Max (max@a.com) on 1/25/11 wrote:
---------------------------
>gallier2 (gallier2@gmx.de) on 1/25/11 wrote:
>---------------------------
>>So, sorry that in the end I answered my question myself, but thinking about the
>>problem from the firmware side made me realize that Linus is right: TRIM is more
>>than useless, adding complexity and overhead for something the firmware already has to do.
>
>TRIM is only useless if the drive is 100% full 100% of the time.

It's useless long before that point. Reread what I wrote.
At best you gain some theoretical/hypothetic time because you could remap sectors a little bit earlier at the cost of a certainty of a loss of time (the command overhead+loss of latency). Not a good trade-off.

> It's useful to
>have free blocks, but it's also useful to be able to use the full capacity.

Yes. It has normaly more physical than logical capacity, so there are always free blocks. If the logical capacity is exhausted it will have to erase the spare areas and shuffle data around, not a good thing for sure, but if the disk is full you already have other problems, much worse than that.

> In a
>way, TRIM makes the drive variable size. To the filesystem it's fixed size, but
>internally the managed size is determined by the free space.
>
Which is always the case, TRIM or no TRIM.

Remember, what does a WRITE command do?
Let's decompose, what would a simplified write command in a SSD look like:


write(data, LBA(x))
{
if(LBA(x) mapped to PBA(y)) {
unmap(LBA(x))
mark_as_free(PBA(y)) // GC will handle it later
}
PBA(z) = get_new_PBA()
map(LBA(x), PBA(z))
write(data, PBA(z))
}


And the if part above is the TRIM command!
Let's imagine a syscall with command going over the SATA wire takes the time A.
checking if a sector is mapped B
the execution of TRIM is C
the time to get a new PBA D
The write itself E.

Sys1 uses TRIM, Sys2 doesn't

When we write a mapped block we will have

SYS1 : trim+write: A+B+C + A+B+D+E
SYS2 : write : A+B+C+D+E


When we write an unmapped block we will have

SYS1 : write: A+B+D+E
SYS2 : write: A+B+D+E


The only variable, that may change between SYS1 and SYS2 is D, the time to get a fresh empty sector. This value may change substantially when the disk reaches a certain capacity limit, but until then that value won't differ a lot between both system. But the impact of the supplemental A+B in SYS1 is always for 100% of the time.