Re: Found on Slashdot: Wear Leveling, RAID Can Wipe Out SSD Advantage

On Mon, Mar 8, 2010 at 22:38, DrFranken <midrange@xxxxxxxxxxxx> wrote:

IBM's SSDs are the first with SAS connectivity and as Jim said are built
with nearly 2x overcapacity and built in wear leveling at the device. I
find it difficult to believe that would eat up the device's performance
as the only time it would come into play is when the device is written
to. Since you have to write somewhere anyway how is that going to slow
it down? Now if the RAID card or the O/S did the wear leveling that's
another problem. We don't do that.

This mostly stems from two things:

a) The need for TRIM on consumer SSDs with no or minimal spare
capacity. Currently, no RAID controllers support TRIM. Without TRIM,
the SSD can't pre-erase blocks without garbage collection, which makes
write performance slower.

b) SSD controller performance used by garbage collection in non-TRIM
enabled devices. Garbage collection will make the drive slower while
it is running.

Note that both of these issues don't exist if a drive has 100% spare
capacity and enough controller channels. Unfortunately, i have no real
idea on how IBM's enterprise SSDs really work, because i haven't
touched any of them and most of the hw tech sites don't get to touch
this stuff, so i can't say for sure that that's the case.