Woodhouse on flash storage
wmb at laptop.org
Tue Oct 6 05:27:26 EDT 2009
Martin Langhoff wrote:
> On Tue, Oct 6, 2009 at 12:07 AM, Mitch Bradley <wmb at laptop.org> wrote:
>> time, I expect the situation to get better and better as the firmware
>> that "gets it right" supplants the earlier tries.
> It's reassuring to hear that at least someone with your understanding
> of HW (and the industry around it) thinks that it will get better
> (good enough?) soon.
> David and Valerie (both with lots of FS work behind them) seem to
> think that there are fundamentally hard problems in implementing FTLs.
Indeed, the problems are fundamentally hard. But they are not intractable.
> Valerie's article lays the problems out, and David argues that FOSS
> dynamics and developers have a better chance of solving them.
Part of my lack of faith in the FOSS dynamics stems from frustration
that the FOSS community did not manage to solve the problem in time to
help XO-1, despite that fact that we knew well in advance that JFFS2 had
already exceeded its scaling limit. I came to the conclusion that, for
all of FOSS's structural advantages, it has one deadly disadvantage -
lack of money. To solve a hard problem to the level where you can ship
the result requires money, because that it the most reliable way to
motivate someone to keep working past the "fun" point. Of course there
are exceptions, but the fact that they are exceptions demonstrates the
general validity of the rule.
At this point I am placing my bet that commercial dynamics will win in
this case. The amount of money that is involved in the transition to
FLASH-based storage is absolutely staggering, enough to fund a lot of
commercial developers, some of whom are probably talented enough to
succeed. In my experience, a hundred-to-one spending advantage is not
easy to overcome, even if you have a "better idea". You might "succeed"
in coming up with a "better solution", but it largely won't matter
because the industry will follow the money spender.
The astute reader will no doubt notice a parallel here with Open Firmware...
> Of course, FS developers crave the fun and direct control :-)
> Given the natural tendency of our userbase to find limits (like
> ENOSPC), the issues that Valerie's paper outlines and the flux things
> are in, the work Wad is doing in testing the cr*p out of SD cards is
> the most important thing we can do.
A large part of my belief that the situation will get better is based on
wad's testing. We have at least two devices that have completed quite a
lot of stress testing without falling over. The controllers inside them
have "gotten it right" to a useful degree. There are a modest number of
vendors of such controllers. All it takes to transform the industry is
for the device vendors to start buying the controller/firmware
combinations that work. That could happen almost overnight.
> I wish we had more company in this.
We have a lot of company - the entire market is waking up to the problem
that the previous generation of FLASH-based storage was dodgy. The
controller vendors have already made great strides toward fixing it.
More information about the Devel