Ricardo B (ricardo.b@xxxxx.xx) on 1/20/11 wrote:
>
>So, your problem is actually with the fact that there's a world full of SSDs with
>buggy and slow implementations of TRIM?

That's the first-order problem. It means that even the good
drives cannot be used, because you generally don't know
ahead of time.

You could do some kind of white-list, but considering that
even for the good drives the upside is debatable and won't
help for many loads, it's not even worth it. The white-list
should be a combination of "for this load" and "for this
drive", which is obviously not something you can automate.

End result: currently the sane option is to just not use
TRIM at all by default. You can enable it (well, on most
OS's you can, at least for the main filesystems), but of
course most people don't.

>You could even argue that TRIM is yet another feature that SSDs manufacturers need to implement and get right.

The thing is, it's pretty low down on the list of things
they need to get right.

I'm still just waiting for the day when you can rely on
the SSD you buy to have good latencies on just regular IO.
Things have become much better over the last year,
but we're still not there.

>But that doesn't mean that TRIM isn't very helpfull for SSDs. When implemented right, of course.

The thing is, it really isn't helpful, even for the good
drives, because of all the crap drives out there.

Can't people understand that simple fact? You cannot just
look at one drive in a vacuum. It doesn't matter one whit
if that drive is perfect, if users can't rely on
it being perfect.

And there is no automatic way to figure out whether TRIM
actually works or not. So you have to treat all drives
as being diseased, because that's the only safe default.

Linus