"John Stoffel" <john@stoffel.org> writes:
"Jeff" == Jeff Moyer <jmoyer@redhat.com> writes: Jeff> I've seen many claims that SSDs are more reliable than Jeff> traditional HDDs, but I haven't seen data to back up those Jeff> claims.
Some of the arguements I've read on SSD reliability are quite interesting. If you do the math on how much data you need to write to destroy an SSD with proper (note!) wear levelling, it's really quite high. But I think the real key is to write data to them in blocks of 512 bytes as much as possible, and to not write smaller amounts if at all possible.
Did you mean to say 512KB? You can't write smaller than the exported block size (which will be either 512 bytes or 4k). Perhaps what you meant to say was that you should try to issue writes that are aligned on and multiples of the erase block size.
Jeff> My main concern is that SSDs are still young, so they haven't Jeff> had much soak time. Whether or not the flash is reliable is Jeff> only a small part of the battle. Buggy firmware is a much more Jeff> likely problem at this stage in the game. No single point of Jeff> failure is a good design point when dealing with your data. ;-)
I think this is a key issue, the firmware and controllers, not the flash chips themselves.
Jeff> I remain cautiously optimistic (though I've already bricked one Jeff> drive).
Details?
During TRIM testing, we managed to get an OCZ Vertex to spew I/O errors, and then finally fail to be recognized by the BIOS upon reboot. Note that this isn't a normal workload (the drive is being pounded on) and TRIM is still very new. So, I am not saying that the drives are not reliable in general. I simply don't know. Cheers, Jeff