[linux-elitists] A modest proposal: SSD disk acceleration in kernel.

Bill Bogstad bogstad@pobox.com
Fri Oct 24 10:29:51 PDT 2008

On Fri, Oct 24, 2008 at 11:26 AM, Karsten M. Self
<karsten@linuxmafia.com> wrote:
> on Fri, Oct 24, 2008 at 10:58:40AM -0400, Ruben Safir (ruben@mrbrklyn.com) wrote:

>> BBad idea with current SSD's bcause of limited life.
> See Linus's blog entry (referenced in OP) for specifics on SSD lifetimes
> and MTBF.  The evidence there is that it's at least comperable to
> rotational media.  I'm guessing with sufficient error detection and
> correction and allowing for degraded volume, probably better.

I'm wondering just how different the failure modes will be as well.
Certainly if we are talking about 'wear' effects it will.  With
spinning platters, it is common wisdom that if you start getting new
bad blocks you should plan on replacement or catastrophic failure
soon.  Flash memory blocks which have reached their max. erase cycle
limit wouldn't seem to imply the same thing.  OTOH, this would also
seem to imply that SSD interfaces that hide these kinds of issues
behind disk-like interfaces aren't good for reliable systems.  Does
the ATA/SATA command set have a way to inform the host that it only
has N writes left on the 'disk'?

OTOH,  it's my understanding that flash is laid out in a row/column
format at lot like RAM.  I suppose if the access electronics go bad,
you could lose the
equivalent of a 'track' of data all at once....

Bill Bogstad

More information about the linux-elitists mailing list