Aaron Spink (aaronspink@notearthlink.net) on 1/26/11 wrote:
---------------------------
>iz (i@z.x) on 1/25/11 wrote:
>---------------------------
>>And when they do they just overwrite them nicely in a sequential way, so TRIM adds
>>nothing because such filesystems don't cause any fragmentation at all.
>>
>Better be careful, one of the anon's is likely to jump all over you for saying that.

Copy on write filesystems I think may potentially make better use of trim than traditional filesystems, but it really would depend on their freespace management. I haven't actually seen any numbers, however.

http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6957655

Here is a slightly interesting one:
http://thread.gmane.org/gmane.comp.file-systems.ext4/22020

Yes, btrfs does trim better than ext4 (only 40% performance drop, rather than 45%).

But there is not much detail of the SSDs involved, and it's just one small workload, so...

So lots of SSDs are bad at trim, that is a given, and there are numbers. Now the assertion people make is that "some" SSDs can magically do trim really really well. It's funny that we haven't seen any numbers for the good ones. You would think vendors who have the lead there would publish numbers.

Other thing is that COW filesystems do cause fragmentation. If you have a perfectly linear file and rewrite in the middle of it, how do you COW without fragmentation? Re-cow the entire file is not really a general purpose option, possibly usable in rare case, however: what do you do with freespace fragmentation?