory the plan is to use ext4 so he might well have some of the
> best options for optimal options for SD. I know from my own experience
> the cheeper SD cards tend to have a lot more issues.

I haven't seen much correlation with price.   At this point, vendors are
mixing and matching controllers and NAND chips frequently.   The 2G
and 4G microSDs from a given manufacturer may well use different
controllers (and FTL firmware).   Some manufacturers even use different
controllers in the same size cards (retailed as identical).

Test and verify seems to be the only way to move forward.