Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
SSD Longevity
http://techreport.com/review/26523/the-ssd-endurance-experiment-casualties-on-the-way-to-a-petabyte
Excellent article.
Should help people put to rest fears about SSDs wearing out because of too many writes.
To TL;DR
SSD cells do have a limited number of write cycles. Consumer SSD drives are typically rated for 200TB worth of writes. Most people never come close to this in the reasonable lifespan of a SSD.
This article takes 6 various consumer-level SSDs and just writes to them non stop, and watched for when and how they failed.
The first drive failed at 735TB... nearly 4x the amount of data that it was rated for, and that was Intel, which didn't actually fail, but stopped working and went read-only once it exhausted all it's build-in spare cells. The TLC-based Samsung 840 (with TLC having notoriously lower write cycle counts and longevity) made it past 900TB of writes. All of the drives which failed gave plenty of warning - well before they actually hit their exhaustion point.
3 of the drives are still working past 1PB worth of writes (Samsgun 840 Pro, Corsair Neutron GTX, and Kingston HyperX 3K.
Also of note, the performance level of the drives didn't appreciably change as they got older. I was somewhat surprised by this as I figured less available over provisioning space would somewhat impact performance (SSD performance does degrade as a function of free space on the drive).
An average SSD write speed of around 300MB/s, to hit the rated 200TB worth of writes would require you to continuously write to the SSD at maximum speed for 8 days. For perspective, a few users posted their total write count from some SSDs in typical gaming scenarios in this thread, and the highest mark was 2 years in, just under 5TB of total writes - no wear near the 200TB rating, and well under the 700TB mark where drives actually started to fail.
Comments
If you compare the amount of data written to the hard disk's size, all the drives in that test were able to write more than 2 900 times their own capacity before failing. If you rewrote the whole hard disk with new data every day, you could do it for more than 8 years before the drive would fail you.
You do need to have the SMART reporting capability activated in the BIOS. Most systems come with it turned off. But yes, relying only on SMART reporting is not a good idea.
2 and they weren't mine so don't take that experience to the bank :P Both happened to work computers and the tech said it was no big deal. That could mean anything.
SMART reporting is good for predicting the predictable failures. Mechanical drives are more prone to unpredictable failure due to moving mechanical parts. That said, all electronics are subject to unpredictable failure. The SATA port on any drive could go bad due to bad circuitry on the SSD board. A voltage spike from a low-quality power supply and/or electrical surge could easily cause damage to any drive.
SMART is a good tool to use. Don't ignore it just because it doesn't catch everything.
To answer Torvaldr's question, Windows can check a drive's SMART data. I wouldn't rely on it, though. Get a SMART reader and check your drive's health manually every once in a while. Uncorrectable errors are signs that the drive needs to be replaced ASAP. Reallocated sectors are signs of wear. Everything else is difficult to interpret due to the differences in manufacturers and can be ignored.
SMART measures things that are easy to measure because they're easy to measure.
-----
Nice article. Thanks for linking it. What I really think it highlights is not that SSDs will last forever, but rather, that something else will kill it before excessive writes do. That "something else" could, of course, occur in a landfill.
I will say this:
This only applies to the SSDs they tested - other makes/models can, and almost certainly will, perform differently.
Also, as SSD manufactures face tighter and tighter margins, expect that amount of overprovision to get cut back to something more realistic. They want your hard drive to last as long as the warranty says, a few weeks beyond that for good measure, and then past that they don't care. That means they will probably start tuning the overprovision down somewhat to be able to cut the manufacturing costs. That may only be pennies per unit, but pennies add up when your making hundreds of thousands of units.
That's just conjecture, but it's what happens in most industries - corners get cut to help the margin and help competitiveness until they realize they've cut just a bit too much. Just for example, the Kingston drive: they tested it with and without compression. Surprise, the compression helps a lot - and that's just a software tweak, it doesn't require hardware. I fully expect the next generation to continue to use compression and possibly cut back on the amount of physical over-provisioning required to get the similar amount of life as a drive without compression enabled.
Not saying SSDs will all of a sudden suck tomorrow, or that any SSD not tested here is horrible. This article has some interesting data about the current generation of SSDs. I don't expect the next generation to be signifcantly different, but just because SSD A-F lasted this long is a good indication, but not proof, that SSDs G-Z will as well.
The way that overprovisioning is usually done is that an SSD will have 128 GB or 256 GB or 512 GB of physical NAND flash, and then they simply decide how much of that to make available to the end user. A "256 GB" SSD makes 256 billion bytes ~ 238.4 GB available to the end user and keeps the other 17.6 GB back for internal use such as overprovisioning. A "240 GB" SSD makes 240 billion bytes ~ 223.5 GB available to the end user and keeps 32.5 GB back.
The amounts available to the end user may change, but I don't expect the internal amounts of NAND flash to change for a given capacity range. If an SSD controller has 8 channels and NAND flash chips have capacities that are powers of 2, then the natural amount of physical NAND flash to hook up for maximum parallelism is 8 times a power of 2--which is itself a power of 2. Intel's early controller had 10 channels, which is why the capacities tended to be 10 times a power of 2, for 80 GB and 160 GB drives.
What really matters for using up the drive is not just how much NAND the controller makes available to you, but how much you actually use. If you never go over 192 GB in use, it doesn't matter whether the controller claims it will let you have 240 GB or 256 GB. That both have the same amount of physical NAND flash onboard is why I don't make much of a distinction between a "240 GB" SSD versus a "256 GB" one.
I don't expect wear leveling algorithms to become worse, either. There's no money to be saved by using an artificially worse algorithm then the best you can implement. The place where you can save money is by buying cheaper physical components--especially NAND flash.
-----
SandForce-based SSDs have been available for purchase for more than four years and have used compression since the beginning. To this day, they're the only ones that do it. Four years is long enough that even if a competitor didn't decide that SSD controller compression was a good idea until after they saw consumer SSDs with it launch, they could still have designed a controller, fabricated it, integrated into SSDs, and made them commercially available by now. Maybe we'll see such products soon, but that no other vendors have done so seems to indicate that they don't think it's an especially great idea.
It's not like other controller vendors can't add features to new controllers. Several have added AES encryption, for example. Of course, perhaps the main advantage of AES encryption for SSDs is that it scrambles 0's and 1's to make it so that you get about half of each and pretty uniformly distributed, so as to make it vanishingly unlikely that most of the cells in some physical area need the same charge and kill the drive that way. I don't know the internal details, but SandForce may have been relying on compression to serve that purpose.
For what it's worth, you can't encrypt with AES and then compress. The AES will scramble things to kill any compression algorithm that isn't sophisticated enough to decrypt AES and then compress. You could compress the data and then encrypt, though; the output of AES encrypted data is the same size as the input, at least if the input is a multiple of 16 bytes.