On 07.11.2017 03:01, Michael Stone wrote: > On Tue, Nov 07, 2017 at 02:20:56AM +0500, Alexander V. Makartsev wrote: >> As I know SSDs are different in a way HDDs write their data. HDDs write >> directly to the free block and are done. SSDs on the other hand will >> have to >> erase free block first to get it ready for writing and then write to >> it. This >> is because how NAND work. It's even worse, if a destination block has >> some data >> on it already and there is some data needs to be written onto it, >> then SSD will >> copy data already written, then it will fully erase destination >> block, then it >> will concatenate copied data with data to be written and finally will >> write >> merged data to the block. > > This is mostly correct. The erasing is done in "erase block" size > units (much larger than disk blocks), while writing is done in > "pages", usually with hundreds of pages per erase block. So a 4kbyte > write might trigger a multi-megabyte erase if there aren't erased > pages available for writing. If an erase block has been partially > written, a write can go to an unwritten page without any additional > steps. In practice, most drive writes go immediately to already-erased > blocks, with the garbage collector erasing blocks while the drive is > idle (rather than actually reading, erasing, and writing for every > write). The drive then keeps track of which physical address is > associated with which logical address--that relationship changes as > blocks are written and rewritten, and it is not the case that the data > on the drive is physically ordered the same way it appears logically. > >> TRIM will erase blocks marked as deleted beforehand even if they are >> not needed >> for writing right now. This overhead is the reason that causes >> increased drive >> wear and decrease it's lifespan. > > No, TRIM does not cause extra erases that wear out a drive. If you > don't TRIM blocks and mark them as unused, the garbage collector will > still erase blocks it doesn't need yet in order to have pages > available for future writes. (Either after the garbage collector > decides to launch a read & rewrite cycle to consolidate free space, or > when all the pages > in the erase block have been rewritten to a new erase block and the > old erase block is known to be empty) The only differences introduced > by using TRIM is that the drive can skip the possible read & rewrite > step, and the drive can increase the free pool so more new pages can > be written before a garbage collection is forced. I still think there are scenarios where NAND ICs will suffer extra writes, when TRIM is constantly used. Also, can you explain what logic is behind this quote from "man 8 fstrim": "Running fstrim frequently, or even using mount -o discard, might negatively affect the lifetime of poor-quality SSD devices."
I think I will find my answers in this huge article on wikipedia when I wade through it eventually, should be quite interesting read: https://en.wikipedia.org/wiki/Write_amplification -- With kindest regards, Alexander. ⢀⣴⠾⠻⢶⣦⠀ ⣾⠁⢠⠒⠀⣿⡁ Debian - The universal operating system ⢿⡄⠘⠷⠚⠋⠀ https://www.debian.org ⠈⠳⣄⠀⠀⠀⠀
