[Ksummit-2008-discuss] Suggested topic: possible

[Matthew Wilcox]

On Wed, Aug 06, 2008 at 08:22:31PM +0300, Eyal Shani wrote:
> I did not, by any means, try to disrespect btrfs, or any other FS development efforts done in the community.
> I am sorry if that was the message I conveyed.

The message you conveyed was actually worse.  It conjured up images of
smoke-filled backrooms where you'd been doing deals out of sight.
Remember you're talking to engineers here, not people who're impressed
by market share and secret sauce.

> I probably should not claim to know what is the right FS to use - that should be left to you Linux experts.
> My main message was that SSDs should NOT be managed as MTDs or FTLs devices in SW - thus claiming to 'know' what is being done to manage flash internally.
> 
> This is the mere beginning of adoption for SSDs as storage devices (as a consumer PC vast market), which is a great technology change.
> Today it is MLC or SLC. Tomorrow you will witness SSD that have SLC/MLC/D3 all together, in one device. Sectors will end up stored according to their context, and expected life cycle.
> Techniques like over-provisioning, that publish capacity X, while actually managing a larger capacity behind the screen is already used today.
> 
> The innovation curve in SSDs will soar, I hope, and taking design assumptions that take into account internal architecture of these devices will, in my opinion, prove very limited in their overtime relevance.

I think that with flash, as with many other technologies, the market
doesn't move so much in a straight line as swing back and forth like a
pendulum.  In graphics, for example, the tradeoffs have changed over the
years.  Sometimes more should be done on the CPU, sometimes more on the
GPU.  Sometimes you want video ram on the card, sometimes it's faster
for the card to access main memory.

I think there's a place for both SSDs that pretend to be a linear array
of sectors as well as ones which expose the characteristics of the raw
flash.  Maybe there's even scope for devices which you can send a magic
sequence to and they say "Oh, right, yeah, I'm not an array of sectors
any more, I'm MLC, deal with me like that".

> Martin, I am not sure what happened in the past. I prefer to ask what we can do for the future, and hopefully the NEAR future.
> I am merely stepping into the Linux arena, and am trying to do the right moves, to enable SSDs. (while irritating you guys already, without trying to...:))

What I think would be hugely useful would be to have a conformance
test-suite for devices.  USB storage has been absolutely terrible for
devices hanging at random, setting strange bits, claiming conformance to
standards the the manufacturers have clearly never even heard of ... I'd
really like to see the industry require a certification test before you
can call a device 'USB'.  Or 'SCSI' for that matter.

-- 
Intel are signing my paycheques ... these opinions are still mine
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."