From what I’ve read on ELI5 already, it seems like SSDs erase data when you trim them. But why do they erase it instead of just waiting for new data to be written? Doesn’t that destroy write cycles? Or am I misunderstanding something?
From what I’ve read on ELI5 already, it seems like SSDs erase data when you trim them. But why do they erase it instead of just waiting for new data to be written? Doesn’t that destroy write cycles? Or am I misunderstanding something?
Comments
>From what I’ve read on ELI5 already, it seems like SSDs erase data when you trim them.
You need to provide a link to a post claiming this, because from my POV, SSDs are like any other hard drive. “Deleted” files aren’t actually erased from storage. Instead, the space they occupy is marked available so that new files can overwrite the old data.
Unlike hard drives, on SSDs, the data must be cleared through a special “erase” operation before writing any new data. (Worse, most SSDs can only erase in blocks of several pages at once. They can’t erase just a single page.)
So the operating system will tell the SSD to mark a page as “user deleted this, I’ll erase it later” (through a mechanism called TRIM), then when the SSD has time, it’ll go around erasing all of those pages and marking them as “erased, ready for new data”.
If the operating system doesn’t tell the SSD that the page has been deleted, then the SSD has to be built with some extra pages that are kept erased (since every normal page could contain real data). Once those extra pages are consumed by a large write operation, speed slows to a crawl as the SSD is forced to erase and write back existing pages before writing new data.
They don’t delete data, they have a “TRIM” operation.
Deleting data from the operating system means that an operation occurs that removes a record from the “file table” that indicates where the data is located.
File systems were developed without knowing anything about the underlying storage layer, whether it is ssd, usb-stick or spinning hard drive.
If SSD do not know what data blocks are free, it will not know how to balance writes because it will not overwrite the “deleted” block, the TRIM operation lets the SSD know that the data block is free to be balanced and re-written to without the OS specifically addressing that block.
The traditional disk operation was to let the OS handle all data block addressing, SSD have an internal mapping that handles wear-leveling so what location the OS writes to isn’t necessarily where the SSD will finally write to.
When you write new data to a hard drive, you can write any data over any other data. You get the fastest performance by leaving the old data in place.
When you write data to an SSD, you can’t directly write new data over old data. You have to zero out the sector before you can write new data to it. So when we erase files off an SSD, we inform the SSD that the data is no longer needed. It will then zero out the data when it doesn’t have anything else to do. It makes things faster when you save new data, as you have already blank sectors you can write directly to.
TRIM on SSDs actually reduces the number of write operations, believe it or not. This is because SSDs are nothing like HDDs. When you need to write data to SSD and there isn’t a readily available place to write data, there is a series of operations that are performed to make a page ready to write.
Think of it like this… you need a clean page to write to, but all of your pages are collections of current data and no-longer-needed (deleted) data. In order to make a clean page, you go and take multiple pages and write all the ‘current’ data from them to a newly formed page which is obviously 100% current data, creating (hopefully) space for a brand new clean page to write new data to.
If you’re not ready with these clean pages to write to (which TRIM makes for you) then you have to do a lot of inefficient writes. The result is what we call Write Amplification, and this can wear out SSDs at a faster rate than if TRIM was in play.
Due to the design of SSDs you cannot overwrite individual bits like you would on an HDD. Data is grouped into „pages” which can be 4kb and this is the smallest amount of data you can write to. Once you write to a page, you need to erase the data before writing again. However, you cannot erase just one page. Pages are grouped into blocks ~128-512 pages. SSDs can only erase whole blocks, so you cannot erase individual pages.
The reason for it is because we don’t want data to be randomly erased. We only need to apply a small voltage to write data, but erasing is intentionally much more difficult and requires a large voltage. Since the voltage is so large we do it at a block level, because it is difficult to be very accurate with a larger voltage.
You cannot arbitrarily flip bits back-and-forth on an SSD. You can read and flip individual bits from their default state but to reset a bit to its default state you have to reset the whole block. So if you’ve flipped one bit and want to flip it back, you have to read the whole block into a cache, reset the entire block and write it all back with your changes. Instead of waiting for the block reset the SSD controller instead uses a different, previously reset block and updates all references to point to this new block, and marks the old block as unused and schedules it for TRIMming. To make sure that there are empty blocks available for writes the SSD controller then clears those scheduled blocks when it is otherwise idle.
In a traditional hard drive, write or overwrite are similar functions. In an SSD, write can only be performed to an empty cell/page. If a page already has data, It must be erased first. But erase functions can only be performed on a block of pages, not an individual block.
Instead of a singe overwrite, a series of processes read the existing data, cache it, erase the entire block, modify the contents to include the new data, then write into the empty pages. This significantly slows down the process compared to direct writes to empty space.
The TRIM function markes the physical storage empty so that it is available for immediate writing. This is less impactful to wear on the drive than the original type of optimization that would cause excessive read/erase/write cycles to the entire disk. TRIM takes place when the drive is not under load.
Unlike hard drives, you cannot write to a non empty memory cell. The cell must be zeroed out before it can be written to. Additionally there is a minimum number of memory cells (a block) that must be retrieved whenever you perform a read/write. Each block then contains the individual pages, which is the smallest unit that can have data assigned to it.
When you write something to an ssd, you must write an entire block, even if the data could fit on a single page in that block. Because data in SSDs is not necessarily continuous due to wear leveling and other practices, there might be old data in that block that needs to also be preserved. Therefore what typically happens when forced to write to a block that already contains data is that the entire block is fetched into memory, data assigned to the empty pages, then the modified block is written to a new empty block (all zeroed out) and the old block is marked for deletion/trimming the next time the SSD is idle.