Hi Gerald, Karim, >From what I understand, wear levelling algorithms included in high end ssd drives are designed to write to different cells. Say you append data to the log file, the ssd won't fill the cell with data until the whole cell is filled. It will write to several different cells at the same time and use those who have been less used. The OS itself doesn't know where the data is physically written.
What bother me the most is not how long the drive will last (hard drive fails aswell), but the fact that the lifetime is pretty much fixed and not variable for a given usage and down to only one thing : cell writes cycles. Say you have two raid mirrors, one with regular hard drives, the other with ssd drives. A regular hard drive has an estimated and expected lifetime, but the failure is pretty much randomized by the mechanical nature of the device. So you don't really need to monitor them so closely. When it fails, you detach/offline it, replace it and resync the mirror. Most of the time S.M.A.R.T. almost always warn you when it is too late after all, but you don't live with the fear that both drive will fail at the same time. In the case of ssd, both drives will handle the same data and usage, with the very same wear levelling algorithm. If the lifetime is fixed and limited to cell writing, we should then expect both drive to fail at the same time. It looks very very bad to me. What I would like to know is how the drive react when it approach the limit of his cells write cycles. Does it warn the OS (hey I'm failing, please replace me !) ? Do you have to just wait for write/read errors ? Does the data just disappear, or is the cell still readable ? Should we do a specific monitoring of these drives ? Are ssd S.M.A.R.T. capable and does it give interesting information ? Strangely, vendors do not communicate a lot on these areas. Regards Thomas On Wed, Nov 4, 2009 at 2:02 PM, ?<gerald.eggenberger at sunwave.ch> wrote: > Hi Karim > > you're obviously not interessed in a discussion about the possibility of > 'negative' technical aspects of SSD's, nor has anybody else returned any > statements on this subject. > > Therefore i apologise to the mailinglist for disturbance. > > regards > G?rald > _______________________________________________ > ug-chosug mailing list > ug-chosug at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/ug-chosug >
