Quote:
Originally Posted by
ChristophB 
The threads above mine described garbage collection in terms that sounded like the SSD just "knows" what the user and file system have discarded. That doesn't happen with HFS.
It's more a function of OS support...
Quote:
My query was to ask them to think through how the SSD "knows". I don't want an SSD that has to have firmware revs every time an vendor revs/fixes the file system software.
Neither TRIM or disks with extra space for internal garbage collection introduces a firmware dependancy. Not sure where you are going with that...
Quote:
Reserve space is a nice temp fix because it'll be used as an alternative to writing over a used block. The SSD will eventually run out of that too.
If it's reserve space, by definition then it can't fill up, right? Otherwise, why would it be called reserve space?
Let's keep it simple. Let's leave out file systems and size and just talk about individual memory cells - because without TRIM that's all an SSD controller knows or cares about.
If I have a disk with an advertised capacity that just happens to work out to 100 memory cells, an I have another 20 cells in reserve, I will always have the aggregate equivalent of twenty free cells. Always - whatever OS and file system is running on the drive simply doesn't even know those cells exist.
So, when the number of totally empty cells passes some pre-defined point, the firmware analyzes all the cells, probably picks the cells with the least amount of data in them, consolidates them into one free cell and then erases the previously partially filled cells and boom - a net gain of free, ready to write at full speed cells.
Kind of a specialized de-frag, but to optimize free space.
With this model, the drive neither knows nor cares about what is being written to it - it just does it's thing in the background, ensuring as many totally empty cells are "on hand" as possible.
Now theoretically, if your drive isn't full and you write lots of big files, TRIM can allow the SSD to optimize more free cells, but as I said I would much rather see a new filesystem and interface developed that allows the OS t more effectively use "disk" - ESP. when the disk isn't disk at all but something totally different with totally different usage and optimization patterns!
I mean hard drives haven't mapped heads/cylinders 1:1 for over 20 years! Isn't it time to move on?