anon (anon@anon.com) on 1/19/11 wrote:
---------------------------
>I really think that depends on the device, and isn't necessarily true at all. It
>_may_ be true, but even when it is, you have to send a request down the wire and
>track it or wait for it, which is not free.
>
Can we agree that if the SSD does anything but either ignore the TRIM command or merely mark the PBA to free but dirty, it is probably a bad design?

And yes you do have a command down the wire and track it, and wait for it to eventually come back, but that shouldn't present any issue at the software level. It also shouldn't really have any performance impact except of the completely margin kind unless the device blocks while it actually erases blocks, but I've already said that would be a bad design.

>I guess if you are unsure about what he meant, you could assume that Linus either
>does know the bare minimum SSD 101, or doesn't. It amazes me when people take the latter route.
>
Unfortunately, considering Linus' hatred of the existence of TRIM and considering he is a smart guy, the only logical conclusion I could come to was that he didn't fully understand the underlying hardware(and it wouldn't be the first time that a very bright, well expereince, extremely good primarily software guy didn't understand an underlying hardware nuance or operation). Now if he is talking about some software interaction with some SSDs that behave poorly/incorrectly with the TRIM command, I might understand, and would like a list of those devices, and make sure to cross them off of my potential purchase list.

>As always, it's more complicated than that.
>
If they aren't reusing the same subroutine for TRIM mark as free/dirty and the mapped LBA write mark as free/dirty then they aren't following either good coding guidelines OR good QA/validation/verification practices.


>
>No.
>

Then enlighten me. The overhead of sending the TRIM command should be minimal and the overhead on the SSD should be minimal. You might be able to save some time by storing in software a to be TRIM LBA list but it would add complexity and overhead there in the case where you later decide to write an LBA that is on the to be TRIM list (cross checking/updating/etc and something you only really would want for SSDs and storage systems that support TRIM).